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;
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         mutex_lock(&extent_root->fs_info->chunk_mutex);
1534         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1535         if (ret == -ENOSPC) {
1536 printk("space info full %Lu\n", flags);
1537                 space_info->full = 1;
1538                 goto out_unlock;
1539         }
1540         BUG_ON(ret);
1541
1542         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1543                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1544         BUG_ON(ret);
1545
1546 out_unlock:
1547         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1548 out:
1549         return ret;
1550 }
1551
1552 static int update_block_group(struct btrfs_trans_handle *trans,
1553                               struct btrfs_root *root,
1554                               u64 bytenr, u64 num_bytes, int alloc,
1555                               int mark_free)
1556 {
1557         struct btrfs_block_group_cache *cache;
1558         struct btrfs_fs_info *info = root->fs_info;
1559         u64 total = num_bytes;
1560         u64 old_val;
1561         u64 byte_in_group;
1562
1563         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1564         while(total) {
1565                 cache = btrfs_lookup_block_group(info, bytenr);
1566                 if (!cache) {
1567                         return -1;
1568                 }
1569                 byte_in_group = bytenr - cache->key.objectid;
1570                 WARN_ON(byte_in_group > cache->key.offset);
1571
1572                 spin_lock(&cache->lock);
1573                 cache->dirty = 1;
1574                 old_val = btrfs_block_group_used(&cache->item);
1575                 num_bytes = min(total, cache->key.offset - byte_in_group);
1576                 if (alloc) {
1577                         old_val += num_bytes;
1578                         cache->space_info->bytes_used += num_bytes;
1579                         btrfs_set_block_group_used(&cache->item, old_val);
1580                         spin_unlock(&cache->lock);
1581                 } else {
1582                         old_val -= num_bytes;
1583                         cache->space_info->bytes_used -= num_bytes;
1584                         btrfs_set_block_group_used(&cache->item, old_val);
1585                         spin_unlock(&cache->lock);
1586                         if (mark_free) {
1587                                 int ret;
1588                                 ret = btrfs_add_free_space(cache, bytenr,
1589                                                            num_bytes);
1590                                 if (ret)
1591                                         return -1;
1592                         }
1593                 }
1594                 total -= num_bytes;
1595                 bytenr += num_bytes;
1596         }
1597         return 0;
1598 }
1599
1600 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1601 {
1602         struct btrfs_block_group_cache *cache;
1603
1604         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1605         if (!cache)
1606                 return 0;
1607
1608         return cache->key.objectid;
1609 }
1610
1611 int btrfs_update_pinned_extents(struct btrfs_root *root,
1612                                 u64 bytenr, u64 num, int pin)
1613 {
1614         u64 len;
1615         struct btrfs_block_group_cache *cache;
1616         struct btrfs_fs_info *fs_info = root->fs_info;
1617
1618         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1619         if (pin) {
1620                 set_extent_dirty(&fs_info->pinned_extents,
1621                                 bytenr, bytenr + num - 1, GFP_NOFS);
1622         } else {
1623                 clear_extent_dirty(&fs_info->pinned_extents,
1624                                 bytenr, bytenr + num - 1, GFP_NOFS);
1625         }
1626         while (num > 0) {
1627                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1628                 BUG_ON(!cache);
1629                 len = min(num, cache->key.offset -
1630                           (bytenr - cache->key.objectid));
1631                 if (pin) {
1632                         spin_lock(&cache->lock);
1633                         cache->pinned += len;
1634                         cache->space_info->bytes_pinned += len;
1635                         spin_unlock(&cache->lock);
1636                         fs_info->total_pinned += len;
1637                 } else {
1638                         spin_lock(&cache->lock);
1639                         cache->pinned -= len;
1640                         cache->space_info->bytes_pinned -= len;
1641                         spin_unlock(&cache->lock);
1642                         fs_info->total_pinned -= len;
1643                 }
1644                 bytenr += len;
1645                 num -= len;
1646         }
1647         return 0;
1648 }
1649
1650 static int update_reserved_extents(struct btrfs_root *root,
1651                                    u64 bytenr, u64 num, int reserve)
1652 {
1653         u64 len;
1654         struct btrfs_block_group_cache *cache;
1655         struct btrfs_fs_info *fs_info = root->fs_info;
1656
1657         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1658         while (num > 0) {
1659                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1660                 BUG_ON(!cache);
1661                 len = min(num, cache->key.offset -
1662                           (bytenr - cache->key.objectid));
1663                 if (reserve) {
1664                         spin_lock(&cache->lock);
1665                         cache->reserved += len;
1666                         cache->space_info->bytes_reserved += len;
1667                         spin_unlock(&cache->lock);
1668                 } else {
1669                         spin_lock(&cache->lock);
1670                         cache->reserved -= len;
1671                         cache->space_info->bytes_reserved -= len;
1672                         spin_unlock(&cache->lock);
1673                 }
1674                 bytenr += len;
1675                 num -= len;
1676         }
1677         return 0;
1678 }
1679
1680 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1681 {
1682         u64 last = 0;
1683         u64 start;
1684         u64 end;
1685         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1686         int ret;
1687
1688         while(1) {
1689                 ret = find_first_extent_bit(pinned_extents, last,
1690                                             &start, &end, EXTENT_DIRTY);
1691                 if (ret)
1692                         break;
1693                 set_extent_dirty(copy, start, end, GFP_NOFS);
1694                 last = end + 1;
1695         }
1696         return 0;
1697 }
1698
1699 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1700                                struct btrfs_root *root,
1701                                struct extent_io_tree *unpin)
1702 {
1703         u64 start;
1704         u64 end;
1705         int ret;
1706         struct btrfs_block_group_cache *cache;
1707
1708         mutex_lock(&root->fs_info->alloc_mutex);
1709         while(1) {
1710                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1711                                             EXTENT_DIRTY);
1712                 if (ret)
1713                         break;
1714                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1715                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1716                 cache = btrfs_lookup_block_group(root->fs_info, start);
1717                 if (cache->cached)
1718                         btrfs_add_free_space(cache, start, end - start + 1);
1719                 if (need_resched()) {
1720                         mutex_unlock(&root->fs_info->alloc_mutex);
1721                         cond_resched();
1722                         mutex_lock(&root->fs_info->alloc_mutex);
1723                 }
1724         }
1725         mutex_unlock(&root->fs_info->alloc_mutex);
1726         return 0;
1727 }
1728
1729 static int finish_current_insert(struct btrfs_trans_handle *trans,
1730                                  struct btrfs_root *extent_root)
1731 {
1732         u64 start;
1733         u64 end;
1734         u64 priv;
1735         struct btrfs_fs_info *info = extent_root->fs_info;
1736         struct btrfs_path *path;
1737         struct btrfs_extent_ref *ref;
1738         struct pending_extent_op *extent_op;
1739         struct btrfs_key key;
1740         struct btrfs_extent_item extent_item;
1741         int ret;
1742         int err = 0;
1743
1744         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1745         btrfs_set_stack_extent_refs(&extent_item, 1);
1746         path = btrfs_alloc_path();
1747
1748         while(1) {
1749                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1750                                             &end, EXTENT_LOCKED);
1751                 if (ret)
1752                         break;
1753
1754                 ret = get_state_private(&info->extent_ins, start, &priv);
1755                 BUG_ON(ret);
1756                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1757
1758                 if (extent_op->type == PENDING_EXTENT_INSERT) {
1759                         key.objectid = start;
1760                         key.offset = end + 1 - start;
1761                         key.type = BTRFS_EXTENT_ITEM_KEY;
1762                         err = btrfs_insert_item(trans, extent_root, &key,
1763                                         &extent_item, sizeof(extent_item));
1764                         BUG_ON(err);
1765
1766                         clear_extent_bits(&info->extent_ins, start, end,
1767                                           EXTENT_LOCKED, GFP_NOFS);
1768
1769                         err = insert_extent_backref(trans, extent_root, path,
1770                                                 start, extent_op->parent,
1771                                                 extent_root->root_key.objectid,
1772                                                 extent_op->generation,
1773                                                 extent_op->level, 0);
1774                         BUG_ON(err);
1775                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1776                         err = lookup_extent_backref(trans, extent_root, path,
1777                                                 start, extent_op->orig_parent,
1778                                                 extent_root->root_key.objectid,
1779                                                 extent_op->orig_generation, 0);
1780                         BUG_ON(err);
1781
1782                         clear_extent_bits(&info->extent_ins, start, end,
1783                                           EXTENT_LOCKED, GFP_NOFS);
1784
1785                         key.objectid = start;
1786                         key.offset = extent_op->parent;
1787                         key.type = BTRFS_EXTENT_REF_KEY;
1788                         err = btrfs_set_item_key_safe(trans, extent_root, path,
1789                                                       &key);
1790                         BUG_ON(err);
1791                         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1792                                              struct btrfs_extent_ref);
1793                         btrfs_set_ref_generation(path->nodes[0], ref,
1794                                                  extent_op->generation);
1795                         btrfs_mark_buffer_dirty(path->nodes[0]);
1796                         btrfs_release_path(extent_root, path);
1797                 } else {
1798                         BUG_ON(1);
1799                 }
1800                 kfree(extent_op);
1801
1802                 if (need_resched()) {
1803                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1804                         cond_resched();
1805                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1806                 }
1807         }
1808         btrfs_free_path(path);
1809         return 0;
1810 }
1811
1812 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1813                           struct btrfs_root *root,
1814                           u64 bytenr, u64 num_bytes, int is_data)
1815 {
1816         int err = 0;
1817         struct extent_buffer *buf;
1818
1819         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1820         if (is_data)
1821                 goto pinit;
1822
1823         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1824         if (!buf)
1825                 goto pinit;
1826
1827         /* we can reuse a block if it hasn't been written
1828          * and it is from this transaction.  We can't
1829          * reuse anything from the tree log root because
1830          * it has tiny sub-transactions.
1831          */
1832         if (btrfs_buffer_uptodate(buf, 0) &&
1833             btrfs_try_tree_lock(buf)) {
1834                 u64 header_owner = btrfs_header_owner(buf);
1835                 u64 header_transid = btrfs_header_generation(buf);
1836                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1837                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
1838                     header_transid == trans->transid &&
1839                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1840                         clean_tree_block(NULL, root, buf);
1841                         btrfs_tree_unlock(buf);
1842                         free_extent_buffer(buf);
1843                         return 1;
1844                 }
1845                 btrfs_tree_unlock(buf);
1846         }
1847         free_extent_buffer(buf);
1848 pinit:
1849         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1850
1851         BUG_ON(err < 0);
1852         return 0;
1853 }
1854
1855 /*
1856  * remove an extent from the root, returns 0 on success
1857  */
1858 static int __free_extent(struct btrfs_trans_handle *trans,
1859                          struct btrfs_root *root,
1860                          u64 bytenr, u64 num_bytes, u64 parent,
1861                          u64 root_objectid, u64 ref_generation,
1862                          u64 owner_objectid, u64 owner_offset,
1863                          int pin, int mark_free)
1864 {
1865         struct btrfs_path *path;
1866         struct btrfs_key key;
1867         struct btrfs_fs_info *info = root->fs_info;
1868         struct btrfs_root *extent_root = info->extent_root;
1869         struct extent_buffer *leaf;
1870         int ret;
1871         int extent_slot = 0;
1872         int found_extent = 0;
1873         int num_to_del = 1;
1874         struct btrfs_extent_item *ei;
1875         u32 refs;
1876
1877         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1878         key.objectid = bytenr;
1879         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1880         key.offset = num_bytes;
1881         path = btrfs_alloc_path();
1882         if (!path)
1883                 return -ENOMEM;
1884
1885         path->reada = 1;
1886         ret = lookup_extent_backref(trans, extent_root, path, bytenr, parent,
1887                                     root_objectid, ref_generation, 1);
1888         if (ret == 0) {
1889                 struct btrfs_key found_key;
1890                 extent_slot = path->slots[0];
1891                 while(extent_slot > 0) {
1892                         extent_slot--;
1893                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1894                                               extent_slot);
1895                         if (found_key.objectid != bytenr)
1896                                 break;
1897                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1898                             found_key.offset == num_bytes) {
1899                                 found_extent = 1;
1900                                 break;
1901                         }
1902                         if (path->slots[0] - extent_slot > 5)
1903                                 break;
1904                 }
1905                 if (!found_extent) {
1906                         ret = remove_extent_backref(trans, extent_root, path);
1907                         BUG_ON(ret);
1908                         btrfs_release_path(extent_root, path);
1909                         ret = btrfs_search_slot(trans, extent_root,
1910                                                 &key, path, -1, 1);
1911                         BUG_ON(ret);
1912                         extent_slot = path->slots[0];
1913                 }
1914         } else {
1915                 btrfs_print_leaf(extent_root, path->nodes[0]);
1916                 WARN_ON(1);
1917                 printk("Unable to find ref byte nr %Lu root %Lu "
1918                        " gen %Lu owner %Lu offset %Lu\n", bytenr,
1919                        root_objectid, ref_generation, owner_objectid,
1920                        owner_offset);
1921         }
1922
1923         leaf = path->nodes[0];
1924         ei = btrfs_item_ptr(leaf, extent_slot,
1925                             struct btrfs_extent_item);
1926         refs = btrfs_extent_refs(leaf, ei);
1927         BUG_ON(refs == 0);
1928         refs -= 1;
1929         btrfs_set_extent_refs(leaf, ei, refs);
1930
1931         btrfs_mark_buffer_dirty(leaf);
1932
1933         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1934                 struct btrfs_extent_ref *ref;
1935                 ref = btrfs_item_ptr(leaf, path->slots[0],
1936                                      struct btrfs_extent_ref);
1937                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
1938                 /* if the back ref and the extent are next to each other
1939                  * they get deleted below in one shot
1940                  */
1941                 path->slots[0] = extent_slot;
1942                 num_to_del = 2;
1943         } else if (found_extent) {
1944                 /* otherwise delete the extent back ref */
1945                 ret = remove_extent_backref(trans, extent_root, path);
1946                 BUG_ON(ret);
1947                 /* if refs are 0, we need to setup the path for deletion */
1948                 if (refs == 0) {
1949                         btrfs_release_path(extent_root, path);
1950                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1951                                                 -1, 1);
1952                         BUG_ON(ret);
1953                 }
1954         }
1955
1956         if (refs == 0) {
1957                 u64 super_used;
1958                 u64 root_used;
1959 #ifdef BIO_RW_DISCARD
1960                 u64 map_length = num_bytes;
1961                 struct btrfs_multi_bio *multi = NULL;
1962 #endif
1963
1964                 if (pin) {
1965                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
1966                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
1967                         if (ret > 0)
1968                                 mark_free = 1;
1969                         BUG_ON(ret < 0);
1970                 }
1971
1972                 /* block accounting for super block */
1973                 spin_lock_irq(&info->delalloc_lock);
1974                 super_used = btrfs_super_bytes_used(&info->super_copy);
1975                 btrfs_set_super_bytes_used(&info->super_copy,
1976                                            super_used - num_bytes);
1977                 spin_unlock_irq(&info->delalloc_lock);
1978
1979                 /* block accounting for root item */
1980                 root_used = btrfs_root_used(&root->root_item);
1981                 btrfs_set_root_used(&root->root_item,
1982                                            root_used - num_bytes);
1983                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1984                                       num_to_del);
1985                 BUG_ON(ret);
1986                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1987                                          mark_free);
1988                 BUG_ON(ret);
1989
1990 #ifdef BIO_RW_DISCARD
1991                 /* Tell the block device(s) that the sectors can be discarded */
1992                 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1993                                       bytenr, &map_length, &multi, 0);
1994                 if (!ret) {
1995                         struct btrfs_bio_stripe *stripe = multi->stripes;
1996                         int i;
1997
1998                         if (map_length > num_bytes)
1999                                 map_length = num_bytes;
2000
2001                         for (i = 0; i < multi->num_stripes; i++, stripe++) {
2002                                 blkdev_issue_discard(stripe->dev->bdev,
2003                                                      stripe->physical >> 9,
2004                                                      map_length >> 9);
2005                         }
2006                         kfree(multi);
2007                 }
2008 #endif
2009         }
2010         btrfs_free_path(path);
2011         finish_current_insert(trans, extent_root);
2012         return ret;
2013 }
2014
2015 /*
2016  * find all the blocks marked as pending in the radix tree and remove
2017  * them from the extent map
2018  */
2019 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2020                                btrfs_root *extent_root)
2021 {
2022         int ret;
2023         int err = 0;
2024         int mark_free = 0;
2025         u64 start;
2026         u64 end;
2027         u64 priv;
2028         struct extent_io_tree *pending_del;
2029         struct extent_io_tree *extent_ins;
2030         struct pending_extent_op *extent_op;
2031
2032         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
2033         extent_ins = &extent_root->fs_info->extent_ins;
2034         pending_del = &extent_root->fs_info->pending_del;
2035
2036         while(1) {
2037                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2038                                             EXTENT_LOCKED);
2039                 if (ret)
2040                         break;
2041
2042                 ret = get_state_private(pending_del, start, &priv);
2043                 BUG_ON(ret);
2044                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2045
2046                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2047                                   GFP_NOFS);
2048
2049                 ret = pin_down_bytes(trans, extent_root, start,
2050                                      end + 1 - start, 0);
2051                 mark_free = ret > 0;
2052                 if (!test_range_bit(extent_ins, start, end,
2053                                     EXTENT_LOCKED, 0)) {
2054 free_extent:
2055                         ret = __free_extent(trans, extent_root,
2056                                             start, end + 1 - start,
2057                                             extent_op->orig_parent,
2058                                             extent_root->root_key.objectid,
2059                                             extent_op->orig_generation,
2060                                             extent_op->level, 0, 0, mark_free);
2061                         kfree(extent_op);
2062                 } else {
2063                         kfree(extent_op);
2064                         ret = get_state_private(extent_ins, start, &priv);
2065                         BUG_ON(ret);
2066                         extent_op = (struct pending_extent_op *)
2067                                                         (unsigned long)priv;
2068
2069                         clear_extent_bits(extent_ins, start, end,
2070                                           EXTENT_LOCKED, GFP_NOFS);
2071
2072                         if (extent_op->type == PENDING_BACKREF_UPDATE)
2073                                 goto free_extent;
2074
2075                         ret = update_block_group(trans, extent_root, start,
2076                                                 end + 1 - start, 0, mark_free);
2077                         BUG_ON(ret);
2078                         kfree(extent_op);
2079                 }
2080                 if (ret)
2081                         err = ret;
2082
2083                 if (need_resched()) {
2084                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
2085                         cond_resched();
2086                         mutex_lock(&extent_root->fs_info->alloc_mutex);
2087                 }
2088         }
2089         return err;
2090 }
2091
2092 /*
2093  * remove an extent from the root, returns 0 on success
2094  */
2095 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2096                                struct btrfs_root *root,
2097                                u64 bytenr, u64 num_bytes, u64 parent,
2098                                u64 root_objectid, u64 ref_generation,
2099                                u64 owner_objectid, u64 owner_offset, int pin)
2100 {
2101         struct btrfs_root *extent_root = root->fs_info->extent_root;
2102         int pending_ret;
2103         int ret;
2104
2105         WARN_ON(num_bytes < root->sectorsize);
2106         if (root == extent_root) {
2107                 struct pending_extent_op *extent_op;
2108
2109                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2110                 BUG_ON(!extent_op);
2111
2112                 extent_op->type = PENDING_EXTENT_DELETE;
2113                 extent_op->bytenr = bytenr;
2114                 extent_op->num_bytes = num_bytes;
2115                 extent_op->parent = parent;
2116                 extent_op->orig_parent = parent;
2117                 extent_op->generation = ref_generation;
2118                 extent_op->orig_generation = ref_generation;
2119                 extent_op->level = (int)owner_objectid;
2120
2121                 set_extent_bits(&root->fs_info->pending_del,
2122                                 bytenr, bytenr + num_bytes - 1,
2123                                 EXTENT_LOCKED, GFP_NOFS);
2124                 set_state_private(&root->fs_info->pending_del,
2125                                   bytenr, (unsigned long)extent_op);
2126                 return 0;
2127         }
2128         /* if metadata always pin */
2129         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2130                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2131                         struct btrfs_block_group_cache *cache;
2132
2133                         /* btrfs_free_reserved_extent */
2134                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2135                         BUG_ON(!cache);
2136                         btrfs_add_free_space(cache, bytenr, num_bytes);
2137                         update_reserved_extents(root, bytenr, num_bytes, 0);
2138                         return 0;
2139                 }
2140                 pin = 1;
2141         }
2142
2143         /* if data pin when any transaction has committed this */
2144         if (ref_generation != trans->transid)
2145                 pin = 1;
2146
2147         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2148                             root_objectid, ref_generation, owner_objectid,
2149                             owner_offset, pin, pin == 0);
2150
2151         finish_current_insert(trans, root->fs_info->extent_root);
2152         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2153         return ret ? ret : pending_ret;
2154 }
2155
2156 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2157                       struct btrfs_root *root,
2158                       u64 bytenr, u64 num_bytes, u64 parent,
2159                       u64 root_objectid, u64 ref_generation,
2160                       u64 owner_objectid, u64 owner_offset, int pin)
2161 {
2162         int ret;
2163
2164         maybe_lock_mutex(root);
2165         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2166                                   root_objectid, ref_generation,
2167                                   owner_objectid, owner_offset, pin);
2168         maybe_unlock_mutex(root);
2169         return ret;
2170 }
2171
2172 static u64 stripe_align(struct btrfs_root *root, u64 val)
2173 {
2174         u64 mask = ((u64)root->stripesize - 1);
2175         u64 ret = (val + mask) & ~mask;
2176         return ret;
2177 }
2178
2179 /*
2180  * walks the btree of allocated extents and find a hole of a given size.
2181  * The key ins is changed to record the hole:
2182  * ins->objectid == block start
2183  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2184  * ins->offset == number of blocks
2185  * Any available blocks before search_start are skipped.
2186  */
2187 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2188                                      struct btrfs_root *orig_root,
2189                                      u64 num_bytes, u64 empty_size,
2190                                      u64 search_start, u64 search_end,
2191                                      u64 hint_byte, struct btrfs_key *ins,
2192                                      u64 exclude_start, u64 exclude_nr,
2193                                      int data)
2194 {
2195         int ret;
2196         u64 orig_search_start;
2197         struct btrfs_root * root = orig_root->fs_info->extent_root;
2198         struct btrfs_fs_info *info = root->fs_info;
2199         u64 total_needed = num_bytes;
2200         u64 *last_ptr = NULL;
2201         struct btrfs_block_group_cache *block_group;
2202         int chunk_alloc_done = 0;
2203         int empty_cluster = 2 * 1024 * 1024;
2204         int allowed_chunk_alloc = 0;
2205
2206         WARN_ON(num_bytes < root->sectorsize);
2207         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2208
2209         if (orig_root->ref_cows || empty_size)
2210                 allowed_chunk_alloc = 1;
2211
2212         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2213                 last_ptr = &root->fs_info->last_alloc;
2214                 empty_cluster = 256 * 1024;
2215         }
2216
2217         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2218                 last_ptr = &root->fs_info->last_data_alloc;
2219
2220         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2221                 last_ptr = &root->fs_info->last_log_alloc;
2222                 if (!last_ptr == 0 && root->fs_info->last_alloc) {
2223                         *last_ptr = root->fs_info->last_alloc + empty_cluster;
2224                 }
2225         }
2226
2227         if (last_ptr) {
2228                 if (*last_ptr)
2229                         hint_byte = *last_ptr;
2230                 else
2231                         empty_size += empty_cluster;
2232         }
2233
2234         search_start = max(search_start, first_logical_byte(root, 0));
2235         orig_search_start = search_start;
2236
2237         search_start = max(search_start, hint_byte);
2238         total_needed += empty_size;
2239
2240 new_group:
2241         block_group = btrfs_lookup_first_block_group(info, search_start);
2242
2243         /*
2244          * Ok this looks a little tricky, buts its really simple.  First if we
2245          * didn't find a block group obviously we want to start over.
2246          * Secondly, if the block group we found does not match the type we
2247          * need, and we have a last_ptr and its not 0, chances are the last
2248          * allocation we made was at the end of the block group, so lets go
2249          * ahead and skip the looking through the rest of the block groups and
2250          * start at the beginning.  This helps with metadata allocations,
2251          * since you are likely to have a bunch of data block groups to search
2252          * through first before you realize that you need to start over, so go
2253          * ahead and start over and save the time.
2254          */
2255         if (!block_group || (!block_group_bits(block_group, data) &&
2256                              last_ptr && *last_ptr)) {
2257                 if (search_start != orig_search_start) {
2258                         if (last_ptr && *last_ptr)
2259                                 *last_ptr = 0;
2260                         search_start = orig_search_start;
2261                         goto new_group;
2262                 } else if (!chunk_alloc_done && allowed_chunk_alloc) {
2263                         ret = do_chunk_alloc(trans, root,
2264                                              num_bytes + 2 * 1024 * 1024,
2265                                              data, 1);
2266                         if (ret < 0)
2267                                 goto error;
2268                         BUG_ON(ret);
2269                         chunk_alloc_done = 1;
2270                         search_start = orig_search_start;
2271                         goto new_group;
2272                 } else {
2273                         ret = -ENOSPC;
2274                         goto error;
2275                 }
2276         }
2277
2278         /*
2279          * this is going to seach through all of the existing block groups it
2280          * can find, so if we don't find something we need to see if we can
2281          * allocate what we need.
2282          */
2283         ret = find_free_space(root, &block_group, &search_start,
2284                               total_needed, data);
2285         if (ret == -ENOSPC) {
2286                 /*
2287                  * instead of allocating, start at the original search start
2288                  * and see if there is something to be found, if not then we
2289                  * allocate
2290                  */
2291                 if (search_start != orig_search_start) {
2292                         if (last_ptr && *last_ptr) {
2293                                 *last_ptr = 0;
2294                                 total_needed += empty_cluster;
2295                         }
2296                         search_start = orig_search_start;
2297                         goto new_group;
2298                 }
2299
2300                 /*
2301                  * we've already allocated, we're pretty screwed
2302                  */
2303                 if (chunk_alloc_done) {
2304                         goto error;
2305                 } else if (!allowed_chunk_alloc && block_group &&
2306                            block_group_bits(block_group, data)) {
2307                         block_group->space_info->force_alloc = 1;
2308                         goto error;
2309                 } else if (!allowed_chunk_alloc) {
2310                         goto error;
2311                 }
2312
2313                 ret = do_chunk_alloc(trans, root, num_bytes + 2 * 1024 * 1024,
2314                                      data, 1);
2315                 if (ret < 0)
2316                         goto error;
2317
2318                 BUG_ON(ret);
2319                 chunk_alloc_done = 1;
2320                 if (block_group)
2321                         search_start = block_group->key.objectid +
2322                                 block_group->key.offset;
2323                 else
2324                         search_start = orig_search_start;
2325                 goto new_group;
2326         }
2327
2328         if (ret)
2329                 goto error;
2330
2331         search_start = stripe_align(root, search_start);
2332         ins->objectid = search_start;
2333         ins->offset = num_bytes;
2334
2335         if (ins->objectid + num_bytes >= search_end) {
2336                 search_start = orig_search_start;
2337                 if (chunk_alloc_done) {
2338                         ret = -ENOSPC;
2339                         goto error;
2340                 }
2341                 goto new_group;
2342         }
2343
2344         if (ins->objectid + num_bytes >
2345             block_group->key.objectid + block_group->key.offset) {
2346                 if (search_start == orig_search_start && chunk_alloc_done) {
2347                         ret = -ENOSPC;
2348                         goto error;
2349                 }
2350                 search_start = block_group->key.objectid +
2351                         block_group->key.offset;
2352                 goto new_group;
2353         }
2354
2355         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2356             ins->objectid < exclude_start + exclude_nr)) {
2357                 search_start = exclude_start + exclude_nr;
2358                 goto new_group;
2359         }
2360
2361         if (!(data & BTRFS_BLOCK_GROUP_DATA))
2362                 trans->block_group = block_group;
2363
2364         ins->offset = num_bytes;
2365         if (last_ptr) {
2366                 *last_ptr = ins->objectid + ins->offset;
2367                 if (*last_ptr ==
2368                     btrfs_super_total_bytes(&root->fs_info->super_copy))
2369                         *last_ptr = 0;
2370         }
2371
2372         ret = 0;
2373 error:
2374         return ret;
2375 }
2376
2377 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2378 {
2379         struct btrfs_block_group_cache *cache;
2380         struct list_head *l;
2381
2382         printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
2383                info->total_bytes - info->bytes_used - info->bytes_pinned -
2384                info->bytes_reserved, (info->full) ? "" : "not ");
2385
2386         spin_lock(&info->lock);
2387         list_for_each(l, &info->block_groups) {
2388                 cache = list_entry(l, struct btrfs_block_group_cache, list);
2389                 spin_lock(&cache->lock);
2390                 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
2391                        "%Lu pinned %Lu reserved\n",
2392                        cache->key.objectid, cache->key.offset,
2393                        btrfs_block_group_used(&cache->item),
2394                        cache->pinned, cache->reserved);
2395                 btrfs_dump_free_space(cache, bytes);
2396                 spin_unlock(&cache->lock);
2397         }
2398         spin_unlock(&info->lock);
2399 }
2400
2401 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2402                                   struct btrfs_root *root,
2403                                   u64 num_bytes, u64 min_alloc_size,
2404                                   u64 empty_size, u64 hint_byte,
2405                                   u64 search_end, struct btrfs_key *ins,
2406                                   u64 data)
2407 {
2408         int ret;
2409         u64 search_start = 0;
2410         u64 alloc_profile;
2411         struct btrfs_fs_info *info = root->fs_info;
2412         struct btrfs_block_group_cache *cache;
2413
2414         if (data) {
2415                 alloc_profile = info->avail_data_alloc_bits &
2416                                 info->data_alloc_profile;
2417                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2418         } else if (root == root->fs_info->chunk_root) {
2419                 alloc_profile = info->avail_system_alloc_bits &
2420                                 info->system_alloc_profile;
2421                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2422         } else {
2423                 alloc_profile = info->avail_metadata_alloc_bits &
2424                                 info->metadata_alloc_profile;
2425                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2426         }
2427 again:
2428         data = reduce_alloc_profile(root, data);
2429         /*
2430          * the only place that sets empty_size is btrfs_realloc_node, which
2431          * is not called recursively on allocations
2432          */
2433         if (empty_size || root->ref_cows) {
2434                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2435                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2436                                      2 * 1024 * 1024,
2437                                      BTRFS_BLOCK_GROUP_METADATA |
2438                                      (info->metadata_alloc_profile &
2439                                       info->avail_metadata_alloc_bits), 0);
2440                 }
2441                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2442                                      num_bytes + 2 * 1024 * 1024, data, 0);
2443         }
2444
2445         WARN_ON(num_bytes < root->sectorsize);
2446         ret = find_free_extent(trans, root, num_bytes, empty_size,
2447                                search_start, search_end, hint_byte, ins,
2448                                trans->alloc_exclude_start,
2449                                trans->alloc_exclude_nr, data);
2450
2451         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2452                 num_bytes = num_bytes >> 1;
2453                 num_bytes = num_bytes & ~(root->sectorsize - 1);
2454                 num_bytes = max(num_bytes, min_alloc_size);
2455                 do_chunk_alloc(trans, root->fs_info->extent_root,
2456                                num_bytes, data, 1);
2457                 goto again;
2458         }
2459         if (ret) {
2460                 struct btrfs_space_info *sinfo;
2461
2462                 sinfo = __find_space_info(root->fs_info, data);
2463                 printk("allocation failed flags %Lu, wanted %Lu\n",
2464                        data, num_bytes);
2465                 dump_space_info(sinfo, num_bytes);
2466                 BUG();
2467         }
2468         cache = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2469         if (!cache) {
2470                 printk(KERN_ERR "Unable to find block group for %Lu\n", ins->objectid);
2471                 return -ENOSPC;
2472         }
2473
2474         ret = btrfs_remove_free_space(cache, ins->objectid, ins->offset);
2475
2476         return ret;
2477 }
2478
2479 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2480 {
2481         struct btrfs_block_group_cache *cache;
2482
2483         maybe_lock_mutex(root);
2484         cache = btrfs_lookup_block_group(root->fs_info, start);
2485         if (!cache) {
2486                 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
2487                 maybe_unlock_mutex(root);
2488                 return -ENOSPC;
2489         }
2490         btrfs_add_free_space(cache, start, len);
2491         update_reserved_extents(root, start, len, 0);
2492         maybe_unlock_mutex(root);
2493         return 0;
2494 }
2495
2496 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2497                                   struct btrfs_root *root,
2498                                   u64 num_bytes, u64 min_alloc_size,
2499                                   u64 empty_size, u64 hint_byte,
2500                                   u64 search_end, struct btrfs_key *ins,
2501                                   u64 data)
2502 {
2503         int ret;
2504         maybe_lock_mutex(root);
2505         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2506                                      empty_size, hint_byte, search_end, ins,
2507                                      data);
2508         update_reserved_extents(root, ins->objectid, ins->offset, 1);
2509         maybe_unlock_mutex(root);
2510         return ret;
2511 }
2512
2513 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2514                                          struct btrfs_root *root, u64 parent,
2515                                          u64 root_objectid, u64 ref_generation,
2516                                          u64 owner, u64 owner_offset,
2517                                          struct btrfs_key *ins)
2518 {
2519         int ret;
2520         int pending_ret;
2521         u64 super_used;
2522         u64 root_used;
2523         u64 num_bytes = ins->offset;
2524         u32 sizes[2];
2525         struct btrfs_fs_info *info = root->fs_info;
2526         struct btrfs_root *extent_root = info->extent_root;
2527         struct btrfs_extent_item *extent_item;
2528         struct btrfs_extent_ref *ref;
2529         struct btrfs_path *path;
2530         struct btrfs_key keys[2];
2531
2532         if (parent == 0)
2533                 parent = ins->objectid;
2534
2535         /* block accounting for super block */
2536         spin_lock_irq(&info->delalloc_lock);
2537         super_used = btrfs_super_bytes_used(&info->super_copy);
2538         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2539         spin_unlock_irq(&info->delalloc_lock);
2540
2541         /* block accounting for root item */
2542         root_used = btrfs_root_used(&root->root_item);
2543         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2544
2545         if (root == extent_root) {
2546                 struct pending_extent_op *extent_op;
2547
2548                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2549                 BUG_ON(!extent_op);
2550
2551                 extent_op->type = PENDING_EXTENT_INSERT;
2552                 extent_op->bytenr = ins->objectid;
2553                 extent_op->num_bytes = ins->offset;
2554                 extent_op->parent = parent;
2555                 extent_op->orig_parent = 0;
2556                 extent_op->generation = ref_generation;
2557                 extent_op->orig_generation = 0;
2558                 extent_op->level = (int)owner;
2559
2560                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2561                                 ins->objectid + ins->offset - 1,
2562                                 EXTENT_LOCKED, GFP_NOFS);
2563                 set_state_private(&root->fs_info->extent_ins,
2564                                   ins->objectid, (unsigned long)extent_op);
2565                 goto update_block;
2566         }
2567
2568         memcpy(&keys[0], ins, sizeof(*ins));
2569         keys[1].objectid = ins->objectid;
2570         keys[1].type = BTRFS_EXTENT_REF_KEY;
2571         keys[1].offset = parent;
2572         sizes[0] = sizeof(*extent_item);
2573         sizes[1] = sizeof(*ref);
2574
2575         path = btrfs_alloc_path();
2576         BUG_ON(!path);
2577
2578         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2579                                        sizes, 2);
2580         BUG_ON(ret);
2581
2582         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2583                                      struct btrfs_extent_item);
2584         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2585         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2586                              struct btrfs_extent_ref);
2587
2588         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2589         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2590         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2591         btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2592         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
2593
2594         btrfs_mark_buffer_dirty(path->nodes[0]);
2595
2596         trans->alloc_exclude_start = 0;
2597         trans->alloc_exclude_nr = 0;
2598         btrfs_free_path(path);
2599         finish_current_insert(trans, extent_root);
2600         pending_ret = del_pending_extents(trans, extent_root);
2601
2602         if (ret)
2603                 goto out;
2604         if (pending_ret) {
2605                 ret = pending_ret;
2606                 goto out;
2607         }
2608
2609 update_block:
2610         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2611         if (ret) {
2612                 printk("update block group failed for %Lu %Lu\n",
2613                        ins->objectid, ins->offset);
2614                 BUG();
2615         }
2616 out:
2617         return ret;
2618 }
2619
2620 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2621                                 struct btrfs_root *root, u64 parent,
2622                                 u64 root_objectid, u64 ref_generation,
2623                                 u64 owner, u64 owner_offset,
2624                                 struct btrfs_key *ins)
2625 {
2626         int ret;
2627
2628         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2629                 return 0;
2630         maybe_lock_mutex(root);
2631         ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2632                                             root_objectid, ref_generation,
2633                                             owner, owner_offset, ins);
2634         update_reserved_extents(root, ins->objectid, ins->offset, 0);
2635         maybe_unlock_mutex(root);
2636         return ret;
2637 }
2638
2639 /*
2640  * this is used by the tree logging recovery code.  It records that
2641  * an extent has been allocated and makes sure to clear the free
2642  * space cache bits as well
2643  */
2644 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2645                                 struct btrfs_root *root, u64 parent,
2646                                 u64 root_objectid, u64 ref_generation,
2647                                 u64 owner, u64 owner_offset,
2648                                 struct btrfs_key *ins)
2649 {
2650         int ret;
2651         struct btrfs_block_group_cache *block_group;
2652
2653         maybe_lock_mutex(root);
2654         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2655         cache_block_group(root, block_group);
2656
2657         ret = btrfs_remove_free_space(block_group, ins->objectid, ins->offset);
2658         BUG_ON(ret);
2659         ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2660                                             root_objectid, ref_generation,
2661                                             owner, owner_offset, ins);
2662         maybe_unlock_mutex(root);
2663         return ret;
2664 }
2665
2666 /*
2667  * finds a free extent and does all the dirty work required for allocation
2668  * returns the key for the extent through ins, and a tree buffer for
2669  * the first block of the extent through buf.
2670  *
2671  * returns 0 if everything worked, non-zero otherwise.
2672  */
2673 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2674                        struct btrfs_root *root,
2675                        u64 num_bytes, u64 parent, u64 min_alloc_size,
2676                        u64 root_objectid, u64 ref_generation,
2677                        u64 owner_objectid, u64 owner_offset,
2678                        u64 empty_size, u64 hint_byte,
2679                        u64 search_end, struct btrfs_key *ins, u64 data)
2680 {
2681         int ret;
2682
2683         maybe_lock_mutex(root);
2684
2685         ret = __btrfs_reserve_extent(trans, root, num_bytes,
2686                                      min_alloc_size, empty_size, hint_byte,
2687                                      search_end, ins, data);
2688         BUG_ON(ret);
2689         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2690                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2691                                         root_objectid, ref_generation,
2692                                         owner_objectid, owner_offset, ins);
2693                 BUG_ON(ret);
2694
2695         } else {
2696                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2697         }
2698         maybe_unlock_mutex(root);
2699         return ret;
2700 }
2701
2702 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2703                                             struct btrfs_root *root,
2704                                             u64 bytenr, u32 blocksize)
2705 {
2706         struct extent_buffer *buf;
2707
2708         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2709         if (!buf)
2710                 return ERR_PTR(-ENOMEM);
2711         btrfs_set_header_generation(buf, trans->transid);
2712         btrfs_tree_lock(buf);
2713         clean_tree_block(trans, root, buf);
2714         btrfs_set_buffer_uptodate(buf);
2715         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2716                 set_extent_dirty(&root->dirty_log_pages, buf->start,
2717                          buf->start + buf->len - 1, GFP_NOFS);
2718         } else {
2719                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2720                          buf->start + buf->len - 1, GFP_NOFS);
2721         }
2722         trans->blocks_used++;
2723         return buf;
2724 }
2725
2726 /*
2727  * helper function to allocate a block for a given tree
2728  * returns the tree buffer or NULL.
2729  */
2730 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2731                                              struct btrfs_root *root,
2732                                              u32 blocksize, u64 parent,
2733                                              u64 root_objectid,
2734                                              u64 ref_generation,
2735                                              int level,
2736                                              u64 hint,
2737                                              u64 empty_size)
2738 {
2739         struct btrfs_key ins;
2740         int ret;
2741         struct extent_buffer *buf;
2742
2743         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
2744                                  root_objectid, ref_generation, level, 0,
2745                                  empty_size, hint, (u64)-1, &ins, 0);
2746         if (ret) {
2747                 BUG_ON(ret > 0);
2748                 return ERR_PTR(ret);
2749         }
2750
2751         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2752         return buf;
2753 }
2754
2755 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2756                         struct btrfs_root *root, struct extent_buffer *leaf)
2757 {
2758         u64 leaf_owner;
2759         u64 leaf_generation;
2760         struct btrfs_key key;
2761         struct btrfs_file_extent_item *fi;
2762         int i;
2763         int nritems;
2764         int ret;
2765
2766         BUG_ON(!btrfs_is_leaf(leaf));
2767         nritems = btrfs_header_nritems(leaf);
2768         leaf_owner = btrfs_header_owner(leaf);
2769         leaf_generation = btrfs_header_generation(leaf);
2770
2771         for (i = 0; i < nritems; i++) {
2772                 u64 disk_bytenr;
2773                 cond_resched();
2774
2775                 btrfs_item_key_to_cpu(leaf, &key, i);
2776                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2777                         continue;
2778                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2779                 if (btrfs_file_extent_type(leaf, fi) ==
2780                     BTRFS_FILE_EXTENT_INLINE)
2781                         continue;
2782                 /*
2783                  * FIXME make sure to insert a trans record that
2784                  * repeats the snapshot del on crash
2785                  */
2786                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2787                 if (disk_bytenr == 0)
2788                         continue;
2789
2790                 mutex_lock(&root->fs_info->alloc_mutex);
2791                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2792                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2793                                 leaf->start, leaf_owner, leaf_generation,
2794                                 key.objectid, key.offset, 0);
2795                 mutex_unlock(&root->fs_info->alloc_mutex);
2796                 BUG_ON(ret);
2797
2798                 atomic_inc(&root->fs_info->throttle_gen);
2799                 wake_up(&root->fs_info->transaction_throttle);
2800                 cond_resched();
2801         }
2802         return 0;
2803 }
2804
2805 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2806                                         struct btrfs_root *root,
2807                                         struct btrfs_leaf_ref *ref)
2808 {
2809         int i;
2810         int ret;
2811         struct btrfs_extent_info *info = ref->extents;
2812
2813         for (i = 0; i < ref->nritems; i++) {
2814                 mutex_lock(&root->fs_info->alloc_mutex);
2815                 ret = __btrfs_free_extent(trans, root, info->bytenr,
2816                                           info->num_bytes, ref->bytenr,
2817                                           ref->owner, ref->generation,
2818                                           info->objectid, info->offset, 0);
2819                 mutex_unlock(&root->fs_info->alloc_mutex);
2820
2821                 atomic_inc(&root->fs_info->throttle_gen);
2822                 wake_up(&root->fs_info->transaction_throttle);
2823                 cond_resched();
2824
2825                 BUG_ON(ret);
2826                 info++;
2827         }
2828
2829         return 0;
2830 }
2831
2832 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2833                               u32 *refs)
2834 {
2835         int ret;
2836
2837         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
2838         BUG_ON(ret);
2839
2840 #if 0 // some debugging code in case we see problems here
2841         /* if the refs count is one, it won't get increased again.  But
2842          * if the ref count is > 1, someone may be decreasing it at
2843          * the same time we are.
2844          */
2845         if (*refs != 1) {
2846                 struct extent_buffer *eb = NULL;
2847                 eb = btrfs_find_create_tree_block(root, start, len);
2848                 if (eb)
2849                         btrfs_tree_lock(eb);
2850
2851                 mutex_lock(&root->fs_info->alloc_mutex);
2852                 ret = lookup_extent_ref(NULL, root, start, len, refs);
2853                 BUG_ON(ret);
2854                 mutex_unlock(&root->fs_info->alloc_mutex);
2855
2856                 if (eb) {
2857                         btrfs_tree_unlock(eb);
2858                         free_extent_buffer(eb);
2859                 }
2860                 if (*refs == 1) {
2861                         printk("block %llu went down to one during drop_snap\n",
2862                                (unsigned long long)start);
2863                 }
2864
2865         }
2866 #endif
2867
2868         cond_resched();
2869         return ret;
2870 }
2871
2872 /*
2873  * helper function for drop_snapshot, this walks down the tree dropping ref
2874  * counts as it goes.
2875  */
2876 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2877                                    struct btrfs_root *root,
2878                                    struct btrfs_path *path, int *level)
2879 {
2880         u64 root_owner;
2881         u64 root_gen;
2882         u64 bytenr;
2883         u64 ptr_gen;
2884         struct extent_buffer *next;
2885         struct extent_buffer *cur;
2886         struct extent_buffer *parent;
2887         struct btrfs_leaf_ref *ref;
2888         u32 blocksize;
2889         int ret;
2890         u32 refs;
2891
2892         WARN_ON(*level < 0);
2893         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2894         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2895                                 path->nodes[*level]->len, &refs);
2896         BUG_ON(ret);
2897         if (refs > 1)
2898                 goto out;
2899
2900         /*
2901          * walk down to the last node level and free all the leaves
2902          */
2903         while(*level >= 0) {
2904                 WARN_ON(*level < 0);
2905                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2906                 cur = path->nodes[*level];
2907
2908                 if (btrfs_header_level(cur) != *level)
2909                         WARN_ON(1);
2910
2911                 if (path->slots[*level] >=
2912                     btrfs_header_nritems(cur))
2913                         break;
2914                 if (*level == 0) {
2915                         ret = btrfs_drop_leaf_ref(trans, root, cur);
2916                         BUG_ON(ret);
2917                         break;
2918                 }
2919                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2920                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2921                 blocksize = btrfs_level_size(root, *level - 1);
2922
2923                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2924                 BUG_ON(ret);
2925                 if (refs != 1) {
2926                         parent = path->nodes[*level];
2927                         root_owner = btrfs_header_owner(parent);
2928                         root_gen = btrfs_header_generation(parent);
2929                         path->slots[*level]++;
2930
2931                         mutex_lock(&root->fs_info->alloc_mutex);
2932                         ret = __btrfs_free_extent(trans, root, bytenr,
2933                                                 blocksize, parent->start,
2934                                                 root_owner, root_gen, 0, 0, 1);
2935                         BUG_ON(ret);
2936                         mutex_unlock(&root->fs_info->alloc_mutex);
2937
2938                         atomic_inc(&root->fs_info->throttle_gen);
2939                         wake_up(&root->fs_info->transaction_throttle);
2940                         cond_resched();
2941
2942                         continue;
2943                 }
2944                 /*
2945                  * at this point, we have a single ref, and since the
2946                  * only place referencing this extent is a dead root
2947                  * the reference count should never go higher.
2948                  * So, we don't need to check it again
2949                  */
2950                 if (*level == 1) {
2951                         ref = btrfs_lookup_leaf_ref(root, bytenr);
2952                         if (ref && ref->generation != ptr_gen) {
2953                                 btrfs_free_leaf_ref(root, ref);
2954                                 ref = NULL;
2955                         }
2956                         if (ref) {
2957                                 ret = cache_drop_leaf_ref(trans, root, ref);
2958                                 BUG_ON(ret);
2959                                 btrfs_remove_leaf_ref(root, ref);
2960                                 btrfs_free_leaf_ref(root, ref);
2961                                 *level = 0;
2962                                 break;
2963                         }
2964                         if (printk_ratelimit())
2965                                 printk("leaf ref miss for bytenr %llu\n",
2966                                        (unsigned long long)bytenr);
2967                 }
2968                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2969                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2970                         free_extent_buffer(next);
2971
2972                         next = read_tree_block(root, bytenr, blocksize,
2973                                                ptr_gen);
2974                         cond_resched();
2975 #if 0
2976                         /*
2977                          * this is a debugging check and can go away
2978                          * the ref should never go all the way down to 1
2979                          * at this point
2980                          */
2981                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2982                                                 &refs);
2983                         BUG_ON(ret);
2984                         WARN_ON(refs != 1);
2985 #endif
2986                 }
2987                 WARN_ON(*level <= 0);
2988                 if (path->nodes[*level-1])
2989                         free_extent_buffer(path->nodes[*level-1]);
2990                 path->nodes[*level-1] = next;
2991                 *level = btrfs_header_level(next);
2992                 path->slots[*level] = 0;
2993                 cond_resched();
2994         }
2995 out:
2996         WARN_ON(*level < 0);
2997         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2998
2999         if (path->nodes[*level] == root->node) {
3000                 parent = path->nodes[*level];
3001                 bytenr = path->nodes[*level]->start;
3002         } else {
3003                 parent = path->nodes[*level + 1];
3004                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3005         }
3006
3007         blocksize = btrfs_level_size(root, *level);
3008         root_owner = btrfs_header_owner(parent);
3009         root_gen = btrfs_header_generation(parent);
3010
3011         mutex_lock(&root->fs_info->alloc_mutex);
3012         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3013                                   parent->start, root_owner, root_gen,
3014                                   0, 0, 1);
3015         mutex_unlock(&root->fs_info->alloc_mutex);
3016         free_extent_buffer(path->nodes[*level]);
3017         path->nodes[*level] = NULL;
3018         *level += 1;
3019         BUG_ON(ret);
3020
3021         cond_resched();
3022         return 0;
3023 }
3024
3025 /*
3026  * helper for dropping snapshots.  This walks back up the tree in the path
3027  * to find the first node higher up where we haven't yet gone through
3028  * all the slots
3029  */
3030 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3031                                  struct btrfs_root *root,
3032                                  struct btrfs_path *path, int *level)
3033 {
3034         u64 root_owner;
3035         u64 root_gen;
3036         struct btrfs_root_item *root_item = &root->root_item;
3037         int i;
3038         int slot;
3039         int ret;
3040
3041         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3042                 slot = path->slots[i];
3043                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3044                         struct extent_buffer *node;
3045                         struct btrfs_disk_key disk_key;
3046                         node = path->nodes[i];
3047                         path->slots[i]++;
3048                         *level = i;
3049                         WARN_ON(*level == 0);
3050                         btrfs_node_key(node, &disk_key, path->slots[i]);
3051                         memcpy(&root_item->drop_progress,
3052                                &disk_key, sizeof(disk_key));
3053                         root_item->drop_level = i;
3054                         return 0;
3055                 } else {
3056                         struct extent_buffer *parent;
3057                         if (path->nodes[*level] == root->node)
3058                                 parent = path->nodes[*level];
3059                         else
3060                                 parent = path->nodes[*level + 1];
3061
3062                         root_owner = btrfs_header_owner(parent);
3063                         root_gen = btrfs_header_generation(parent);
3064                         ret = btrfs_free_extent(trans, root,
3065                                                 path->nodes[*level]->start,
3066                                                 path->nodes[*level]->len,
3067                                                 parent->start,
3068                                                 root_owner, root_gen, 0, 0, 1);
3069                         BUG_ON(ret);
3070                         free_extent_buffer(path->nodes[*level]);
3071                         path->nodes[*level] = NULL;
3072                         *level = i + 1;
3073                 }
3074         }
3075         return 1;
3076 }
3077
3078 /*
3079  * drop the reference count on the tree rooted at 'snap'.  This traverses
3080  * the tree freeing any blocks that have a ref count of zero after being
3081  * decremented.
3082  */
3083 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3084                         *root)
3085 {
3086         int ret = 0;
3087         int wret;
3088         int level;
3089         struct btrfs_path *path;
3090         int i;
3091         int orig_level;
3092         struct btrfs_root_item *root_item = &root->root_item;
3093
3094         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3095         path = btrfs_alloc_path();
3096         BUG_ON(!path);
3097
3098         level = btrfs_header_level(root->node);
3099         orig_level = level;
3100         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3101                 path->nodes[level] = root->node;
3102                 extent_buffer_get(root->node);
3103                 path->slots[level] = 0;
3104         } else {
3105                 struct btrfs_key key;
3106                 struct btrfs_disk_key found_key;
3107                 struct extent_buffer *node;
3108
3109                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3110                 level = root_item->drop_level;
3111                 path->lowest_level = level;
3112                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3113                 if (wret < 0) {
3114                         ret = wret;
3115                         goto out;
3116                 }
3117                 node = path->nodes[level];
3118                 btrfs_node_key(node, &found_key, path->slots[level]);
3119                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3120                                sizeof(found_key)));
3121                 /*
3122                  * unlock our path, this is safe because only this
3123                  * function is allowed to delete this snapshot
3124                  */
3125                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3126                         if (path->nodes[i] && path->locks[i]) {
3127                                 path->locks[i] = 0;
3128                                 btrfs_tree_unlock(path->nodes[i]);
3129                         }
3130                 }
3131         }
3132         while(1) {
3133                 wret = walk_down_tree(trans, root, path, &level);
3134                 if (wret > 0)
3135                         break;
3136                 if (wret < 0)
3137                         ret = wret;
3138
3139                 wret = walk_up_tree(trans, root, path, &level);
3140                 if (wret > 0)
3141                         break;
3142                 if (wret < 0)
3143                         ret = wret;
3144                 if (trans->transaction->in_commit) {
3145                         ret = -EAGAIN;
3146                         break;
3147                 }
3148                 atomic_inc(&root->fs_info->throttle_gen);
3149                 wake_up(&root->fs_info->transaction_throttle);
3150         }
3151         for (i = 0; i <= orig_level; i++) {
3152                 if (path->nodes[i]) {
3153                         free_extent_buffer(path->nodes[i]);
3154                         path->nodes[i] = NULL;
3155                 }
3156         }
3157 out:
3158         btrfs_free_path(path);
3159         return ret;
3160 }
3161
3162 static unsigned long calc_ra(unsigned long start, unsigned long last,
3163                              unsigned long nr)
3164 {
3165         return min(last, start + nr - 1);
3166 }
3167
3168 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3169                                          u64 len)
3170 {
3171         u64 page_start;
3172         u64 page_end;
3173         unsigned long first_index;
3174         unsigned long last_index;
3175         unsigned long i;
3176         struct page *page;
3177         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3178         struct file_ra_state *ra;
3179         struct btrfs_ordered_extent *ordered;
3180         unsigned int total_read = 0;
3181         unsigned int total_dirty = 0;
3182         int ret = 0;
3183
3184         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3185
3186         mutex_lock(&inode->i_mutex);
3187         first_index = start >> PAGE_CACHE_SHIFT;
3188         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3189
3190         /* make sure the dirty trick played by the caller work */
3191         ret = invalidate_inode_pages2_range(inode->i_mapping,
3192                                             first_index, last_index);
3193         if (ret)
3194                 goto out_unlock;
3195
3196         file_ra_state_init(ra, inode->i_mapping);
3197
3198         for (i = first_index ; i <= last_index; i++) {
3199                 if (total_read % ra->ra_pages == 0) {
3200                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3201                                        calc_ra(i, last_index, ra->ra_pages));
3202                 }
3203                 total_read++;
3204 again:
3205                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3206                         BUG_ON(1);
3207                 page = grab_cache_page(inode->i_mapping, i);
3208                 if (!page) {
3209                         ret = -ENOMEM;
3210                         goto out_unlock;
3211                 }
3212                 if (!PageUptodate(page)) {
3213                         btrfs_readpage(NULL, page);
3214                         lock_page(page);
3215                         if (!PageUptodate(page)) {
3216                                 unlock_page(page);
3217                                 page_cache_release(page);
3218                                 ret = -EIO;
3219                                 goto out_unlock;
3220                         }
3221                 }
3222                 wait_on_page_writeback(page);
3223
3224                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3225                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3226                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3227
3228                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3229                 if (ordered) {
3230                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3231                         unlock_page(page);
3232                         page_cache_release(page);
3233                         btrfs_start_ordered_extent(inode, ordered, 1);
3234                         btrfs_put_ordered_extent(ordered);
3235                         goto again;
3236                 }
3237                 set_page_extent_mapped(page);
3238
3239                 btrfs_set_extent_delalloc(inode, page_start, page_end);
3240                 if (i == first_index)
3241                         set_extent_bits(io_tree, page_start, page_end,
3242                                         EXTENT_BOUNDARY, GFP_NOFS);
3243
3244                 set_page_dirty(page);
3245                 total_dirty++;
3246
3247                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3248                 unlock_page(page);
3249                 page_cache_release(page);
3250         }
3251
3252 out_unlock:
3253         kfree(ra);
3254         mutex_unlock(&inode->i_mutex);
3255         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3256         return ret;
3257 }
3258
3259 static int noinline relocate_data_extent(struct inode *reloc_inode,
3260                                          struct btrfs_key *extent_key,
3261                                          u64 offset)
3262 {
3263         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3264         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3265         struct extent_map *em;
3266
3267         em = alloc_extent_map(GFP_NOFS);
3268         BUG_ON(!em || IS_ERR(em));
3269
3270         em->start = extent_key->objectid - offset;
3271         em->len = extent_key->offset;
3272         em->block_start = extent_key->objectid;
3273         em->bdev = root->fs_info->fs_devices->latest_bdev;
3274         set_bit(EXTENT_FLAG_PINNED, &em->flags);
3275
3276         /* setup extent map to cheat btrfs_readpage */
3277         mutex_lock(&BTRFS_I(reloc_inode)->extent_mutex);
3278         while (1) {
3279                 int ret;
3280                 spin_lock(&em_tree->lock);
3281                 ret = add_extent_mapping(em_tree, em);
3282                 spin_unlock(&em_tree->lock);
3283                 if (ret != -EEXIST) {
3284                         free_extent_map(em);
3285                         break;
3286                 }
3287                 btrfs_drop_extent_cache(reloc_inode, em->start,
3288                                         em->start + em->len - 1, 0);
3289         }
3290         mutex_unlock(&BTRFS_I(reloc_inode)->extent_mutex);
3291
3292         return relocate_inode_pages(reloc_inode, extent_key->objectid - offset,
3293                                     extent_key->offset);
3294 }
3295
3296 struct btrfs_ref_path {
3297         u64 extent_start;
3298         u64 nodes[BTRFS_MAX_LEVEL];
3299         u64 root_objectid;
3300         u64 root_generation;
3301         u64 owner_objectid;
3302         u64 owner_offset;
3303         u32 num_refs;
3304         int lowest_level;
3305         int current_level;
3306 };
3307
3308 struct disk_extent {
3309         u64 disk_bytenr;
3310         u64 disk_num_bytes;
3311         u64 offset;
3312         u64 num_bytes;
3313 };
3314
3315 static int is_cowonly_root(u64 root_objectid)
3316 {
3317         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3318             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3319             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3320             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3321             root_objectid == BTRFS_TREE_LOG_OBJECTID)
3322                 return 1;
3323         return 0;
3324 }
3325
3326 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
3327                                     struct btrfs_root *extent_root,
3328                                     struct btrfs_ref_path *ref_path,
3329                                     int first_time)
3330 {
3331         struct extent_buffer *leaf;
3332         struct btrfs_path *path;
3333         struct btrfs_extent_ref *ref;
3334         struct btrfs_key key;
3335         struct btrfs_key found_key;
3336         u64 bytenr;
3337         u32 nritems;
3338         int level;
3339         int ret = 1;
3340
3341         path = btrfs_alloc_path();
3342         if (!path)
3343                 return -ENOMEM;
3344
3345         mutex_lock(&extent_root->fs_info->alloc_mutex);
3346
3347         if (first_time) {
3348                 ref_path->lowest_level = -1;
3349                 ref_path->current_level = -1;
3350                 goto walk_up;
3351         }
3352 walk_down:
3353         level = ref_path->current_level - 1;
3354         while (level >= -1) {
3355                 u64 parent;
3356                 if (level < ref_path->lowest_level)
3357                         break;
3358
3359                 if (level >= 0) {
3360                         bytenr = ref_path->nodes[level];
3361                 } else {
3362                         bytenr = ref_path->extent_start;
3363                 }
3364                 BUG_ON(bytenr == 0);
3365
3366                 parent = ref_path->nodes[level + 1];
3367                 ref_path->nodes[level + 1] = 0;
3368                 ref_path->current_level = level;
3369                 BUG_ON(parent == 0);
3370
3371                 key.objectid = bytenr;
3372                 key.offset = parent + 1;
3373                 key.type = BTRFS_EXTENT_REF_KEY;
3374
3375                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3376                 if (ret < 0)
3377                         goto out;
3378                 BUG_ON(ret == 0);
3379
3380                 leaf = path->nodes[0];
3381                 nritems = btrfs_header_nritems(leaf);
3382                 if (path->slots[0] >= nritems) {
3383                         ret = btrfs_next_leaf(extent_root, path);
3384                         if (ret < 0)
3385                                 goto out;
3386                         if (ret > 0)
3387                                 goto next;
3388                         leaf = path->nodes[0];
3389                 }
3390
3391                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3392                 if (found_key.objectid == bytenr &&
3393                                 found_key.type == BTRFS_EXTENT_REF_KEY)
3394                         goto found;
3395 next:
3396                 level--;
3397                 btrfs_release_path(extent_root, path);
3398                 if (need_resched()) {
3399                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
3400                         cond_resched();
3401                         mutex_lock(&extent_root->fs_info->alloc_mutex);
3402                 }
3403         }
3404         /* reached lowest level */
3405         ret = 1;
3406         goto out;
3407 walk_up:
3408         level = ref_path->current_level;
3409         while (level < BTRFS_MAX_LEVEL - 1) {
3410                 u64 ref_objectid;
3411                 if (level >= 0) {
3412                         bytenr = ref_path->nodes[level];
3413                 } else {
3414                         bytenr = ref_path->extent_start;
3415                 }
3416                 BUG_ON(bytenr == 0);
3417
3418                 key.objectid = bytenr;
3419                 key.offset = 0;
3420                 key.type = BTRFS_EXTENT_REF_KEY;
3421
3422                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3423                 if (ret < 0)
3424                         goto out;
3425
3426                 leaf = path->nodes[0];
3427                 nritems = btrfs_header_nritems(leaf);
3428                 if (path->slots[0] >= nritems) {
3429                         ret = btrfs_next_leaf(extent_root, path);
3430                         if (ret < 0)
3431                                 goto out;
3432                         if (ret > 0) {
3433                                 /* the extent was freed by someone */
3434                                 if (ref_path->lowest_level == level)
3435                                         goto out;
3436                                 btrfs_release_path(extent_root, path);
3437                                 goto walk_down;
3438                         }
3439                         leaf = path->nodes[0];
3440                 }
3441
3442                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3443                 if (found_key.objectid != bytenr ||
3444                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
3445                         /* the extent was freed by someone */
3446                         if (ref_path->lowest_level == level) {
3447                                 ret = 1;
3448                                 goto out;
3449                         }
3450                         btrfs_release_path(extent_root, path);
3451                         goto walk_down;
3452                 }
3453 found:
3454                 ref = btrfs_item_ptr(leaf, path->slots[0],
3455                                 struct btrfs_extent_ref);
3456                 ref_objectid = btrfs_ref_objectid(leaf, ref);
3457                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3458                         if (first_time) {
3459                                 level = (int)ref_objectid;
3460                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
3461                                 ref_path->lowest_level = level;
3462                                 ref_path->current_level = level;
3463                                 ref_path->nodes[level] = bytenr;
3464                         } else {
3465                                 WARN_ON(ref_objectid != level);
3466                         }
3467                 } else {
3468                         WARN_ON(level != -1);
3469                 }
3470                 first_time = 0;
3471
3472                 if (ref_path->lowest_level == level) {
3473                         ref_path->owner_objectid = ref_objectid;
3474                         ref_path->owner_offset = btrfs_ref_offset(leaf, ref);
3475                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
3476                 }
3477
3478                 /*
3479                  * the block is tree root or the block isn't in reference
3480                  * counted tree.
3481                  */
3482                 if (found_key.objectid == found_key.offset ||
3483                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
3484                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3485                         ref_path->root_generation =
3486                                 btrfs_ref_generation(leaf, ref);
3487                         if (level < 0) {
3488                                 /* special reference from the tree log */
3489                                 ref_path->nodes[0] = found_key.offset;
3490                                 ref_path->current_level = 0;
3491                         }
3492                         ret = 0;
3493                         goto out;
3494                 }
3495
3496                 level++;
3497                 BUG_ON(ref_path->nodes[level] != 0);
3498                 ref_path->nodes[level] = found_key.offset;
3499                 ref_path->current_level = level;
3500
3501                 /*
3502                  * the reference was created in the running transaction,
3503                  * no need to continue walking up.
3504                  */
3505                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
3506                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3507                         ref_path->root_generation =
3508                                 btrfs_ref_generation(leaf, ref);
3509                         ret = 0;
3510                         goto out;
3511                 }
3512
3513                 btrfs_release_path(extent_root, path);
3514                 if (need_resched()) {
3515                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
3516                         cond_resched();
3517                         mutex_lock(&extent_root->fs_info->alloc_mutex);
3518                 }
3519         }
3520         /* reached max tree level, but no tree root found. */
3521         BUG();
3522 out:
3523         mutex_unlock(&extent_root->fs_info->alloc_mutex);
3524         btrfs_free_path(path);
3525         return ret;
3526 }
3527
3528 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
3529                                 struct btrfs_root *extent_root,
3530                                 struct btrfs_ref_path *ref_path,
3531                                 u64 extent_start)
3532 {
3533         memset(ref_path, 0, sizeof(*ref_path));
3534         ref_path->extent_start = extent_start;
3535
3536         return __next_ref_path(trans, extent_root, ref_path, 1);
3537 }
3538
3539 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
3540                                struct btrfs_root *extent_root,
3541                                struct btrfs_ref_path *ref_path)
3542 {
3543         return __next_ref_path(trans, extent_root, ref_path, 0);
3544 }
3545
3546 static int noinline get_new_locations(struct inode *reloc_inode,
3547                                       struct btrfs_key *extent_key,
3548                                       u64 offset, int no_fragment,
3549                                       struct disk_extent **extents,
3550                                       int *nr_extents)
3551 {
3552         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3553         struct btrfs_path *path;
3554         struct btrfs_file_extent_item *fi;
3555         struct extent_buffer *leaf;
3556         struct disk_extent *exts = *extents;
3557         struct btrfs_key found_key;
3558         u64 cur_pos;
3559         u64 last_byte;
3560         u32 nritems;
3561         int nr = 0;
3562         int max = *nr_extents;
3563         int ret;
3564
3565         WARN_ON(!no_fragment && *extents);
3566         if (!exts) {
3567                 max = 1;
3568                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
3569                 if (!exts)
3570                         return -ENOMEM;
3571         }
3572
3573         path = btrfs_alloc_path();
3574         BUG_ON(!path);
3575
3576         cur_pos = extent_key->objectid - offset;
3577         last_byte = extent_key->objectid + extent_key->offset;
3578         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
3579                                        cur_pos, 0);
3580         if (ret < 0)
3581                 goto out;
3582         if (ret > 0) {
3583                 ret = -ENOENT;
3584                 goto out;
3585         }
3586
3587         while (1) {
3588                 leaf = path->nodes[0];
3589                 nritems = btrfs_header_nritems(leaf);
3590                 if (path->slots[0] >= nritems) {
3591                         ret = btrfs_next_leaf(root, path);
3592                         if (ret < 0)
3593                                 goto out;
3594                         if (ret > 0)
3595                                 break;
3596                         leaf = path->nodes[0];
3597                 }
3598
3599                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3600                 if (found_key.offset != cur_pos ||
3601                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
3602                     found_key.objectid != reloc_inode->i_ino)
3603                         break;
3604
3605                 fi = btrfs_item_ptr(leaf, path->slots[0],
3606                                     struct btrfs_file_extent_item);
3607                 if (btrfs_file_extent_type(leaf, fi) !=
3608                     BTRFS_FILE_EXTENT_REG ||
3609                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
3610                         break;
3611
3612                 if (nr == max) {
3613                         struct disk_extent *old = exts;
3614                         max *= 2;
3615                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
3616                         memcpy(exts, old, sizeof(*exts) * nr);
3617                         if (old != *extents)
3618                                 kfree(old);
3619                 }
3620
3621                 exts[nr].disk_bytenr =
3622                         btrfs_file_extent_disk_bytenr(leaf, fi);
3623                 exts[nr].disk_num_bytes =
3624                         btrfs_file_extent_disk_num_bytes(leaf, fi);
3625                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
3626                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3627                 WARN_ON(exts[nr].offset > 0);
3628                 WARN_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
3629
3630                 cur_pos += exts[nr].num_bytes;
3631                 nr++;
3632
3633                 if (cur_pos + offset >= last_byte)
3634                         break;
3635
3636                 if (no_fragment) {
3637                         ret = 1;
3638                         goto out;
3639                 }
3640                 path->slots[0]++;
3641         }
3642
3643         WARN_ON(cur_pos + offset > last_byte);
3644         if (cur_pos + offset < last_byte) {
3645                 ret = -ENOENT;
3646                 goto out;
3647         }
3648         ret = 0;
3649 out:
3650         btrfs_free_path(path);
3651         if (ret) {
3652                 if (exts != *extents)
3653                         kfree(exts);
3654         } else {
3655                 *extents = exts;
3656                 *nr_extents = nr;
3657         }
3658         return ret;
3659 }
3660
3661 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
3662                                         struct btrfs_root *root,
3663                                         struct btrfs_path *path,
3664                                         struct btrfs_key *extent_key,
3665                                         struct btrfs_key *leaf_key,
3666                                         struct btrfs_ref_path *ref_path,
3667                                         struct disk_extent *new_extents,
3668                                         int nr_extents)
3669 {
3670         struct extent_buffer *leaf;
3671         struct btrfs_file_extent_item *fi;
3672         struct inode *inode = NULL;
3673         struct btrfs_key key;
3674         u64 lock_start = 0;
3675         u64 lock_end = 0;
3676         u64 num_bytes;
3677         u64 ext_offset;
3678         u64 first_pos;
3679         u32 nritems;
3680         int extent_locked = 0;
3681         int ret;
3682
3683         first_pos = ref_path->owner_offset;
3684         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3685                 key.objectid = ref_path->owner_objectid;
3686                 key.offset = ref_path->owner_offset;
3687                 key.type = BTRFS_EXTENT_DATA_KEY;
3688         } else {
3689                 memcpy(&key, leaf_key, sizeof(key));
3690         }
3691
3692         while (1) {
3693                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3694                 if (ret < 0)
3695                         goto out;
3696
3697                 leaf = path->nodes[0];
3698                 nritems = btrfs_header_nritems(leaf);
3699 next:
3700                 if (extent_locked && ret > 0) {
3701                         /*
3702                          * the file extent item was modified by someone
3703                          * before the extent got locked.
3704                          */
3705                         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3706                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3707                                       lock_end, GFP_NOFS);
3708                         extent_locked = 0;
3709                 }
3710
3711                 if (path->slots[0] >= nritems) {
3712                         if (ref_path->owner_objectid ==
3713                             BTRFS_MULTIPLE_OBJECTIDS)
3714                                 break;
3715
3716                         BUG_ON(extent_locked);
3717                         ret = btrfs_next_leaf(root, path);
3718                         if (ret < 0)
3719                                 goto out;
3720                         if (ret > 0)
3721                                 break;
3722                         leaf = path->nodes[0];
3723                         nritems = btrfs_header_nritems(leaf);
3724                 }
3725
3726                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3727
3728                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3729                         if ((key.objectid > ref_path->owner_objectid) ||
3730                             (key.objectid == ref_path->owner_objectid &&
3731                              key.type > BTRFS_EXTENT_DATA_KEY) ||
3732                             (key.offset >= first_pos + extent_key->offset))
3733                                 break;
3734                 }
3735
3736                 if (inode && key.objectid != inode->i_ino) {
3737                         BUG_ON(extent_locked);
3738                         btrfs_release_path(root, path);
3739                         mutex_unlock(&inode->i_mutex);
3740                         iput(inode);
3741                         inode = NULL;
3742                         continue;
3743                 }
3744
3745                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
3746                         path->slots[0]++;
3747                         ret = 1;
3748                         goto next;
3749                 }
3750                 fi = btrfs_item_ptr(leaf, path->slots[0],
3751                                     struct btrfs_file_extent_item);
3752                 if ((btrfs_file_extent_type(leaf, fi) !=
3753                      BTRFS_FILE_EXTENT_REG) ||
3754                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3755                      extent_key->objectid)) {
3756                         path->slots[0]++;
3757                         ret = 1;
3758                         goto next;
3759                 }
3760
3761                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3762                 ext_offset = btrfs_file_extent_offset(leaf, fi);
3763
3764                 if (first_pos > key.offset - ext_offset)
3765                         first_pos = key.offset - ext_offset;
3766
3767                 if (!extent_locked) {
3768                         lock_start = key.offset;
3769                         lock_end = lock_start + num_bytes - 1;
3770                 } else {
3771                         BUG_ON(lock_start != key.offset);
3772                         BUG_ON(lock_end - lock_start + 1 < num_bytes);
3773                 }
3774
3775                 if (!inode) {
3776                         btrfs_release_path(root, path);
3777
3778                         inode = btrfs_iget_locked(root->fs_info->sb,
3779                                                   key.objectid, root);
3780                         if (inode->i_state & I_NEW) {
3781                                 BTRFS_I(inode)->root = root;
3782                                 BTRFS_I(inode)->location.objectid =
3783                                         key.objectid;
3784                                 BTRFS_I(inode)->location.type =
3785                                         BTRFS_INODE_ITEM_KEY;
3786                                 BTRFS_I(inode)->location.offset = 0;
3787                                 btrfs_read_locked_inode(inode);
3788                                 unlock_new_inode(inode);
3789                         }
3790                         /*
3791                          * some code call btrfs_commit_transaction while
3792                          * holding the i_mutex, so we can't use mutex_lock
3793                          * here.
3794                          */
3795                         if (is_bad_inode(inode) ||
3796                             !mutex_trylock(&inode->i_mutex)) {
3797                                 iput(inode);
3798                                 inode = NULL;
3799                                 key.offset = (u64)-1;
3800                                 goto skip;
3801                         }
3802                 }
3803
3804                 if (!extent_locked) {
3805                         struct btrfs_ordered_extent *ordered;
3806
3807                         btrfs_release_path(root, path);
3808
3809                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3810                                     lock_end, GFP_NOFS);
3811                         ordered = btrfs_lookup_first_ordered_extent(inode,
3812                                                                     lock_end);
3813                         if (ordered &&
3814                             ordered->file_offset <= lock_end &&
3815                             ordered->file_offset + ordered->len > lock_start) {
3816                                 unlock_extent(&BTRFS_I(inode)->io_tree,
3817                                               lock_start, lock_end, GFP_NOFS);
3818                                 btrfs_start_ordered_extent(inode, ordered, 1);
3819                                 btrfs_put_ordered_extent(ordered);
3820                                 key.offset += num_bytes;
3821                                 goto skip;
3822                         }
3823                         if (ordered)
3824                                 btrfs_put_ordered_extent(ordered);
3825
3826                         mutex_lock(&BTRFS_I(inode)->extent_mutex);
3827                         extent_locked = 1;
3828                         continue;
3829                 }
3830
3831                 if (nr_extents == 1) {
3832                         /* update extent pointer in place */
3833                         btrfs_set_file_extent_generation(leaf, fi,
3834                                                 trans->transid);
3835                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
3836                                                 new_extents[0].disk_bytenr);
3837                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3838                                                 new_extents[0].disk_num_bytes);
3839                         ext_offset += new_extents[0].offset;
3840                         btrfs_set_file_extent_offset(leaf, fi, ext_offset);
3841                         btrfs_mark_buffer_dirty(leaf);
3842
3843                         btrfs_drop_extent_cache(inode, key.offset,
3844                                                 key.offset + num_bytes - 1, 0);
3845
3846                         ret = btrfs_inc_extent_ref(trans, root,
3847                                                 new_extents[0].disk_bytenr,
3848                                                 new_extents[0].disk_num_bytes,
3849                                                 leaf->start,
3850                                                 root->root_key.objectid,
3851                                                 trans->transid,
3852                                                 key.objectid, key.offset);
3853                         BUG_ON(ret);
3854
3855                         ret = btrfs_free_extent(trans, root,
3856                                                 extent_key->objectid,
3857                                                 extent_key->offset,
3858                                                 leaf->start,
3859                                                 btrfs_header_owner(leaf),
3860                                                 btrfs_header_generation(leaf),
3861                                                 key.objectid, key.offset, 0);
3862                         BUG_ON(ret);
3863
3864                         btrfs_release_path(root, path);
3865                         key.offset += num_bytes;
3866                 } else {
3867                         u64 alloc_hint;
3868                         u64 extent_len;
3869                         int i;
3870                         /*
3871                          * drop old extent pointer at first, then insert the
3872                          * new pointers one bye one
3873                          */
3874                         btrfs_release_path(root, path);
3875                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
3876                                                  key.offset + num_bytes,
3877                                                  key.offset, &alloc_hint);
3878                         BUG_ON(ret);
3879
3880                         for (i = 0; i < nr_extents; i++) {
3881                                 if (ext_offset >= new_extents[i].num_bytes) {
3882                                         ext_offset -= new_extents[i].num_bytes;
3883                                         continue;
3884                                 }
3885                                 extent_len = min(new_extents[i].num_bytes -
3886                                                  ext_offset, num_bytes);
3887
3888                                 ret = btrfs_insert_empty_item(trans, root,
3889                                                               path, &key,
3890                                                               sizeof(*fi));
3891                                 BUG_ON(ret);
3892
3893                                 leaf = path->nodes[0];
3894                                 fi = btrfs_item_ptr(leaf, path->slots[0],
3895                                                 struct btrfs_file_extent_item);
3896                                 btrfs_set_file_extent_generation(leaf, fi,
3897                                                         trans->transid);
3898                                 btrfs_set_file_extent_type(leaf, fi,
3899                                                         BTRFS_FILE_EXTENT_REG);
3900                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3901                                                 new_extents[i].disk_bytenr);
3902                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3903                                                 new_extents[i].disk_num_bytes);
3904                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3905                                                         extent_len);
3906                                 ext_offset += new_extents[i].offset;
3907                                 btrfs_set_file_extent_offset(leaf, fi,
3908                                                         ext_offset);
3909                                 btrfs_mark_buffer_dirty(leaf);
3910
3911                                 btrfs_drop_extent_cache(inode, key.offset,
3912                                                 key.offset + extent_len - 1, 0);
3913
3914                                 ret = btrfs_inc_extent_ref(trans, root,
3915                                                 new_extents[i].disk_bytenr,
3916                                                 new_extents[i].disk_num_bytes,
3917                                                 leaf->start,
3918                                                 root->root_key.objectid,
3919                                                 trans->transid,
3920                                                 key.objectid, key.offset);
3921                                 BUG_ON(ret);
3922                                 btrfs_release_path(root, path);
3923
3924                                 inode->i_blocks += extent_len >> 9;
3925
3926                                 ext_offset = 0;
3927                                 num_bytes -= extent_len;
3928                                 key.offset += extent_len;
3929
3930                                 if (num_bytes == 0)
3931                                         break;
3932                         }
3933                         BUG_ON(i >= nr_extents);
3934                 }
3935
3936                 if (extent_locked) {
3937                         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3938                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3939                                       lock_end, GFP_NOFS);
3940                         extent_locked = 0;
3941                 }
3942 skip:
3943                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
3944                     key.offset >= first_pos + extent_key->offset)
3945                         break;
3946
3947                 cond_resched();
3948         }
3949         ret = 0;
3950 out:
3951         btrfs_release_path(root, path);
3952         if (inode) {
3953                 mutex_unlock(&inode->i_mutex);
3954                 if (extent_locked) {
3955                         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3956                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3957                                       lock_end, GFP_NOFS);
3958                 }
3959                 iput(inode);
3960         }
3961         return ret;
3962 }
3963
3964 int btrfs_add_reloc_mapping(struct btrfs_root *root, u64 orig_bytenr,
3965                             u64 num_bytes, u64 new_bytenr)
3966 {
3967         set_extent_bits(&root->fs_info->reloc_mapping_tree,
3968                         orig_bytenr, orig_bytenr + num_bytes - 1,
3969                         EXTENT_LOCKED, GFP_NOFS);
3970         set_state_private(&root->fs_info->reloc_mapping_tree,
3971                           orig_bytenr, new_bytenr);
3972         return 0;
3973 }
3974
3975 int btrfs_get_reloc_mapping(struct btrfs_root *root, u64 orig_bytenr,
3976                             u64 num_bytes, u64 *new_bytenr)
3977 {
3978         u64 bytenr;
3979         u64 cur_bytenr = orig_bytenr;
3980         u64 prev_bytenr = orig_bytenr;
3981         int ret;
3982
3983         while (1) {
3984                 ret = get_state_private(&root->fs_info->reloc_mapping_tree,
3985                                         cur_bytenr, &bytenr);
3986                 if (ret)
3987                         break;
3988                 prev_bytenr = cur_bytenr;
3989                 cur_bytenr = bytenr;
3990         }
3991
3992         if (orig_bytenr == cur_bytenr)
3993                 return -ENOENT;
3994
3995         if (prev_bytenr != orig_bytenr) {
3996                 set_state_private(&root->fs_info->reloc_mapping_tree,
3997                                   orig_bytenr, cur_bytenr);
3998         }
3999         *new_bytenr = cur_bytenr;
4000         return 0;
4001 }
4002
4003 void btrfs_free_reloc_mappings(struct btrfs_root *root)
4004 {
4005         clear_extent_bits(&root->fs_info->reloc_mapping_tree,
4006                           0, (u64)-1, -1, GFP_NOFS);
4007 }
4008
4009 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4010                                struct btrfs_root *root,
4011                                struct extent_buffer *buf, u64 orig_start)
4012 {
4013         int level;
4014         int ret;
4015
4016         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4017         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4018
4019         level = btrfs_header_level(buf);
4020         if (level == 0) {
4021                 struct btrfs_leaf_ref *ref;
4022                 struct btrfs_leaf_ref *orig_ref;
4023
4024                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4025                 if (!orig_ref)
4026                         return -ENOENT;
4027
4028                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4029                 if (!ref) {
4030                         btrfs_free_leaf_ref(root, orig_ref);
4031                         return -ENOMEM;
4032                 }
4033
4034                 ref->nritems = orig_ref->nritems;
4035                 memcpy(ref->extents, orig_ref->extents,
4036                         sizeof(ref->extents[0]) * ref->nritems);
4037
4038                 btrfs_free_leaf_ref(root, orig_ref);
4039
4040                 ref->root_gen = trans->transid;
4041                 ref->bytenr = buf->start;
4042                 ref->owner = btrfs_header_owner(buf);
4043                 ref->generation = btrfs_header_generation(buf);
4044                 ret = btrfs_add_leaf_ref(root, ref, 0);
4045                 WARN_ON(ret);
4046                 btrfs_free_leaf_ref(root, ref);
4047         }
4048         return 0;
4049 }
4050
4051 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4052                                         struct extent_buffer *leaf,
4053                                         struct btrfs_block_group_cache *group,
4054                                         struct btrfs_root *target_root)
4055 {
4056         struct btrfs_key key;
4057         struct inode *inode = NULL;
4058         struct btrfs_file_extent_item *fi;
4059         u64 num_bytes;
4060         u64 skip_objectid = 0;
4061         u32 nritems;
4062         u32 i;
4063
4064         nritems = btrfs_header_nritems(leaf);
4065         for (i = 0; i < nritems; i++) {
4066                 btrfs_item_key_to_cpu(leaf, &key, i);
4067                 if (key.objectid == skip_objectid ||
4068                     key.type != BTRFS_EXTENT_DATA_KEY)
4069                         continue;
4070                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4071                 if (btrfs_file_extent_type(leaf, fi) ==
4072                     BTRFS_FILE_EXTENT_INLINE)
4073                         continue;
4074                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4075                         continue;
4076                 if (!inode || inode->i_ino != key.objectid) {
4077                         iput(inode);
4078                         inode = btrfs_ilookup(target_root->fs_info->sb,
4079                                               key.objectid, target_root, 1);
4080                 }
4081                 if (!inode) {
4082                         skip_objectid = key.objectid;
4083                         continue;
4084                 }
4085                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4086
4087                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4088                             key.offset + num_bytes - 1, GFP_NOFS);
4089                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
4090                 btrfs_drop_extent_cache(inode, key.offset,
4091                                         key.offset + num_bytes - 1, 1);
4092                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
4093                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4094                               key.offset + num_bytes - 1, GFP_NOFS);
4095                 cond_resched();
4096         }
4097         iput(inode);
4098         return 0;
4099 }
4100
4101 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4102                                         struct btrfs_root *root,
4103                                         struct extent_buffer *leaf,
4104                                         struct btrfs_block_group_cache *group,
4105                                         struct inode *reloc_inode)
4106 {
4107         struct btrfs_key key;
4108         struct btrfs_key extent_key;
4109         struct btrfs_file_extent_item *fi;
4110         struct btrfs_leaf_ref *ref;
4111         struct disk_extent *new_extent;
4112         u64 bytenr;
4113         u64 num_bytes;
4114         u32 nritems;
4115         u32 i;
4116         int ext_index;
4117         int nr_extent;
4118         int ret;
4119
4120         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4121         BUG_ON(!new_extent);
4122
4123         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4124         BUG_ON(!ref);
4125
4126         ext_index = -1;
4127         nritems = btrfs_header_nritems(leaf);
4128         for (i = 0; i < nritems; i++) {
4129                 btrfs_item_key_to_cpu(leaf, &key, i);
4130                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4131                         continue;
4132                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4133                 if (btrfs_file_extent_type(leaf, fi) ==
4134                     BTRFS_FILE_EXTENT_INLINE)
4135                         continue;
4136                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4137                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4138                 if (bytenr == 0)
4139                         continue;
4140
4141                 ext_index++;
4142                 if (bytenr >= group->key.objectid + group->key.offset ||
4143                     bytenr + num_bytes <= group->key.objectid)
4144                         continue;
4145
4146                 extent_key.objectid = bytenr;
4147                 extent_key.offset = num_bytes;
4148                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4149                 nr_extent = 1;
4150                 ret = get_new_locations(reloc_inode, &extent_key,
4151                                         group->key.objectid, 1,
4152                                         &new_extent, &nr_extent);
4153                 if (ret > 0)
4154                         continue;
4155                 BUG_ON(ret < 0);
4156
4157                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4158                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4159                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4160                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4161
4162                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
4163                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4164                                                 new_extent->disk_bytenr);
4165                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4166                                                 new_extent->disk_num_bytes);
4167                 new_extent->offset += btrfs_file_extent_offset(leaf, fi);
4168                 btrfs_set_file_extent_offset(leaf, fi, new_extent->offset);
4169                 btrfs_mark_buffer_dirty(leaf);
4170
4171                 ret = btrfs_inc_extent_ref(trans, root,
4172                                         new_extent->disk_bytenr,
4173                                         new_extent->disk_num_bytes,
4174                                         leaf->start,
4175                                         root->root_key.objectid,
4176                                         trans->transid,
4177                                         key.objectid, key.offset);
4178                 BUG_ON(ret);
4179                 ret = btrfs_free_extent(trans, root,
4180                                         bytenr, num_bytes, leaf->start,
4181                                         btrfs_header_owner(leaf),
4182                                         btrfs_header_generation(leaf),
4183                                         key.objectid, key.offset, 0);
4184                 BUG_ON(ret);
4185                 cond_resched();
4186         }
4187         kfree(new_extent);
4188         BUG_ON(ext_index + 1 != ref->nritems);
4189         btrfs_free_leaf_ref(root, ref);
4190         return 0;
4191 }
4192
4193 int btrfs_free_reloc_root(struct btrfs_root *root)
4194 {
4195         struct btrfs_root *reloc_root;
4196
4197         if (root->reloc_root) {
4198                 reloc_root = root->reloc_root;
4199                 root->reloc_root = NULL;
4200                 list_add(&reloc_root->dead_list,
4201                          &root->fs_info->dead_reloc_roots);
4202         }
4203         return 0;
4204 }
4205
4206 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4207 {
4208         struct btrfs_trans_handle *trans;
4209         struct btrfs_root *reloc_root;
4210         struct btrfs_root *prev_root = NULL;
4211         struct list_head dead_roots;
4212         int ret;
4213         unsigned long nr;
4214
4215         INIT_LIST_HEAD(&dead_roots);
4216         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4217
4218         while (!list_empty(&dead_roots)) {
4219                 reloc_root = list_entry(dead_roots.prev,
4220                                         struct btrfs_root, dead_list);
4221                 list_del_init(&reloc_root->dead_list);
4222
4223                 BUG_ON(reloc_root->commit_root != NULL);
4224                 while (1) {
4225                         trans = btrfs_join_transaction(root, 1);
4226                         BUG_ON(!trans);
4227
4228                         mutex_lock(&root->fs_info->drop_mutex);
4229                         ret = btrfs_drop_snapshot(trans, reloc_root);
4230                         if (ret != -EAGAIN)
4231                                 break;
4232                         mutex_unlock(&root->fs_info->drop_mutex);
4233
4234                         nr = trans->blocks_used;
4235                         ret = btrfs_end_transaction(trans, root);
4236                         BUG_ON(ret);
4237                         btrfs_btree_balance_dirty(root, nr);
4238                 }
4239
4240                 free_extent_buffer(reloc_root->node);
4241
4242                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4243                                      &reloc_root->root_key);
4244                 BUG_ON(ret);
4245                 mutex_unlock(&root->fs_info->drop_mutex);
4246
4247                 nr = trans->blocks_used;
4248                 ret = btrfs_end_transaction(trans, root);
4249                 BUG_ON(ret);
4250                 btrfs_btree_balance_dirty(root, nr);
4251
4252                 kfree(prev_root);
4253                 prev_root = reloc_root;
4254         }
4255         if (prev_root) {
4256                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4257                 kfree(prev_root);
4258         }
4259         return 0;
4260 }
4261
4262 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4263 {
4264         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4265         return 0;
4266 }
4267
4268 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4269 {
4270         struct btrfs_root *reloc_root;
4271         struct btrfs_trans_handle *trans;
4272         struct btrfs_key location;
4273         int found;
4274         int ret;
4275
4276         mutex_lock(&root->fs_info->tree_reloc_mutex);
4277         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4278         BUG_ON(ret);
4279         found = !list_empty(&root->fs_info->dead_reloc_roots);
4280         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4281
4282         if (found) {
4283                 trans = btrfs_start_transaction(root, 1);
4284                 BUG_ON(!trans);
4285                 ret = btrfs_commit_transaction(trans, root);
4286                 BUG_ON(ret);
4287         }
4288
4289         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4290         location.offset = (u64)-1;
4291         location.type = BTRFS_ROOT_ITEM_KEY;
4292
4293         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4294         BUG_ON(!reloc_root);
4295         btrfs_orphan_cleanup(reloc_root);
4296         return 0;
4297 }
4298
4299 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
4300                                     struct btrfs_root *root)
4301 {
4302         struct btrfs_root *reloc_root;
4303         struct extent_buffer *eb;
4304         struct btrfs_root_item *root_item;
4305         struct btrfs_key root_key;
4306         int ret;
4307
4308         BUG_ON(!root->ref_cows);
4309         if (root->reloc_root)
4310                 return 0;
4311
4312         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4313         BUG_ON(!root_item);
4314
4315         ret = btrfs_copy_root(trans, root, root->commit_root,
4316                               &eb, BTRFS_TREE_RELOC_OBJECTID);
4317         BUG_ON(ret);
4318
4319         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4320         root_key.offset = root->root_key.objectid;
4321         root_key.type = BTRFS_ROOT_ITEM_KEY;
4322
4323         memcpy(root_item, &root->root_item, sizeof(root_item));
4324         btrfs_set_root_refs(root_item, 0);
4325         btrfs_set_root_bytenr(root_item, eb->start);
4326         btrfs_set_root_level(root_item, btrfs_header_level(eb));
4327         memset(&root_item->drop_progress, 0, sizeof(root_item->drop_progress));
4328         root_item->drop_level = 0;
4329
4330         btrfs_tree_unlock(eb);
4331         free_extent_buffer(eb);
4332
4333         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4334                                 &root_key, root_item);
4335         BUG_ON(ret);
4336         kfree(root_item);
4337
4338         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4339                                                  &root_key);
4340         BUG_ON(!reloc_root);
4341         reloc_root->last_trans = trans->transid;
4342         reloc_root->commit_root = NULL;
4343         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4344
4345         root->reloc_root = reloc_root;
4346         return 0;
4347 }
4348
4349 /*
4350  * Core function of space balance.
4351  *
4352  * The idea is using reloc trees to relocate tree blocks in reference
4353  * counted roots. There is one reloc tree for each subvol, all reloc
4354  * trees share same key objectid. Reloc trees are snapshots of the
4355  * latest committed roots (subvol root->commit_root). To relocate a tree
4356  * block referenced by a subvol, the code COW the block through the reloc
4357  * tree, then update pointer in the subvol to point to the new block.
4358  * Since all reloc trees share same key objectid, we can easily do special
4359  * handing to share tree blocks between reloc trees. Once a tree block has
4360  * been COWed in one reloc tree, we can use the result when the same block
4361  * is COWed again through other reloc trees.
4362  */
4363 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
4364                                       struct btrfs_root *root,
4365                                       struct btrfs_path *path,
4366                                       struct btrfs_key *first_key,
4367                                       struct btrfs_ref_path *ref_path,
4368                                       struct btrfs_block_group_cache *group,
4369                                       struct inode *reloc_inode)
4370 {
4371         struct btrfs_root *reloc_root;
4372         struct extent_buffer *eb = NULL;
4373         struct btrfs_key *keys;
4374         u64 *nodes;
4375         int level;
4376         int lowest_merge;
4377         int lowest_level = 0;
4378         int update_refs;
4379         int ret;
4380
4381         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
4382                 lowest_level = ref_path->owner_objectid;
4383
4384         if (is_cowonly_root(ref_path->root_objectid)) {
4385                 path->lowest_level = lowest_level;
4386                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
4387                 BUG_ON(ret < 0);
4388                 path->lowest_level = 0;
4389                 btrfs_release_path(root, path);
4390                 return 0;
4391         }
4392
4393         keys = kzalloc(sizeof(*keys) * BTRFS_MAX_LEVEL, GFP_NOFS);
4394         BUG_ON(!keys);
4395         nodes = kzalloc(sizeof(*nodes) * BTRFS_MAX_LEVEL, GFP_NOFS);
4396         BUG_ON(!nodes);
4397
4398         mutex_lock(&root->fs_info->tree_reloc_mutex);
4399         ret = init_reloc_tree(trans, root);
4400         BUG_ON(ret);
4401         reloc_root = root->reloc_root;
4402
4403         path->lowest_level = lowest_level;
4404         ret = btrfs_search_slot(trans, reloc_root, first_key, path, 0, 0);
4405         BUG_ON(ret);
4406         /*
4407          * get relocation mapping for tree blocks in the path
4408          */
4409         lowest_merge = BTRFS_MAX_LEVEL;
4410         for (level = BTRFS_MAX_LEVEL - 1; level >= lowest_level; level--) {
4411                 u64 new_bytenr;
4412                 eb = path->nodes[level];
4413                 if (!eb || eb == reloc_root->node)
4414                         continue;
4415                 ret = btrfs_get_reloc_mapping(reloc_root, eb->start, eb->len,
4416                                               &new_bytenr);
4417                 if (ret)
4418                         continue;
4419                 if (level == 0)
4420                         btrfs_item_key_to_cpu(eb, &keys[level], 0);
4421                 else
4422                         btrfs_node_key_to_cpu(eb, &keys[level], 0);
4423                 nodes[level] = new_bytenr;
4424                 lowest_merge = level;
4425         }
4426
4427         update_refs = 0;
4428         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4429                 eb = path->nodes[0];
4430                 if (btrfs_header_generation(eb) < trans->transid)
4431                         update_refs = 1;
4432         }
4433
4434         btrfs_release_path(reloc_root, path);
4435         /*
4436          * merge tree blocks that already relocated in other reloc trees
4437          */
4438         if (lowest_merge != BTRFS_MAX_LEVEL) {
4439                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
4440                                        lowest_merge);
4441                 BUG_ON(ret < 0);
4442         }
4443         /*
4444          * cow any tree blocks that still haven't been relocated
4445          */
4446         ret = btrfs_search_slot(trans, reloc_root, first_key, path, 0, 1);
4447         BUG_ON(ret);
4448         /*
4449          * if we are relocating data block group, update extent pointers
4450          * in the newly created tree leaf.
4451          */
4452         eb = path->nodes[0];
4453         if (update_refs && nodes[0] != eb->start) {
4454                 ret = replace_extents_in_leaf(trans, reloc_root, eb, group,
4455                                               reloc_inode);
4456                 BUG_ON(ret);
4457         }
4458
4459         memset(keys, 0, sizeof(*keys) * BTRFS_MAX_LEVEL);
4460         memset(nodes, 0, sizeof(*nodes) * BTRFS_MAX_LEVEL);
4461         for (level = BTRFS_MAX_LEVEL - 1; level >= lowest_level; level--) {
4462                 eb = path->nodes[level];
4463                 if (!eb || eb == reloc_root->node)
4464                         continue;
4465                 BUG_ON(btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID);
4466                 nodes[level] = eb->start;
4467                 if (level == 0)
4468                         btrfs_item_key_to_cpu(eb, &keys[level], 0);
4469                 else
4470                         btrfs_node_key_to_cpu(eb, &keys[level], 0);
4471         }
4472
4473         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4474                 eb = path->nodes[0];
4475                 extent_buffer_get(eb);
4476         }
4477         btrfs_release_path(reloc_root, path);
4478         /*
4479          * replace tree blocks in the fs tree with tree blocks in
4480          * the reloc tree.
4481          */
4482         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
4483         BUG_ON(ret < 0);
4484
4485         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4486                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
4487                 BUG_ON(ret);
4488                 free_extent_buffer(eb);
4489         }
4490         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4491
4492         path->lowest_level = 0;
4493         kfree(nodes);
4494         kfree(keys);
4495         return 0;
4496 }
4497
4498 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
4499                                         struct btrfs_root *root,
4500                                         struct btrfs_path *path,
4501                                         struct btrfs_key *first_key,
4502                                         struct btrfs_ref_path *ref_path)
4503 {
4504         int ret;
4505         int needs_lock = 0;
4506
4507         if (root == root->fs_info->extent_root ||
4508             root == root->fs_info->chunk_root ||
4509             root == root->fs_info->dev_root) {
4510                 needs_lock = 1;
4511                 mutex_lock(&root->fs_info->alloc_mutex);
4512         }
4513
4514         ret = relocate_one_path(trans, root, path, first_key,
4515                                 ref_path, NULL, NULL);
4516         BUG_ON(ret);
4517
4518         if (root == root->fs_info->extent_root)
4519                 btrfs_extent_post_op(trans, root);
4520         if (needs_lock)
4521                 mutex_unlock(&root->fs_info->alloc_mutex);
4522
4523         return 0;
4524 }
4525
4526 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
4527                                     struct btrfs_root *extent_root,
4528                                     struct btrfs_path *path,
4529                                     struct btrfs_key *extent_key)
4530 {
4531         int ret;
4532
4533         mutex_lock(&extent_root->fs_info->alloc_mutex);
4534         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
4535         if (ret)
4536                 goto out;
4537         ret = btrfs_del_item(trans, extent_root, path);
4538 out:
4539         btrfs_release_path(extent_root, path);
4540         mutex_unlock(&extent_root->fs_info->alloc_mutex);
4541         return ret;
4542 }
4543
4544 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
4545                                                 struct btrfs_ref_path *ref_path)
4546 {
4547         struct btrfs_key root_key;
4548
4549         root_key.objectid = ref_path->root_objectid;
4550         root_key.type = BTRFS_ROOT_ITEM_KEY;
4551         if (is_cowonly_root(ref_path->root_objectid))
4552                 root_key.offset = 0;
4553         else
4554                 root_key.offset = (u64)-1;
4555
4556         return btrfs_read_fs_root_no_name(fs_info, &root_key);
4557 }
4558
4559 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
4560                                         struct btrfs_path *path,
4561                                         struct btrfs_key *extent_key,
4562                                         struct btrfs_block_group_cache *group,
4563                                         struct inode *reloc_inode, int pass)
4564 {
4565         struct btrfs_trans_handle *trans;
4566         struct btrfs_root *found_root;
4567         struct btrfs_ref_path *ref_path = NULL;
4568         struct disk_extent *new_extents = NULL;
4569         int nr_extents = 0;
4570         int loops;
4571         int ret;
4572         int level;
4573         struct btrfs_key first_key;
4574         u64 prev_block = 0;
4575
4576         mutex_unlock(&extent_root->fs_info->alloc_mutex);
4577
4578         trans = btrfs_start_transaction(extent_root, 1);
4579         BUG_ON(!trans);
4580
4581         if (extent_key->objectid == 0) {
4582                 ret = del_extent_zero(trans, extent_root, path, extent_key);
4583                 goto out;
4584         }
4585
4586         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
4587         if (!ref_path) {
4588                ret = -ENOMEM;
4589                goto out;
4590         }
4591
4592         for (loops = 0; ; loops++) {
4593                 if (loops == 0) {
4594                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
4595                                                    extent_key->objectid);
4596                 } else {
4597                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
4598                 }
4599                 if (ret < 0)
4600                         goto out;
4601                 if (ret > 0)
4602                         break;
4603
4604                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4605                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
4606                         continue;
4607
4608                 found_root = read_ref_root(extent_root->fs_info, ref_path);
4609                 BUG_ON(!found_root);
4610                 /*
4611                  * for reference counted tree, only process reference paths
4612                  * rooted at the latest committed root.
4613                  */
4614                 if (found_root->ref_cows &&
4615                     ref_path->root_generation != found_root->root_key.offset)
4616                         continue;
4617
4618                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4619                         if (pass == 0) {
4620                                 /*
4621                                  * copy data extents to new locations
4622                                  */
4623                                 u64 group_start = group->key.objectid;
4624                                 ret = relocate_data_extent(reloc_inode,
4625                                                            extent_key,
4626                                                            group_start);
4627                                 if (ret < 0)
4628                                         goto out;
4629                                 break;
4630                         }
4631                         level = 0;
4632                 } else {
4633                         level = ref_path->owner_objectid;
4634                 }
4635
4636                 if (prev_block != ref_path->nodes[level]) {
4637                         struct extent_buffer *eb;
4638                         u64 block_start = ref_path->nodes[level];
4639                         u64 block_size = btrfs_level_size(found_root, level);
4640
4641                         eb = read_tree_block(found_root, block_start,
4642                                              block_size, 0);
4643                         btrfs_tree_lock(eb);
4644                         BUG_ON(level != btrfs_header_level(eb));
4645
4646                         if (level == 0)
4647                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
4648                         else
4649                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
4650
4651                         btrfs_tree_unlock(eb);
4652                         free_extent_buffer(eb);
4653                         prev_block = block_start;
4654                 }
4655
4656                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4657                     pass >= 2) {
4658                         /*
4659                          * use fallback method to process the remaining
4660                          * references.
4661                          */
4662                         if (!new_extents) {
4663                                 u64 group_start = group->key.objectid;
4664                                 ret = get_new_locations(reloc_inode,
4665                                                         extent_key,
4666                                                         group_start, 0,
4667                                                         &new_extents,
4668                                                         &nr_extents);
4669                                 if (ret < 0)
4670                                         goto out;
4671                         }
4672                         btrfs_record_root_in_trans(found_root);
4673                         ret = replace_one_extent(trans, found_root,
4674                                                 path, extent_key,
4675                                                 &first_key, ref_path,
4676                                                 new_extents, nr_extents);
4677                         if (ret < 0)
4678                                 goto out;
4679                         continue;
4680                 }
4681
4682                 btrfs_record_root_in_trans(found_root);
4683                 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4684                         ret = relocate_tree_block(trans, found_root, path,
4685                                                   &first_key, ref_path);
4686                 } else {
4687                         /*
4688                          * try to update data extent references while
4689                          * keeping metadata shared between snapshots.
4690                          */
4691                         ret = relocate_one_path(trans, found_root, path,
4692                                                 &first_key, ref_path,
4693                                                 group, reloc_inode);
4694                 }
4695                 if (ret < 0)
4696                         goto out;
4697         }
4698         ret = 0;
4699 out:
4700         btrfs_end_transaction(trans, extent_root);
4701         kfree(new_extents);
4702         kfree(ref_path);
4703         mutex_lock(&extent_root->fs_info->alloc_mutex);
4704         return ret;
4705 }
4706
4707 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
4708 {
4709         u64 num_devices;
4710         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
4711                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
4712
4713         num_devices = root->fs_info->fs_devices->num_devices;
4714         if (num_devices == 1) {
4715                 stripped |= BTRFS_BLOCK_GROUP_DUP;
4716                 stripped = flags & ~stripped;
4717
4718                 /* turn raid0 into single device chunks */
4719                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
4720                         return stripped;
4721
4722                 /* turn mirroring into duplication */
4723                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
4724                              BTRFS_BLOCK_GROUP_RAID10))
4725                         return stripped | BTRFS_BLOCK_GROUP_DUP;
4726                 return flags;
4727         } else {
4728                 /* they already had raid on here, just return */
4729                 if (flags & stripped)
4730                         return flags;
4731
4732                 stripped |= BTRFS_BLOCK_GROUP_DUP;
4733                 stripped = flags & ~stripped;
4734
4735                 /* switch duplicated blocks with raid1 */
4736                 if (flags & BTRFS_BLOCK_GROUP_DUP)
4737                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
4738
4739                 /* turn single device chunks into raid0 */
4740                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
4741         }
4742         return flags;
4743 }
4744
4745 int __alloc_chunk_for_shrink(struct btrfs_root *root,
4746                      struct btrfs_block_group_cache *shrink_block_group,
4747                      int force)
4748 {
4749         struct btrfs_trans_handle *trans;
4750         u64 new_alloc_flags;
4751         u64 calc;
4752
4753         spin_lock(&shrink_block_group->lock);
4754         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
4755                 spin_unlock(&shrink_block_group->lock);
4756                 mutex_unlock(&root->fs_info->alloc_mutex);
4757
4758                 trans = btrfs_start_transaction(root, 1);
4759                 mutex_lock(&root->fs_info->alloc_mutex);
4760                 spin_lock(&shrink_block_group->lock);
4761
4762                 new_alloc_flags = update_block_group_flags(root,
4763                                                    shrink_block_group->flags);
4764                 if (new_alloc_flags != shrink_block_group->flags) {
4765                         calc =
4766                              btrfs_block_group_used(&shrink_block_group->item);
4767                 } else {
4768                         calc = shrink_block_group->key.offset;
4769                 }
4770                 spin_unlock(&shrink_block_group->lock);
4771
4772                 do_chunk_alloc(trans, root->fs_info->extent_root,
4773                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
4774
4775                 mutex_unlock(&root->fs_info->alloc_mutex);
4776                 btrfs_end_transaction(trans, root);
4777                 mutex_lock(&root->fs_info->alloc_mutex);
4778         } else
4779                 spin_unlock(&shrink_block_group->lock);
4780         return 0;
4781 }
4782
4783 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4784                                  struct btrfs_root *root,
4785                                  u64 objectid, u64 size)
4786 {
4787         struct btrfs_path *path;
4788         struct btrfs_inode_item *item;
4789         struct extent_buffer *leaf;
4790         int ret;
4791
4792         path = btrfs_alloc_path();
4793         if (!path)
4794                 return -ENOMEM;
4795
4796         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4797         if (ret)
4798                 goto out;
4799
4800         leaf = path->nodes[0];
4801         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4802         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4803         btrfs_set_inode_generation(leaf, item, 1);
4804         btrfs_set_inode_size(leaf, item, size);
4805         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4806         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM);
4807         btrfs_mark_buffer_dirty(leaf);
4808         btrfs_release_path(root, path);
4809 out:
4810         btrfs_free_path(path);
4811         return ret;
4812 }
4813
4814 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
4815                                         struct btrfs_block_group_cache *group)
4816 {
4817         struct inode *inode = NULL;
4818         struct btrfs_trans_handle *trans;
4819         struct btrfs_root *root;
4820         struct btrfs_key root_key;
4821         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
4822         int err = 0;
4823
4824         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4825         root_key.type = BTRFS_ROOT_ITEM_KEY;
4826         root_key.offset = (u64)-1;
4827         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
4828         if (IS_ERR(root))
4829                 return ERR_CAST(root);
4830
4831         trans = btrfs_start_transaction(root, 1);
4832         BUG_ON(!trans);
4833
4834         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
4835         if (err)
4836                 goto out;
4837
4838         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
4839         BUG_ON(err);
4840
4841         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
4842                                        group->key.offset, 0);
4843         BUG_ON(err);
4844
4845         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
4846         if (inode->i_state & I_NEW) {
4847                 BTRFS_I(inode)->root = root;
4848                 BTRFS_I(inode)->location.objectid = objectid;
4849                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
4850                 BTRFS_I(inode)->location.offset = 0;
4851                 btrfs_read_locked_inode(inode);
4852                 unlock_new_inode(inode);
4853                 BUG_ON(is_bad_inode(inode));
4854         } else {
4855                 BUG_ON(1);
4856         }
4857
4858         err = btrfs_orphan_add(trans, inode);
4859 out:
4860         btrfs_end_transaction(trans, root);
4861         if (err) {
4862                 if (inode)
4863                         iput(inode);
4864                 inode = ERR_PTR(err);
4865         }
4866         return inode;
4867 }
4868
4869 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
4870 {
4871         struct btrfs_trans_handle *trans;
4872         struct btrfs_path *path;
4873         struct btrfs_fs_info *info = root->fs_info;
4874         struct extent_buffer *leaf;
4875         struct inode *reloc_inode;
4876         struct btrfs_block_group_cache *block_group;
4877         struct btrfs_key key;
4878         u64 cur_byte;
4879         u64 total_found;
4880         u32 nritems;
4881         int ret;
4882         int progress;
4883         int pass = 0;
4884
4885         root = root->fs_info->extent_root;
4886
4887         block_group = btrfs_lookup_block_group(info, group_start);
4888         BUG_ON(!block_group);
4889
4890         printk("btrfs relocating block group %llu flags %llu\n",
4891                (unsigned long long)block_group->key.objectid,
4892                (unsigned long long)block_group->flags);
4893
4894         path = btrfs_alloc_path();
4895         BUG_ON(!path);
4896
4897         reloc_inode = create_reloc_inode(info, block_group);
4898         BUG_ON(IS_ERR(reloc_inode));
4899
4900         mutex_lock(&root->fs_info->alloc_mutex);
4901
4902         __alloc_chunk_for_shrink(root, block_group, 1);
4903         block_group->ro = 1;
4904         block_group->space_info->total_bytes -= block_group->key.offset;
4905
4906         mutex_unlock(&root->fs_info->alloc_mutex);
4907
4908         btrfs_start_delalloc_inodes(info->tree_root);
4909         btrfs_wait_ordered_extents(info->tree_root, 0);
4910 again:
4911         total_found = 0;
4912         progress = 0;
4913         key.objectid = block_group->key.objectid;
4914         key.offset = 0;
4915         key.type = 0;
4916         cur_byte = key.objectid;
4917
4918         trans = btrfs_start_transaction(info->tree_root, 1);
4919         btrfs_commit_transaction(trans, info->tree_root);
4920
4921         mutex_lock(&root->fs_info->cleaner_mutex);
4922         btrfs_clean_old_snapshots(info->tree_root);
4923         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
4924         mutex_unlock(&root->fs_info->cleaner_mutex);
4925
4926         mutex_lock(&root->fs_info->alloc_mutex);
4927
4928         while(1) {
4929                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4930                 if (ret < 0)
4931                         goto out;
4932 next:
4933                 leaf = path->nodes[0];
4934                 nritems = btrfs_header_nritems(leaf);
4935                 if (path->slots[0] >= nritems) {
4936                         ret = btrfs_next_leaf(root, path);
4937                         if (ret < 0)
4938                                 goto out;
4939                         if (ret == 1) {
4940                                 ret = 0;
4941                                 break;
4942                         }
4943                         leaf = path->nodes[0];
4944                         nritems = btrfs_header_nritems(leaf);
4945                 }
4946
4947                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4948
4949                 if (key.objectid >= block_group->key.objectid +
4950                     block_group->key.offset)
4951                         break;
4952
4953                 if (progress && need_resched()) {
4954                         btrfs_release_path(root, path);
4955                         mutex_unlock(&root->fs_info->alloc_mutex);
4956                         cond_resched();
4957                         mutex_lock(&root->fs_info->alloc_mutex);
4958                         progress = 0;
4959                         continue;
4960                 }
4961                 progress = 1;
4962
4963                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
4964                     key.objectid + key.offset <= cur_byte) {
4965                         path->slots[0]++;
4966                         goto next;
4967                 }
4968
4969                 total_found++;
4970                 cur_byte = key.objectid + key.offset;
4971                 btrfs_release_path(root, path);
4972
4973                 __alloc_chunk_for_shrink(root, block_group, 0);
4974                 ret = relocate_one_extent(root, path, &key, block_group,
4975                                           reloc_inode, pass);
4976                 BUG_ON(ret < 0);
4977
4978                 key.objectid = cur_byte;
4979                 key.type = 0;
4980                 key.offset = 0;
4981         }
4982
4983         btrfs_release_path(root, path);
4984         mutex_unlock(&root->fs_info->alloc_mutex);
4985
4986         if (pass == 0) {
4987                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
4988                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
4989                 WARN_ON(reloc_inode->i_mapping->nrpages);
4990         }
4991
4992         if (total_found > 0) {
4993                 printk("btrfs found %llu extents in pass %d\n",
4994                        (unsigned long long)total_found, pass);
4995                 pass++;
4996                 goto again;
4997         }
4998
4999         /* delete reloc_inode */
5000         iput(reloc_inode);
5001
5002         /* unpin extents in this range */
5003         trans = btrfs_start_transaction(info->tree_root, 1);
5004         btrfs_commit_transaction(trans, info->tree_root);
5005
5006         mutex_lock(&root->fs_info->alloc_mutex);
5007
5008         spin_lock(&block_group->lock);
5009         WARN_ON(block_group->pinned > 0);
5010         WARN_ON(block_group->reserved > 0);
5011         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5012         spin_unlock(&block_group->lock);
5013         ret = 0;
5014 out:
5015         mutex_unlock(&root->fs_info->alloc_mutex);
5016         btrfs_free_path(path);
5017         return ret;
5018 }
5019
5020 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5021                            struct btrfs_key *key)
5022 {
5023         int ret = 0;
5024         struct btrfs_key found_key;
5025         struct extent_buffer *leaf;
5026         int slot;
5027
5028         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5029         if (ret < 0)
5030                 goto out;
5031
5032         while(1) {
5033                 slot = path->slots[0];
5034                 leaf = path->nodes[0];
5035                 if (slot >= btrfs_header_nritems(leaf)) {
5036                         ret = btrfs_next_leaf(root, path);
5037                         if (ret == 0)
5038                                 continue;
5039                         if (ret < 0)
5040                                 goto out;
5041                         break;
5042                 }
5043                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5044
5045                 if (found_key.objectid >= key->objectid &&
5046                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5047                         ret = 0;
5048                         goto out;
5049                 }
5050                 path->slots[0]++;
5051         }
5052         ret = -ENOENT;
5053 out:
5054         return ret;
5055 }
5056
5057 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5058 {
5059         struct btrfs_block_group_cache *block_group;
5060         struct rb_node *n;
5061
5062         mutex_lock(&info->alloc_mutex);
5063         spin_lock(&info->block_group_cache_lock);
5064         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5065                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5066                                        cache_node);
5067
5068                 spin_unlock(&info->block_group_cache_lock);
5069                 btrfs_remove_free_space_cache(block_group);
5070                 spin_lock(&info->block_group_cache_lock);
5071
5072                 rb_erase(&block_group->cache_node,
5073                          &info->block_group_cache_tree);
5074                 spin_lock(&block_group->space_info->lock);
5075                 list_del(&block_group->list);
5076                 spin_unlock(&block_group->space_info->lock);
5077                 kfree(block_group);
5078         }
5079         spin_unlock(&info->block_group_cache_lock);
5080         mutex_unlock(&info->alloc_mutex);
5081         return 0;
5082 }
5083
5084 int btrfs_read_block_groups(struct btrfs_root *root)
5085 {
5086         struct btrfs_path *path;
5087         int ret;
5088         struct btrfs_block_group_cache *cache;
5089         struct btrfs_fs_info *info = root->fs_info;
5090         struct btrfs_space_info *space_info;
5091         struct btrfs_key key;
5092         struct btrfs_key found_key;
5093         struct extent_buffer *leaf;
5094
5095         root = info->extent_root;
5096         key.objectid = 0;
5097         key.offset = 0;
5098         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5099         path = btrfs_alloc_path();
5100         if (!path)
5101                 return -ENOMEM;
5102
5103         mutex_lock(&root->fs_info->alloc_mutex);
5104         while(1) {
5105                 ret = find_first_block_group(root, path, &key);
5106                 if (ret > 0) {
5107                         ret = 0;
5108                         goto error;
5109                 }
5110                 if (ret != 0)
5111                         goto error;
5112
5113                 leaf = path->nodes[0];
5114                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5115                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5116                 if (!cache) {
5117                         ret = -ENOMEM;
5118                         break;
5119                 }
5120
5121                 spin_lock_init(&cache->lock);
5122                 INIT_LIST_HEAD(&cache->list);
5123                 read_extent_buffer(leaf, &cache->item,
5124                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5125                                    sizeof(cache->item));
5126                 memcpy(&cache->key, &found_key, sizeof(found_key));
5127
5128                 key.objectid = found_key.objectid + found_key.offset;
5129                 btrfs_release_path(root, path);
5130                 cache->flags = btrfs_block_group_flags(&cache->item);
5131
5132                 ret = update_space_info(info, cache->flags, found_key.offset,
5133                                         btrfs_block_group_used(&cache->item),
5134                                         &space_info);
5135                 BUG_ON(ret);
5136                 cache->space_info = space_info;
5137                 spin_lock(&space_info->lock);
5138                 list_add(&cache->list, &space_info->block_groups);
5139                 spin_unlock(&space_info->lock);
5140
5141                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5142                 BUG_ON(ret);
5143         }
5144         ret = 0;
5145 error:
5146         btrfs_free_path(path);
5147         mutex_unlock(&root->fs_info->alloc_mutex);
5148         return ret;
5149 }
5150
5151 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5152                            struct btrfs_root *root, u64 bytes_used,
5153                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5154                            u64 size)
5155 {
5156         int ret;
5157         struct btrfs_root *extent_root;
5158         struct btrfs_block_group_cache *cache;
5159
5160         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
5161         extent_root = root->fs_info->extent_root;
5162
5163         root->fs_info->last_trans_new_blockgroup = trans->transid;
5164
5165         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5166         if (!cache)
5167                 return -ENOMEM;
5168
5169         cache->key.objectid = chunk_offset;
5170         cache->key.offset = size;
5171         spin_lock_init(&cache->lock);
5172         INIT_LIST_HEAD(&cache->list);
5173         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5174
5175         btrfs_set_block_group_used(&cache->item, bytes_used);
5176         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5177         cache->flags = type;
5178         btrfs_set_block_group_flags(&cache->item, type);
5179
5180         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5181                                 &cache->space_info);
5182         BUG_ON(ret);
5183         spin_lock(&cache->space_info->lock);
5184         list_add(&cache->list, &cache->space_info->block_groups);
5185         spin_unlock(&cache->space_info->lock);
5186
5187         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5188         BUG_ON(ret);
5189
5190         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5191                                 sizeof(cache->item));
5192         BUG_ON(ret);
5193
5194         finish_current_insert(trans, extent_root);
5195         ret = del_pending_extents(trans, extent_root);
5196         BUG_ON(ret);
5197         set_avail_alloc_bits(extent_root->fs_info, type);
5198
5199         return 0;
5200 }
5201
5202 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5203                              struct btrfs_root *root, u64 group_start)
5204 {
5205         struct btrfs_path *path;
5206         struct btrfs_block_group_cache *block_group;
5207         struct btrfs_key key;
5208         int ret;
5209
5210         BUG_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
5211         root = root->fs_info->extent_root;
5212
5213         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5214         BUG_ON(!block_group);
5215
5216         memcpy(&key, &block_group->key, sizeof(key));
5217
5218         path = btrfs_alloc_path();
5219         BUG_ON(!path);
5220
5221         btrfs_remove_free_space_cache(block_group);
5222         rb_erase(&block_group->cache_node,
5223                  &root->fs_info->block_group_cache_tree);
5224         spin_lock(&block_group->space_info->lock);
5225         list_del(&block_group->list);
5226         spin_unlock(&block_group->space_info->lock);
5227
5228         /*
5229         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5230         kfree(shrink_block_group);
5231         */
5232
5233         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5234         if (ret > 0)
5235                 ret = -EIO;
5236         if (ret < 0)
5237                 goto out;
5238
5239         ret = btrfs_del_item(trans, root, path);
5240 out:
5241         btrfs_free_path(path);
5242         return ret;
5243 }