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

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