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         if (key.objectid != bytenr) {
 772                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
 773                 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
 774                 BUG();
 775         }
 776         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
 777
 778         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 779         refs = btrfs_extent_refs(l, item);
 780         btrfs_set_extent_refs(l, item, refs + 1);
 781         btrfs_mark_buffer_dirty(path->nodes[0]);
 782
 783         btrfs_release_path(root->fs_info->extent_root, path);
 784
 785         path->reada = 1;
 786         ret = insert_extent_backref(trans, root->fs_info->extent_root,
 787                                     path, bytenr, parent,
 788                                     ref_root, ref_generation,
 789                                     owner_objectid);
 790         BUG_ON(ret);
 791         finish_current_insert(trans, root->fs_info->extent_root, 0);
 792         del_pending_extents(trans, root->fs_info->extent_root, 0);
 793
 794         btrfs_free_path(path);
 795         return 0;
 796 }
 797
 798 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 799                          struct btrfs_root *root,
 800                          u64 bytenr, u64 num_bytes, u64 parent,
 801                          u64 ref_root, u64 ref_generation,
 802                          u64 owner_objectid)
 803 {
 804         int ret;
 805         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
 806             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
 807                 return 0;
 808         ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
 809                                      0, ref_root, 0, ref_generation,
 810                                      owner_objectid);
 811         return ret;
 812 }
 813
 814 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 815                          struct btrfs_root *root)
 816 {
 817         finish_current_insert(trans, root->fs_info->extent_root, 1);
 818         del_pending_extents(trans, root->fs_info->extent_root, 1);
 819         return 0;
 820 }
 821
 822 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
 823                             struct btrfs_root *root, u64 bytenr,
 824                             u64 num_bytes, u32 *refs)
 825 {
 826         struct btrfs_path *path;
 827         int ret;
 828         struct btrfs_key key;
 829         struct extent_buffer *l;
 830         struct btrfs_extent_item *item;
 831
 832         WARN_ON(num_bytes < root->sectorsize);
 833         path = btrfs_alloc_path();
 834         path->reada = 1;
 835         key.objectid = bytenr;
 836         key.offset = num_bytes;
 837         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 838         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 839                                 0, 0);
 840         if (ret < 0)
 841                 goto out;
 842         if (ret != 0) {
 843                 btrfs_print_leaf(root, path->nodes[0]);
 844                 printk("failed to find block number %Lu\n", bytenr);
 845                 BUG();
 846         }
 847         l = path->nodes[0];
 848         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 849         *refs = btrfs_extent_refs(l, item);
 850 out:
 851         btrfs_free_path(path);
 852         return 0;
 853 }
 854
 855 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
 856                           struct btrfs_root *root, u64 bytenr)
 857 {
 858         struct btrfs_root *extent_root = root->fs_info->extent_root;
 859         struct btrfs_path *path;
 860         struct extent_buffer *leaf;
 861         struct btrfs_extent_ref *ref_item;
 862         struct btrfs_key key;
 863         struct btrfs_key found_key;
 864         u64 ref_root;
 865         u64 last_snapshot;
 866         u32 nritems;
 867         int ret;
 868
 869         key.objectid = bytenr;
 870         key.offset = (u64)-1;
 871         key.type = BTRFS_EXTENT_ITEM_KEY;
 872
 873         path = btrfs_alloc_path();
 874         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 875         if (ret < 0)
 876                 goto out;
 877         BUG_ON(ret == 0);
 878
 879         ret = -ENOENT;
 880         if (path->slots[0] == 0)
 881                 goto out;
 882
 883         path->slots[0]--;
 884         leaf = path->nodes[0];
 885         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 886
 887         if (found_key.objectid != bytenr ||
 888             found_key.type != BTRFS_EXTENT_ITEM_KEY)
 889                 goto out;
 890
 891         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
 892         while (1) {
 893                 leaf = path->nodes[0];
 894                 nritems = btrfs_header_nritems(leaf);
 895                 if (path->slots[0] >= nritems) {
 896                         ret = btrfs_next_leaf(extent_root, path);
 897                         if (ret < 0)
 898                                 goto out;
 899                         if (ret == 0)
 900                                 continue;
 901                         break;
 902                 }
 903                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 904                 if (found_key.objectid != bytenr)
 905                         break;
 906
 907                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 908                         path->slots[0]++;
 909                         continue;
 910                 }
 911
 912                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
 913                                           struct btrfs_extent_ref);
 914                 ref_root = btrfs_ref_root(leaf, ref_item);
 915                 if (ref_root != root->root_key.objectid &&
 916                     ref_root != BTRFS_TREE_LOG_OBJECTID) {
 917                         ret = 1;
 918                         goto out;
 919                 }
 920                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
 921                         ret = 1;
 922                         goto out;
 923                 }
 924
 925                 path->slots[0]++;
 926         }
 927         ret = 0;
 928 out:
 929         btrfs_free_path(path);
 930         return ret;
 931 }
 932
 933 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 934                     struct extent_buffer *buf, u32 nr_extents)
 935 {
 936         struct btrfs_key key;
 937         struct btrfs_file_extent_item *fi;
 938         u64 root_gen;
 939         u32 nritems;
 940         int i;
 941         int level;
 942         int ret = 0;
 943         int shared = 0;
 944
 945         if (!root->ref_cows)
 946                 return 0;
 947
 948         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
 949                 shared = 0;
 950                 root_gen = root->root_key.offset;
 951         } else {
 952                 shared = 1;
 953                 root_gen = trans->transid - 1;
 954         }
 955
 956         level = btrfs_header_level(buf);
 957         nritems = btrfs_header_nritems(buf);
 958
 959         if (level == 0) {
 960                 struct btrfs_leaf_ref *ref;
 961                 struct btrfs_extent_info *info;
 962
 963                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
 964                 if (!ref) {
 965                         ret = -ENOMEM;
 966                         goto out;
 967                 }
 968
 969                 ref->root_gen = root_gen;
 970                 ref->bytenr = buf->start;
 971                 ref->owner = btrfs_header_owner(buf);
 972                 ref->generation = btrfs_header_generation(buf);
 973                 ref->nritems = nr_extents;
 974                 info = ref->extents;
 975
 976                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
 977                         u64 disk_bytenr;
 978                         btrfs_item_key_to_cpu(buf, &key, i);
 979                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 980                                 continue;
 981                         fi = btrfs_item_ptr(buf, i,
 982                                             struct btrfs_file_extent_item);
 983                         if (btrfs_file_extent_type(buf, fi) ==
 984                             BTRFS_FILE_EXTENT_INLINE)
 985                                 continue;
 986                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 987                         if (disk_bytenr == 0)
 988                                 continue;
 989
 990                         info->bytenr = disk_bytenr;
 991                         info->num_bytes =
 992                                 btrfs_file_extent_disk_num_bytes(buf, fi);
 993                         info->objectid = key.objectid;
 994                         info->offset = key.offset;
 995                         info++;
 996                 }
 997
 998                 ret = btrfs_add_leaf_ref(root, ref, shared);
 999                 if (ret == -EEXIST && shared) {
1000                         struct btrfs_leaf_ref *old;
1001                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1002                         BUG_ON(!old);
1003                         btrfs_remove_leaf_ref(root, old);
1004                         btrfs_free_leaf_ref(root, old);
1005                         ret = btrfs_add_leaf_ref(root, ref, shared);
1006                 }
1007                 WARN_ON(ret);
1008                 btrfs_free_leaf_ref(root, ref);
1009         }
1010 out:
1011         return ret;
1012 }
1013
1014 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1015                   struct extent_buffer *orig_buf, struct extent_buffer *buf,
1016                   u32 *nr_extents)
1017 {
1018         u64 bytenr;
1019         u64 ref_root;
1020         u64 orig_root;
1021         u64 ref_generation;
1022         u64 orig_generation;
1023         u32 nritems;
1024         u32 nr_file_extents = 0;
1025         struct btrfs_key key;
1026         struct btrfs_file_extent_item *fi;
1027         int i;
1028         int level;
1029         int ret = 0;
1030         int faili = 0;
1031         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1032                             u64, u64, u64, u64, u64, u64, u64, u64);
1033
1034         ref_root = btrfs_header_owner(buf);
1035         ref_generation = btrfs_header_generation(buf);
1036         orig_root = btrfs_header_owner(orig_buf);
1037         orig_generation = btrfs_header_generation(orig_buf);
1038
1039         nritems = btrfs_header_nritems(buf);
1040         level = btrfs_header_level(buf);
1041
1042         if (root->ref_cows) {
1043                 process_func = __btrfs_inc_extent_ref;
1044         } else {
1045                 if (level == 0 &&
1046                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1047                         goto out;
1048                 if (level != 0 &&
1049                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1050                         goto out;
1051                 process_func = __btrfs_update_extent_ref;
1052         }
1053
1054         for (i = 0; i < nritems; i++) {
1055                 cond_resched();
1056                 if (level == 0) {
1057                         btrfs_item_key_to_cpu(buf, &key, i);
1058                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1059                                 continue;
1060                         fi = btrfs_item_ptr(buf, i,
1061                                             struct btrfs_file_extent_item);
1062                         if (btrfs_file_extent_type(buf, fi) ==
1063                             BTRFS_FILE_EXTENT_INLINE)
1064                                 continue;
1065                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1066                         if (bytenr == 0)
1067                                 continue;
1068
1069                         nr_file_extents++;
1070
1071                         ret = process_func(trans, root, bytenr,
1072                                            orig_buf->start, buf->start,
1073                                            orig_root, ref_root,
1074                                            orig_generation, ref_generation,
1075                                            key.objectid);
1076
1077                         if (ret) {
1078                                 faili = i;
1079                                 WARN_ON(1);
1080                                 goto fail;
1081                         }
1082                 } else {
1083                         bytenr = btrfs_node_blockptr(buf, i);
1084                         ret = process_func(trans, root, bytenr,
1085                                            orig_buf->start, buf->start,
1086                                            orig_root, ref_root,
1087                                            orig_generation, ref_generation,
1088                                            level - 1);
1089                         if (ret) {
1090                                 faili = i;
1091                                 WARN_ON(1);
1092                                 goto fail;
1093                         }
1094                 }
1095         }
1096 out:
1097         if (nr_extents) {
1098                 if (level == 0)
1099                         *nr_extents = nr_file_extents;
1100                 else
1101                         *nr_extents = nritems;
1102         }
1103         return 0;
1104 fail:
1105         WARN_ON(1);
1106         return ret;
1107 }
1108
1109 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1110                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1111                      struct extent_buffer *buf, int start_slot, int nr)
1112
1113 {
1114         u64 bytenr;
1115         u64 ref_root;
1116         u64 orig_root;
1117         u64 ref_generation;
1118         u64 orig_generation;
1119         struct btrfs_key key;
1120         struct btrfs_file_extent_item *fi;
1121         int i;
1122         int ret;
1123         int slot;
1124         int level;
1125
1126         BUG_ON(start_slot < 0);
1127         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1128
1129         ref_root = btrfs_header_owner(buf);
1130         ref_generation = btrfs_header_generation(buf);
1131         orig_root = btrfs_header_owner(orig_buf);
1132         orig_generation = btrfs_header_generation(orig_buf);
1133         level = btrfs_header_level(buf);
1134
1135         if (!root->ref_cows) {
1136                 if (level == 0 &&
1137                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1138                         return 0;
1139                 if (level != 0 &&
1140                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1141                         return 0;
1142         }
1143
1144         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1145                 cond_resched();
1146                 if (level == 0) {
1147                         btrfs_item_key_to_cpu(buf, &key, slot);
1148                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1149                                 continue;
1150                         fi = btrfs_item_ptr(buf, slot,
1151                                             struct btrfs_file_extent_item);
1152                         if (btrfs_file_extent_type(buf, fi) ==
1153                             BTRFS_FILE_EXTENT_INLINE)
1154                                 continue;
1155                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1156                         if (bytenr == 0)
1157                                 continue;
1158                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1159                                             orig_buf->start, buf->start,
1160                                             orig_root, ref_root,
1161                                             orig_generation, ref_generation,
1162                                             key.objectid);
1163                         if (ret)
1164                                 goto fail;
1165                 } else {
1166                         bytenr = btrfs_node_blockptr(buf, slot);
1167                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1168                                             orig_buf->start, buf->start,
1169                                             orig_root, ref_root,
1170                                             orig_generation, ref_generation,
1171                                             level - 1);
1172                         if (ret)
1173                                 goto fail;
1174                 }
1175         }
1176         return 0;
1177 fail:
1178         WARN_ON(1);
1179         return -1;
1180 }
1181
1182 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1183                                  struct btrfs_root *root,
1184                                  struct btrfs_path *path,
1185                                  struct btrfs_block_group_cache *cache)
1186 {
1187         int ret;
1188         int pending_ret;
1189         struct btrfs_root *extent_root = root->fs_info->extent_root;
1190         unsigned long bi;
1191         struct extent_buffer *leaf;
1192
1193         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1194         if (ret < 0)
1195                 goto fail;
1196         BUG_ON(ret);
1197
1198         leaf = path->nodes[0];
1199         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1200         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1201         btrfs_mark_buffer_dirty(leaf);
1202         btrfs_release_path(extent_root, path);
1203 fail:
1204         finish_current_insert(trans, extent_root, 0);
1205         pending_ret = del_pending_extents(trans, extent_root, 0);
1206         if (ret)
1207                 return ret;
1208         if (pending_ret)
1209                 return pending_ret;
1210         return 0;
1211
1212 }
1213
1214 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1215                                    struct btrfs_root *root)
1216 {
1217         struct btrfs_block_group_cache *cache, *entry;
1218         struct rb_node *n;
1219         int err = 0;
1220         int werr = 0;
1221         struct btrfs_path *path;
1222         u64 last = 0;
1223
1224         path = btrfs_alloc_path();
1225         if (!path)
1226                 return -ENOMEM;
1227
1228         while(1) {
1229                 cache = NULL;
1230                 spin_lock(&root->fs_info->block_group_cache_lock);
1231                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1232                      n; n = rb_next(n)) {
1233                         entry = rb_entry(n, struct btrfs_block_group_cache,
1234                                          cache_node);
1235                         if (entry->dirty) {
1236                                 cache = entry;
1237                                 break;
1238                         }
1239                 }
1240                 spin_unlock(&root->fs_info->block_group_cache_lock);
1241
1242                 if (!cache)
1243                         break;
1244
1245                 cache->dirty = 0;
1246                 last += cache->key.offset;
1247
1248                 err = write_one_cache_group(trans, root,
1249                                             path, cache);
1250                 /*
1251                  * if we fail to write the cache group, we want
1252                  * to keep it marked dirty in hopes that a later
1253                  * write will work
1254                  */
1255                 if (err) {
1256                         werr = err;
1257                         continue;
1258                 }
1259         }
1260         btrfs_free_path(path);
1261         return werr;
1262 }
1263
1264 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1265                              u64 total_bytes, u64 bytes_used,
1266                              struct btrfs_space_info **space_info)
1267 {
1268         struct btrfs_space_info *found;
1269
1270         found = __find_space_info(info, flags);
1271         if (found) {
1272                 spin_lock(&found->lock);
1273                 found->total_bytes += total_bytes;
1274                 found->bytes_used += bytes_used;
1275                 found->full = 0;
1276                 spin_unlock(&found->lock);
1277                 *space_info = found;
1278                 return 0;
1279         }
1280         found = kmalloc(sizeof(*found), GFP_NOFS);
1281         if (!found)
1282                 return -ENOMEM;
1283
1284         list_add(&found->list, &info->space_info);
1285         INIT_LIST_HEAD(&found->block_groups);
1286         init_rwsem(&found->groups_sem);
1287         spin_lock_init(&found->lock);
1288         found->flags = flags;
1289         found->total_bytes = total_bytes;
1290         found->bytes_used = bytes_used;
1291         found->bytes_pinned = 0;
1292         found->bytes_reserved = 0;
1293         found->full = 0;
1294         found->force_alloc = 0;
1295         *space_info = found;
1296         return 0;
1297 }
1298
1299 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1300 {
1301         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1302                                    BTRFS_BLOCK_GROUP_RAID1 |
1303                                    BTRFS_BLOCK_GROUP_RAID10 |
1304                                    BTRFS_BLOCK_GROUP_DUP);
1305         if (extra_flags) {
1306                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1307                         fs_info->avail_data_alloc_bits |= extra_flags;
1308                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1309                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1310                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1311                         fs_info->avail_system_alloc_bits |= extra_flags;
1312         }
1313 }
1314
1315 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1316 {
1317         u64 num_devices = root->fs_info->fs_devices->num_devices;
1318
1319         if (num_devices == 1)
1320                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1321         if (num_devices < 4)
1322                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1323
1324         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1325             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1326                       BTRFS_BLOCK_GROUP_RAID10))) {
1327                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1328         }
1329
1330         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1331             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1332                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1333         }
1334
1335         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1336             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1337              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1338              (flags & BTRFS_BLOCK_GROUP_DUP)))
1339                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1340         return flags;
1341 }
1342
1343 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1344                           struct btrfs_root *extent_root, u64 alloc_bytes,
1345                           u64 flags, int force)
1346 {
1347         struct btrfs_space_info *space_info;
1348         u64 thresh;
1349         u64 start;
1350         u64 num_bytes;
1351         int ret = 0, waited = 0;
1352
1353         flags = reduce_alloc_profile(extent_root, flags);
1354
1355         space_info = __find_space_info(extent_root->fs_info, flags);
1356         if (!space_info) {
1357                 ret = update_space_info(extent_root->fs_info, flags,
1358                                         0, 0, &space_info);
1359                 BUG_ON(ret);
1360         }
1361         BUG_ON(!space_info);
1362
1363         spin_lock(&space_info->lock);
1364         if (space_info->force_alloc) {
1365                 force = 1;
1366                 space_info->force_alloc = 0;
1367         }
1368         if (space_info->full) {
1369                 spin_unlock(&space_info->lock);
1370                 goto out;
1371         }
1372
1373         thresh = div_factor(space_info->total_bytes, 6);
1374         if (!force &&
1375            (space_info->bytes_used + space_info->bytes_pinned +
1376             space_info->bytes_reserved + alloc_bytes) < thresh) {
1377                 spin_unlock(&space_info->lock);
1378                 goto out;
1379         }
1380
1381         spin_unlock(&space_info->lock);
1382
1383         ret = mutex_trylock(&extent_root->fs_info->chunk_mutex);
1384         if (!ret && !force) {
1385                 goto out;
1386         } else if (!ret) {
1387                 mutex_lock(&extent_root->fs_info->chunk_mutex);
1388                 waited = 1;
1389         }
1390
1391         if (waited) {
1392                 spin_lock(&space_info->lock);
1393                 if (space_info->full) {
1394                         spin_unlock(&space_info->lock);
1395                         goto out_unlock;
1396                 }
1397                 spin_unlock(&space_info->lock);
1398         }
1399
1400         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1401         if (ret) {
1402 printk("space info full %Lu\n", flags);
1403                 space_info->full = 1;
1404                 goto out_unlock;
1405         }
1406
1407         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1408                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1409         BUG_ON(ret);
1410 out_unlock:
1411         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1412 out:
1413         return ret;
1414 }
1415
1416 static int update_block_group(struct btrfs_trans_handle *trans,
1417                               struct btrfs_root *root,
1418                               u64 bytenr, u64 num_bytes, int alloc,
1419                               int mark_free)
1420 {
1421         struct btrfs_block_group_cache *cache;
1422         struct btrfs_fs_info *info = root->fs_info;
1423         u64 total = num_bytes;
1424         u64 old_val;
1425         u64 byte_in_group;
1426
1427         while(total) {
1428                 cache = btrfs_lookup_block_group(info, bytenr);
1429                 if (!cache) {
1430                         return -1;
1431                 }
1432                 byte_in_group = bytenr - cache->key.objectid;
1433                 WARN_ON(byte_in_group > cache->key.offset);
1434
1435                 spin_lock(&cache->space_info->lock);
1436                 spin_lock(&cache->lock);
1437                 cache->dirty = 1;
1438                 old_val = btrfs_block_group_used(&cache->item);
1439                 num_bytes = min(total, cache->key.offset - byte_in_group);
1440                 if (alloc) {
1441                         old_val += num_bytes;
1442                         cache->space_info->bytes_used += num_bytes;
1443                         btrfs_set_block_group_used(&cache->item, old_val);
1444                         spin_unlock(&cache->lock);
1445                         spin_unlock(&cache->space_info->lock);
1446                 } else {
1447                         old_val -= num_bytes;
1448                         cache->space_info->bytes_used -= num_bytes;
1449                         btrfs_set_block_group_used(&cache->item, old_val);
1450                         spin_unlock(&cache->lock);
1451                         spin_unlock(&cache->space_info->lock);
1452                         if (mark_free) {
1453                                 int ret;
1454                                 ret = btrfs_add_free_space(cache, bytenr,
1455                                                            num_bytes);
1456                                 if (ret)
1457                                         return -1;
1458                         }
1459                 }
1460                 total -= num_bytes;
1461                 bytenr += num_bytes;
1462         }
1463         return 0;
1464 }
1465
1466 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1467 {
1468         struct btrfs_block_group_cache *cache;
1469
1470         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1471         if (!cache)
1472                 return 0;
1473
1474         return cache->key.objectid;
1475 }
1476
1477 int btrfs_update_pinned_extents(struct btrfs_root *root,
1478                                 u64 bytenr, u64 num, int pin)
1479 {
1480         u64 len;
1481         struct btrfs_block_group_cache *cache;
1482         struct btrfs_fs_info *fs_info = root->fs_info;
1483
1484         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
1485         if (pin) {
1486                 set_extent_dirty(&fs_info->pinned_extents,
1487                                 bytenr, bytenr + num - 1, GFP_NOFS);
1488         } else {
1489                 clear_extent_dirty(&fs_info->pinned_extents,
1490                                 bytenr, bytenr + num - 1, GFP_NOFS);
1491         }
1492         while (num > 0) {
1493                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1494                 BUG_ON(!cache);
1495                 len = min(num, cache->key.offset -
1496                           (bytenr - cache->key.objectid));
1497                 if (pin) {
1498                         spin_lock(&cache->space_info->lock);
1499                         spin_lock(&cache->lock);
1500                         cache->pinned += len;
1501                         cache->space_info->bytes_pinned += len;
1502                         spin_unlock(&cache->lock);
1503                         spin_unlock(&cache->space_info->lock);
1504                         fs_info->total_pinned += len;
1505                 } else {
1506                         spin_lock(&cache->space_info->lock);
1507                         spin_lock(&cache->lock);
1508                         cache->pinned -= len;
1509                         cache->space_info->bytes_pinned -= len;
1510                         spin_unlock(&cache->lock);
1511                         spin_unlock(&cache->space_info->lock);
1512                         fs_info->total_pinned -= len;
1513                 }
1514                 bytenr += len;
1515                 num -= len;
1516         }
1517         return 0;
1518 }
1519
1520 static int update_reserved_extents(struct btrfs_root *root,
1521                                    u64 bytenr, u64 num, int reserve)
1522 {
1523         u64 len;
1524         struct btrfs_block_group_cache *cache;
1525         struct btrfs_fs_info *fs_info = root->fs_info;
1526
1527         while (num > 0) {
1528                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1529                 BUG_ON(!cache);
1530                 len = min(num, cache->key.offset -
1531                           (bytenr - cache->key.objectid));
1532
1533                 spin_lock(&cache->space_info->lock);
1534                 spin_lock(&cache->lock);
1535                 if (reserve) {
1536                         cache->reserved += len;
1537                         cache->space_info->bytes_reserved += len;
1538                 } else {
1539                         cache->reserved -= len;
1540                         cache->space_info->bytes_reserved -= len;
1541                 }
1542                 spin_unlock(&cache->lock);
1543                 spin_unlock(&cache->space_info->lock);
1544                 bytenr += len;
1545                 num -= len;
1546         }
1547         return 0;
1548 }
1549
1550 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1551 {
1552         u64 last = 0;
1553         u64 start;
1554         u64 end;
1555         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1556         int ret;
1557
1558         mutex_lock(&root->fs_info->pinned_mutex);
1559         while(1) {
1560                 ret = find_first_extent_bit(pinned_extents, last,
1561                                             &start, &end, EXTENT_DIRTY);
1562                 if (ret)
1563                         break;
1564                 set_extent_dirty(copy, start, end, GFP_NOFS);
1565                 last = end + 1;
1566         }
1567         mutex_unlock(&root->fs_info->pinned_mutex);
1568         return 0;
1569 }
1570
1571 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1572                                struct btrfs_root *root,
1573                                struct extent_io_tree *unpin)
1574 {
1575         u64 start;
1576         u64 end;
1577         int ret;
1578         struct btrfs_block_group_cache *cache;
1579
1580         mutex_lock(&root->fs_info->pinned_mutex);
1581         while(1) {
1582                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1583                                             EXTENT_DIRTY);
1584                 if (ret)
1585                         break;
1586                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1587                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1588                 cache = btrfs_lookup_block_group(root->fs_info, start);
1589                 if (cache->cached)
1590                         btrfs_add_free_space(cache, start, end - start + 1);
1591                 if (need_resched()) {
1592                         mutex_unlock(&root->fs_info->pinned_mutex);
1593                         cond_resched();
1594                         mutex_lock(&root->fs_info->pinned_mutex);
1595                 }
1596         }
1597         mutex_unlock(&root->fs_info->pinned_mutex);
1598         return 0;
1599 }
1600
1601 static int finish_current_insert(struct btrfs_trans_handle *trans,
1602                                  struct btrfs_root *extent_root, int all)
1603 {
1604         u64 start;
1605         u64 end;
1606         u64 priv;
1607         u64 search = 0;
1608         struct btrfs_fs_info *info = extent_root->fs_info;
1609         struct btrfs_path *path;
1610         struct btrfs_extent_ref *ref;
1611         struct pending_extent_op *extent_op;
1612         struct btrfs_key key;
1613         struct btrfs_extent_item extent_item;
1614         int ret;
1615         int err = 0;
1616
1617         btrfs_set_stack_extent_refs(&extent_item, 1);
1618         path = btrfs_alloc_path();
1619
1620         while(1) {
1621                 mutex_lock(&info->extent_ins_mutex);
1622                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
1623                                             &end, EXTENT_WRITEBACK);
1624                 if (ret) {
1625                         mutex_unlock(&info->extent_ins_mutex);
1626                         if (search && all) {
1627                                 search = 0;
1628                                 continue;
1629                         }
1630                         break;
1631                 }
1632
1633                 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
1634                 if (!ret) {
1635                         search = end + 1;
1636                         mutex_unlock(&info->extent_ins_mutex);
1637                         cond_resched();
1638                         continue;
1639                 }
1640                 BUG_ON(ret < 0);
1641
1642                 ret = get_state_private(&info->extent_ins, start, &priv);
1643                 BUG_ON(ret);
1644                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1645
1646                 mutex_unlock(&info->extent_ins_mutex);
1647
1648                 if (extent_op->type == PENDING_EXTENT_INSERT) {
1649                         key.objectid = start;
1650                         key.offset = end + 1 - start;
1651                         key.type = BTRFS_EXTENT_ITEM_KEY;
1652                         err = btrfs_insert_item(trans, extent_root, &key,
1653                                         &extent_item, sizeof(extent_item));
1654                         BUG_ON(err);
1655
1656                         mutex_lock(&info->extent_ins_mutex);
1657                         clear_extent_bits(&info->extent_ins, start, end,
1658                                           EXTENT_WRITEBACK, GFP_NOFS);
1659                         mutex_unlock(&info->extent_ins_mutex);
1660
1661                         err = insert_extent_backref(trans, extent_root, path,
1662                                                 start, extent_op->parent,
1663                                                 extent_root->root_key.objectid,
1664                                                 extent_op->generation,
1665                                                 extent_op->level);
1666                         BUG_ON(err);
1667                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1668                         err = lookup_extent_backref(trans, extent_root, path,
1669                                                 start, extent_op->orig_parent,
1670                                                 extent_root->root_key.objectid,
1671                                                 extent_op->orig_generation,
1672                                                 extent_op->level, 0);
1673                         BUG_ON(err);
1674
1675                         mutex_lock(&info->extent_ins_mutex);
1676                         clear_extent_bits(&info->extent_ins, start, end,
1677                                           EXTENT_WRITEBACK, GFP_NOFS);
1678                         mutex_unlock(&info->extent_ins_mutex);
1679
1680                         key.objectid = start;
1681                         key.offset = extent_op->parent;
1682                         key.type = BTRFS_EXTENT_REF_KEY;
1683                         err = btrfs_set_item_key_safe(trans, extent_root, path,
1684                                                       &key);
1685                         BUG_ON(err);
1686                         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1687                                              struct btrfs_extent_ref);
1688                         btrfs_set_ref_generation(path->nodes[0], ref,
1689                                                  extent_op->generation);
1690                         btrfs_mark_buffer_dirty(path->nodes[0]);
1691                         btrfs_release_path(extent_root, path);
1692                 } else {
1693                         BUG_ON(1);
1694                 }
1695                 kfree(extent_op);
1696                 unlock_extent(&info->extent_ins, start, end, GFP_NOFS);
1697                 if (all)
1698                         search = 0;
1699                 else
1700                         search = end + 1;
1701
1702                 cond_resched();
1703         }
1704         btrfs_free_path(path);
1705         return 0;
1706 }
1707
1708 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1709                           struct btrfs_root *root,
1710                           u64 bytenr, u64 num_bytes, int is_data)
1711 {
1712         int err = 0;
1713         struct extent_buffer *buf;
1714
1715         if (is_data)
1716                 goto pinit;
1717
1718         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1719         if (!buf)
1720                 goto pinit;
1721
1722         /* we can reuse a block if it hasn't been written
1723          * and it is from this transaction.  We can't
1724          * reuse anything from the tree log root because
1725          * it has tiny sub-transactions.
1726          */
1727         if (btrfs_buffer_uptodate(buf, 0) &&
1728             btrfs_try_tree_lock(buf)) {
1729                 u64 header_owner = btrfs_header_owner(buf);
1730                 u64 header_transid = btrfs_header_generation(buf);
1731                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1732                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
1733                     header_transid == trans->transid &&
1734                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1735                         clean_tree_block(NULL, root, buf);
1736                         btrfs_tree_unlock(buf);
1737                         free_extent_buffer(buf);
1738                         return 1;
1739                 }
1740                 btrfs_tree_unlock(buf);
1741         }
1742         free_extent_buffer(buf);
1743 pinit:
1744         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1745
1746         BUG_ON(err < 0);
1747         return 0;
1748 }
1749
1750 /*
1751  * remove an extent from the root, returns 0 on success
1752  */
1753 static int __free_extent(struct btrfs_trans_handle *trans,
1754                          struct btrfs_root *root,
1755                          u64 bytenr, u64 num_bytes, u64 parent,
1756                          u64 root_objectid, u64 ref_generation,
1757                          u64 owner_objectid, int pin, int mark_free)
1758 {
1759         struct btrfs_path *path;
1760         struct btrfs_key key;
1761         struct btrfs_fs_info *info = root->fs_info;
1762         struct btrfs_root *extent_root = info->extent_root;
1763         struct extent_buffer *leaf;
1764         int ret;
1765         int extent_slot = 0;
1766         int found_extent = 0;
1767         int num_to_del = 1;
1768         struct btrfs_extent_item *ei;
1769         u32 refs;
1770
1771         key.objectid = bytenr;
1772         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1773         key.offset = num_bytes;
1774         path = btrfs_alloc_path();
1775         if (!path)
1776                 return -ENOMEM;
1777
1778         path->reada = 1;
1779         ret = lookup_extent_backref(trans, extent_root, path,
1780                                     bytenr, parent, root_objectid,
1781                                     ref_generation, owner_objectid, 1);
1782         if (ret == 0) {
1783                 struct btrfs_key found_key;
1784                 extent_slot = path->slots[0];
1785                 while(extent_slot > 0) {
1786                         extent_slot--;
1787                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1788                                               extent_slot);
1789                         if (found_key.objectid != bytenr)
1790                                 break;
1791                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1792                             found_key.offset == num_bytes) {
1793                                 found_extent = 1;
1794                                 break;
1795                         }
1796                         if (path->slots[0] - extent_slot > 5)
1797                                 break;
1798                 }
1799                 if (!found_extent) {
1800                         ret = remove_extent_backref(trans, extent_root, path);
1801                         BUG_ON(ret);
1802                         btrfs_release_path(extent_root, path);
1803                         ret = btrfs_search_slot(trans, extent_root,
1804                                                 &key, path, -1, 1);
1805                         BUG_ON(ret);
1806                         extent_slot = path->slots[0];
1807                 }
1808         } else {
1809                 btrfs_print_leaf(extent_root, path->nodes[0]);
1810                 WARN_ON(1);
1811                 printk("Unable to find ref byte nr %Lu root %Lu "
1812                        "gen %Lu owner %Lu\n", bytenr,
1813                        root_objectid, ref_generation, owner_objectid);
1814         }
1815
1816         leaf = path->nodes[0];
1817         ei = btrfs_item_ptr(leaf, extent_slot,
1818                             struct btrfs_extent_item);
1819         refs = btrfs_extent_refs(leaf, ei);
1820         BUG_ON(refs == 0);
1821         refs -= 1;
1822         btrfs_set_extent_refs(leaf, ei, refs);
1823
1824         btrfs_mark_buffer_dirty(leaf);
1825
1826         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1827                 struct btrfs_extent_ref *ref;
1828                 ref = btrfs_item_ptr(leaf, path->slots[0],
1829                                      struct btrfs_extent_ref);
1830                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
1831                 /* if the back ref and the extent are next to each other
1832                  * they get deleted below in one shot
1833                  */
1834                 path->slots[0] = extent_slot;
1835                 num_to_del = 2;
1836         } else if (found_extent) {
1837                 /* otherwise delete the extent back ref */
1838                 ret = remove_extent_backref(trans, extent_root, path);
1839                 BUG_ON(ret);
1840                 /* if refs are 0, we need to setup the path for deletion */
1841                 if (refs == 0) {
1842                         btrfs_release_path(extent_root, path);
1843                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1844                                                 -1, 1);
1845                         BUG_ON(ret);
1846                 }
1847         }
1848
1849         if (refs == 0) {
1850                 u64 super_used;
1851                 u64 root_used;
1852 #ifdef BIO_RW_DISCARD
1853                 u64 map_length = num_bytes;
1854                 struct btrfs_multi_bio *multi = NULL;
1855 #endif
1856
1857                 if (pin) {
1858                         mutex_lock(&root->fs_info->pinned_mutex);
1859                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
1860                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
1861                         mutex_unlock(&root->fs_info->pinned_mutex);
1862                         if (ret > 0)
1863                                 mark_free = 1;
1864                         BUG_ON(ret < 0);
1865                 }
1866
1867                 /* block accounting for super block */
1868                 spin_lock_irq(&info->delalloc_lock);
1869                 super_used = btrfs_super_bytes_used(&info->super_copy);
1870                 btrfs_set_super_bytes_used(&info->super_copy,
1871                                            super_used - num_bytes);
1872                 spin_unlock_irq(&info->delalloc_lock);
1873
1874                 /* block accounting for root item */
1875                 root_used = btrfs_root_used(&root->root_item);
1876                 btrfs_set_root_used(&root->root_item,
1877                                            root_used - num_bytes);
1878                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1879                                       num_to_del);
1880                 BUG_ON(ret);
1881                 btrfs_release_path(extent_root, path);
1882                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1883                                          mark_free);
1884                 BUG_ON(ret);
1885
1886 #ifdef BIO_RW_DISCARD
1887                 /* Tell the block device(s) that the sectors can be discarded */
1888                 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1889                                       bytenr, &map_length, &multi, 0);
1890                 if (!ret) {
1891                         struct btrfs_bio_stripe *stripe = multi->stripes;
1892                         int i;
1893
1894                         if (map_length > num_bytes)
1895                                 map_length = num_bytes;
1896
1897                         for (i = 0; i < multi->num_stripes; i++, stripe++) {
1898                                 blkdev_issue_discard(stripe->dev->bdev,
1899                                                      stripe->physical >> 9,
1900                                                      map_length >> 9);
1901                         }
1902                         kfree(multi);
1903                 }
1904 #endif
1905         }
1906         btrfs_free_path(path);
1907         finish_current_insert(trans, extent_root, 0);
1908         return ret;
1909 }
1910
1911 /*
1912  * find all the blocks marked as pending in the radix tree and remove
1913  * them from the extent map
1914  */
1915 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1916                                btrfs_root *extent_root, int all)
1917 {
1918         int ret;
1919         int err = 0;
1920         u64 start;
1921         u64 end;
1922         u64 priv;
1923         u64 search = 0;
1924         struct extent_io_tree *pending_del;
1925         struct extent_io_tree *extent_ins;
1926         struct pending_extent_op *extent_op;
1927         struct btrfs_fs_info *info = extent_root->fs_info;
1928
1929         extent_ins = &extent_root->fs_info->extent_ins;
1930         pending_del = &extent_root->fs_info->pending_del;
1931
1932         while(1) {
1933                 mutex_lock(&info->extent_ins_mutex);
1934                 ret = find_first_extent_bit(pending_del, search, &start, &end,
1935                                             EXTENT_WRITEBACK);
1936                 if (ret) {
1937                         mutex_unlock(&info->extent_ins_mutex);
1938                         if (all && search) {
1939                                 search = 0;
1940                                 continue;
1941                         }
1942                         break;
1943                 }
1944
1945                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
1946                 if (!ret) {
1947                         search = end+1;
1948                         mutex_unlock(&info->extent_ins_mutex);
1949                         cond_resched();
1950                         continue;
1951                 }
1952                 BUG_ON(ret < 0);
1953
1954                 ret = get_state_private(pending_del, start, &priv);
1955                 BUG_ON(ret);
1956                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1957
1958                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1959                                   GFP_NOFS);
1960                 if (!test_range_bit(extent_ins, start, end,
1961                                     EXTENT_WRITEBACK, 0)) {
1962                         mutex_unlock(&info->extent_ins_mutex);
1963 free_extent:
1964                         ret = __free_extent(trans, extent_root,
1965                                             start, end + 1 - start,
1966                                             extent_op->orig_parent,
1967                                             extent_root->root_key.objectid,
1968                                             extent_op->orig_generation,
1969                                             extent_op->level, 1, 0);
1970                         kfree(extent_op);
1971                 } else {
1972                         kfree(extent_op);
1973
1974                         ret = get_state_private(&info->extent_ins, start,
1975                                                 &priv);
1976                         BUG_ON(ret);
1977                         extent_op = (struct pending_extent_op *)
1978                                                 (unsigned long)priv;
1979
1980                         clear_extent_bits(&info->extent_ins, start, end,
1981                                           EXTENT_WRITEBACK, GFP_NOFS);
1982
1983                         mutex_unlock(&info->extent_ins_mutex);
1984
1985                         if (extent_op->type == PENDING_BACKREF_UPDATE)
1986                                 goto free_extent;
1987
1988                         mutex_lock(&extent_root->fs_info->pinned_mutex);
1989                         ret = pin_down_bytes(trans, extent_root, start,
1990                                              end + 1 - start, 0);
1991                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
1992
1993                         ret = update_block_group(trans, extent_root, start,
1994                                                 end + 1 - start, 0, ret > 0);
1995
1996                         BUG_ON(ret);
1997                         kfree(extent_op);
1998                 }
1999                 if (ret)
2000                         err = ret;
2001                 unlock_extent(extent_ins, start, end, GFP_NOFS);
2002
2003                 if (all)
2004                         search = 0;
2005                 else
2006                         search = end + 1;
2007                 cond_resched();
2008         }
2009         return err;
2010 }
2011
2012 /*
2013  * remove an extent from the root, returns 0 on success
2014  */
2015 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2016                                struct btrfs_root *root,
2017                                u64 bytenr, u64 num_bytes, u64 parent,
2018                                u64 root_objectid, u64 ref_generation,
2019                                u64 owner_objectid, int pin)
2020 {
2021         struct btrfs_root *extent_root = root->fs_info->extent_root;
2022         int pending_ret;
2023         int ret;
2024
2025         WARN_ON(num_bytes < root->sectorsize);
2026         if (root == extent_root) {
2027                 struct pending_extent_op *extent_op;
2028
2029                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2030                 BUG_ON(!extent_op);
2031
2032                 extent_op->type = PENDING_EXTENT_DELETE;
2033                 extent_op->bytenr = bytenr;
2034                 extent_op->num_bytes = num_bytes;
2035                 extent_op->parent = parent;
2036                 extent_op->orig_parent = parent;
2037                 extent_op->generation = ref_generation;
2038                 extent_op->orig_generation = ref_generation;
2039                 extent_op->level = (int)owner_objectid;
2040
2041                 mutex_lock(&root->fs_info->extent_ins_mutex);
2042                 set_extent_bits(&root->fs_info->pending_del,
2043                                 bytenr, bytenr + num_bytes - 1,
2044                                 EXTENT_WRITEBACK, GFP_NOFS);
2045                 set_state_private(&root->fs_info->pending_del,
2046                                   bytenr, (unsigned long)extent_op);
2047                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2048                 return 0;
2049         }
2050         /* if metadata always pin */
2051         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2052                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2053                         struct btrfs_block_group_cache *cache;
2054
2055                         /* btrfs_free_reserved_extent */
2056                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2057                         BUG_ON(!cache);
2058                         btrfs_add_free_space(cache, bytenr, num_bytes);
2059                         update_reserved_extents(root, bytenr, num_bytes, 0);
2060                         return 0;
2061                 }
2062                 pin = 1;
2063         }
2064
2065         /* if data pin when any transaction has committed this */
2066         if (ref_generation != trans->transid)
2067                 pin = 1;
2068
2069         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2070                             root_objectid, ref_generation,
2071                             owner_objectid, pin, pin == 0);
2072
2073         finish_current_insert(trans, root->fs_info->extent_root, 0);
2074         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2075         return ret ? ret : pending_ret;
2076 }
2077
2078 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2079                       struct btrfs_root *root,
2080                       u64 bytenr, u64 num_bytes, u64 parent,
2081                       u64 root_objectid, u64 ref_generation,
2082                       u64 owner_objectid, int pin)
2083 {
2084         int ret;
2085
2086         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2087                                   root_objectid, ref_generation,
2088                                   owner_objectid, pin);
2089         return ret;
2090 }
2091
2092 static u64 stripe_align(struct btrfs_root *root, u64 val)
2093 {
2094         u64 mask = ((u64)root->stripesize - 1);
2095         u64 ret = (val + mask) & ~mask;
2096         return ret;
2097 }
2098
2099 /*
2100  * walks the btree of allocated extents and find a hole of a given size.
2101  * The key ins is changed to record the hole:
2102  * ins->objectid == block start
2103  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2104  * ins->offset == number of blocks
2105  * Any available blocks before search_start are skipped.
2106  */
2107 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2108                                      struct btrfs_root *orig_root,
2109                                      u64 num_bytes, u64 empty_size,
2110                                      u64 search_start, u64 search_end,
2111                                      u64 hint_byte, struct btrfs_key *ins,
2112                                      u64 exclude_start, u64 exclude_nr,
2113                                      int data)
2114 {
2115         int ret = 0;
2116         struct btrfs_root * root = orig_root->fs_info->extent_root;
2117         u64 total_needed = num_bytes;
2118         u64 *last_ptr = NULL;
2119         u64 last_wanted = 0;
2120         struct btrfs_block_group_cache *block_group = NULL;
2121         int chunk_alloc_done = 0;
2122         int empty_cluster = 2 * 1024 * 1024;
2123         int allowed_chunk_alloc = 0;
2124         struct list_head *head = NULL, *cur = NULL;
2125         int loop = 0;
2126         struct btrfs_space_info *space_info;
2127
2128         WARN_ON(num_bytes < root->sectorsize);
2129         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2130         ins->objectid = 0;
2131         ins->offset = 0;
2132
2133         if (orig_root->ref_cows || empty_size)
2134                 allowed_chunk_alloc = 1;
2135
2136         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2137                 last_ptr = &root->fs_info->last_alloc;
2138                 empty_cluster = 64 * 1024;
2139         }
2140
2141         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2142                 last_ptr = &root->fs_info->last_data_alloc;
2143
2144         if (last_ptr) {
2145                 if (*last_ptr) {
2146                         hint_byte = *last_ptr;
2147                         last_wanted = *last_ptr;
2148                 } else
2149                         empty_size += empty_cluster;
2150         } else {
2151                 empty_cluster = 0;
2152         }
2153         search_start = max(search_start, first_logical_byte(root, 0));
2154         search_start = max(search_start, hint_byte);
2155         total_needed += empty_size;
2156
2157         if (search_start != last_wanted)
2158                 last_wanted = 0;
2159
2160         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2161         if (!block_group)
2162                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2163                                                              search_start);
2164         space_info = __find_space_info(root->fs_info, data);
2165
2166         down_read(&space_info->groups_sem);
2167         while (1) {
2168                 struct btrfs_free_space *free_space;
2169                 /*
2170                  * the only way this happens if our hint points to a block
2171                  * group thats not of the proper type, while looping this
2172                  * should never happen
2173                  */
2174                 WARN_ON(!block_group);
2175                 mutex_lock(&block_group->alloc_mutex);
2176                 if (unlikely(!block_group_bits(block_group, data)))
2177                         goto new_group;
2178
2179                 ret = cache_block_group(root, block_group);
2180                 if (ret) {
2181                         mutex_unlock(&block_group->alloc_mutex);
2182                         break;
2183                 }
2184
2185                 if (block_group->ro)
2186                         goto new_group;
2187
2188                 free_space = btrfs_find_free_space(block_group, search_start,
2189                                                    total_needed);
2190                 if (free_space) {
2191                         u64 start = block_group->key.objectid;
2192                         u64 end = block_group->key.objectid +
2193                                 block_group->key.offset;
2194
2195                         search_start = stripe_align(root, free_space->offset);
2196
2197                         /* move on to the next group */
2198                         if (search_start + num_bytes >= search_end)
2199                                 goto new_group;
2200
2201                         /* move on to the next group */
2202                         if (search_start + num_bytes > end)
2203                                 goto new_group;
2204
2205                         if (last_wanted && search_start != last_wanted) {
2206                                 total_needed += empty_cluster;
2207                                 last_wanted = 0;
2208                                 /*
2209                                  * if search_start is still in this block group
2210                                  * then we just re-search this block group
2211                                  */
2212                                 if (search_start >= start &&
2213                                     search_start < end) {
2214                                         mutex_unlock(&block_group->alloc_mutex);
2215                                         continue;
2216                                 }
2217
2218                                 /* else we go to the next block group */
2219                                 goto new_group;
2220                         }
2221
2222                         if (exclude_nr > 0 &&
2223                             (search_start + num_bytes > exclude_start &&
2224                              search_start < exclude_start + exclude_nr)) {
2225                                 search_start = exclude_start + exclude_nr;
2226                                 /*
2227                                  * if search_start is still in this block group
2228                                  * then we just re-search this block group
2229                                  */
2230                                 if (search_start >= start &&
2231                                     search_start < end) {
2232                                         mutex_unlock(&block_group->alloc_mutex);
2233                                         last_wanted = 0;
2234                                         continue;
2235                                 }
2236
2237                                 /* else we go to the next block group */
2238                                 goto new_group;
2239                         }
2240
2241                         ins->objectid = search_start;
2242                         ins->offset = num_bytes;
2243
2244                         btrfs_remove_free_space_lock(block_group, search_start,
2245                                                      num_bytes);
2246                         /* we are all good, lets return */
2247                         mutex_unlock(&block_group->alloc_mutex);
2248                         break;
2249                 }
2250 new_group:
2251                 last_wanted = 0;
2252                 if (loop > 0) {
2253                         total_needed -= empty_cluster;
2254                         empty_cluster = 0;
2255                 }
2256                 mutex_unlock(&block_group->alloc_mutex);
2257                 /*
2258                  * Here's how this works.
2259                  * loop == 0: we were searching a block group via a hint
2260                  *              and didn't find anything, so we start at
2261                  *              the head of the block groups and keep searching
2262                  * loop == 1: we're searching through all of the block groups
2263                  *              if we hit the head again we have searched
2264                  *              all of the block groups for this space and we
2265                  *              need to try and allocate, if we cant error out.
2266                  * loop == 2: we allocated more space and are looping through
2267                  *              all of the block groups again.
2268                  */
2269                 if (loop == 0) {
2270                         head = &space_info->block_groups;
2271                         cur = head->next;
2272                         loop++;
2273                 } else if (loop == 1 && cur == head) {
2274                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2275                                 up_read(&space_info->groups_sem);
2276                                 ret = do_chunk_alloc(trans, root, num_bytes +
2277                                                      2 * 1024 * 1024, data, 1);
2278                                 if (ret < 0)
2279                                         break;
2280                                 down_read(&space_info->groups_sem);
2281                                 loop++;
2282                                 head = &space_info->block_groups;
2283                                 cur = head->next;
2284                                 chunk_alloc_done = 1;
2285                         } else if (!allowed_chunk_alloc) {
2286                                 space_info->force_alloc = 1;
2287                                 break;
2288                         } else {
2289                                 break;
2290                         }
2291                 } else if (cur == head) {
2292                         break;
2293                 }
2294
2295                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2296                                          list);
2297                 search_start = block_group->key.objectid;
2298                 cur = cur->next;
2299         }
2300
2301         /* we found what we needed */
2302         if (ins->objectid) {
2303                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2304                         trans->block_group = block_group;
2305
2306                 if (last_ptr)
2307                         *last_ptr = ins->objectid + ins->offset;
2308                 ret = 0;
2309         } else if (!ret) {
2310                 ret = -ENOSPC;
2311         }
2312
2313         up_read(&space_info->groups_sem);
2314         return ret;
2315 }
2316
2317 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2318 {
2319         struct btrfs_block_group_cache *cache;
2320         struct list_head *l;
2321
2322         printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
2323                info->total_bytes - info->bytes_used - info->bytes_pinned -
2324                info->bytes_reserved, (info->full) ? "" : "not ");
2325
2326         down_read(&info->groups_sem);
2327         list_for_each(l, &info->block_groups) {
2328                 cache = list_entry(l, struct btrfs_block_group_cache, list);
2329                 spin_lock(&cache->lock);
2330                 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
2331                        "%Lu pinned %Lu reserved\n",
2332                        cache->key.objectid, cache->key.offset,
2333                        btrfs_block_group_used(&cache->item),
2334                        cache->pinned, cache->reserved);
2335                 btrfs_dump_free_space(cache, bytes);
2336                 spin_unlock(&cache->lock);
2337         }
2338         up_read(&info->groups_sem);
2339 }
2340
2341 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2342                                   struct btrfs_root *root,
2343                                   u64 num_bytes, u64 min_alloc_size,
2344                                   u64 empty_size, u64 hint_byte,
2345                                   u64 search_end, struct btrfs_key *ins,
2346                                   u64 data)
2347 {
2348         int ret;
2349         u64 search_start = 0;
2350         u64 alloc_profile;
2351         struct btrfs_fs_info *info = root->fs_info;
2352
2353         if (data) {
2354                 alloc_profile = info->avail_data_alloc_bits &
2355                                 info->data_alloc_profile;
2356                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2357         } else if (root == root->fs_info->chunk_root) {
2358                 alloc_profile = info->avail_system_alloc_bits &
2359                                 info->system_alloc_profile;
2360                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2361         } else {
2362                 alloc_profile = info->avail_metadata_alloc_bits &
2363                                 info->metadata_alloc_profile;
2364                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2365         }
2366 again:
2367         data = reduce_alloc_profile(root, data);
2368         /*
2369          * the only place that sets empty_size is btrfs_realloc_node, which
2370          * is not called recursively on allocations
2371          */
2372         if (empty_size || root->ref_cows) {
2373                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2374                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2375                                      2 * 1024 * 1024,
2376                                      BTRFS_BLOCK_GROUP_METADATA |
2377                                      (info->metadata_alloc_profile &
2378                                       info->avail_metadata_alloc_bits), 0);
2379                 }
2380                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2381                                      num_bytes + 2 * 1024 * 1024, data, 0);
2382         }
2383
2384         WARN_ON(num_bytes < root->sectorsize);
2385         ret = find_free_extent(trans, root, num_bytes, empty_size,
2386                                search_start, search_end, hint_byte, ins,
2387                                trans->alloc_exclude_start,
2388                                trans->alloc_exclude_nr, data);
2389
2390         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2391                 num_bytes = num_bytes >> 1;
2392                 num_bytes = num_bytes & ~(root->sectorsize - 1);
2393                 num_bytes = max(num_bytes, min_alloc_size);
2394                 do_chunk_alloc(trans, root->fs_info->extent_root,
2395                                num_bytes, data, 1);
2396                 goto again;
2397         }
2398         if (ret) {
2399                 struct btrfs_space_info *sinfo;
2400
2401                 sinfo = __find_space_info(root->fs_info, data);
2402                 printk("allocation failed flags %Lu, wanted %Lu\n",
2403                        data, num_bytes);
2404                 dump_space_info(sinfo, num_bytes);
2405                 BUG();
2406         }
2407
2408         return ret;
2409 }
2410
2411 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2412 {
2413         struct btrfs_block_group_cache *cache;
2414
2415         cache = btrfs_lookup_block_group(root->fs_info, start);
2416         if (!cache) {
2417                 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
2418                 return -ENOSPC;
2419         }
2420         btrfs_add_free_space(cache, start, len);
2421         update_reserved_extents(root, start, len, 0);
2422         return 0;
2423 }
2424
2425 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2426                                   struct btrfs_root *root,
2427                                   u64 num_bytes, u64 min_alloc_size,
2428                                   u64 empty_size, u64 hint_byte,
2429                                   u64 search_end, struct btrfs_key *ins,
2430                                   u64 data)
2431 {
2432         int ret;
2433         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2434                                      empty_size, hint_byte, search_end, ins,
2435                                      data);
2436         update_reserved_extents(root, ins->objectid, ins->offset, 1);
2437         return ret;
2438 }
2439
2440 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2441                                          struct btrfs_root *root, u64 parent,
2442                                          u64 root_objectid, u64 ref_generation,
2443                                          u64 owner, struct btrfs_key *ins)
2444 {
2445         int ret;
2446         int pending_ret;
2447         u64 super_used;
2448         u64 root_used;
2449         u64 num_bytes = ins->offset;
2450         u32 sizes[2];
2451         struct btrfs_fs_info *info = root->fs_info;
2452         struct btrfs_root *extent_root = info->extent_root;
2453         struct btrfs_extent_item *extent_item;
2454         struct btrfs_extent_ref *ref;
2455         struct btrfs_path *path;
2456         struct btrfs_key keys[2];
2457
2458         if (parent == 0)
2459                 parent = ins->objectid;
2460
2461         /* block accounting for super block */
2462         spin_lock_irq(&info->delalloc_lock);
2463         super_used = btrfs_super_bytes_used(&info->super_copy);
2464         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2465         spin_unlock_irq(&info->delalloc_lock);
2466
2467         /* block accounting for root item */
2468         root_used = btrfs_root_used(&root->root_item);
2469         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2470
2471         if (root == extent_root) {
2472                 struct pending_extent_op *extent_op;
2473
2474                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2475                 BUG_ON(!extent_op);
2476
2477                 extent_op->type = PENDING_EXTENT_INSERT;
2478                 extent_op->bytenr = ins->objectid;
2479                 extent_op->num_bytes = ins->offset;
2480                 extent_op->parent = parent;
2481                 extent_op->orig_parent = 0;
2482                 extent_op->generation = ref_generation;
2483                 extent_op->orig_generation = 0;
2484                 extent_op->level = (int)owner;
2485
2486                 mutex_lock(&root->fs_info->extent_ins_mutex);
2487                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2488                                 ins->objectid + ins->offset - 1,
2489                                 EXTENT_WRITEBACK, GFP_NOFS);
2490                 set_state_private(&root->fs_info->extent_ins,
2491                                   ins->objectid, (unsigned long)extent_op);
2492                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2493                 goto update_block;
2494         }
2495
2496         memcpy(&keys[0], ins, sizeof(*ins));
2497         keys[1].objectid = ins->objectid;
2498         keys[1].type = BTRFS_EXTENT_REF_KEY;
2499         keys[1].offset = parent;
2500         sizes[0] = sizeof(*extent_item);
2501         sizes[1] = sizeof(*ref);
2502
2503         path = btrfs_alloc_path();
2504         BUG_ON(!path);
2505
2506         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2507                                        sizes, 2);
2508         BUG_ON(ret);
2509
2510         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2511                                      struct btrfs_extent_item);
2512         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2513         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2514                              struct btrfs_extent_ref);
2515
2516         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2517         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2518         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2519         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
2520
2521         btrfs_mark_buffer_dirty(path->nodes[0]);
2522
2523         trans->alloc_exclude_start = 0;
2524         trans->alloc_exclude_nr = 0;
2525         btrfs_free_path(path);
2526         finish_current_insert(trans, extent_root, 0);
2527         pending_ret = del_pending_extents(trans, extent_root, 0);
2528
2529         if (ret)
2530                 goto out;
2531         if (pending_ret) {
2532                 ret = pending_ret;
2533                 goto out;
2534         }
2535
2536 update_block:
2537         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2538         if (ret) {
2539                 printk("update block group failed for %Lu %Lu\n",
2540                        ins->objectid, ins->offset);
2541                 BUG();
2542         }
2543 out:
2544         return ret;
2545 }
2546
2547 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2548                                 struct btrfs_root *root, u64 parent,
2549                                 u64 root_objectid, u64 ref_generation,
2550                                 u64 owner, struct btrfs_key *ins)
2551 {
2552         int ret;
2553
2554         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2555                 return 0;
2556         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2557                                             ref_generation, owner, ins);
2558         update_reserved_extents(root, ins->objectid, ins->offset, 0);
2559         return ret;
2560 }
2561
2562 /*
2563  * this is used by the tree logging recovery code.  It records that
2564  * an extent has been allocated and makes sure to clear the free
2565  * space cache bits as well
2566  */
2567 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2568                                 struct btrfs_root *root, u64 parent,
2569                                 u64 root_objectid, u64 ref_generation,
2570                                 u64 owner, struct btrfs_key *ins)
2571 {
2572         int ret;
2573         struct btrfs_block_group_cache *block_group;
2574
2575         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2576         mutex_lock(&block_group->alloc_mutex);
2577         cache_block_group(root, block_group);
2578
2579         ret = btrfs_remove_free_space_lock(block_group, ins->objectid,
2580                                            ins->offset);
2581         mutex_unlock(&block_group->alloc_mutex);
2582         BUG_ON(ret);
2583         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2584                                             ref_generation, owner, ins);
2585         return ret;
2586 }
2587
2588 /*
2589  * finds a free extent and does all the dirty work required for allocation
2590  * returns the key for the extent through ins, and a tree buffer for
2591  * the first block of the extent through buf.
2592  *
2593  * returns 0 if everything worked, non-zero otherwise.
2594  */
2595 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2596                        struct btrfs_root *root,
2597                        u64 num_bytes, u64 parent, u64 min_alloc_size,
2598                        u64 root_objectid, u64 ref_generation,
2599                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
2600                        u64 search_end, struct btrfs_key *ins, u64 data)
2601 {
2602         int ret;
2603
2604         ret = __btrfs_reserve_extent(trans, root, num_bytes,
2605                                      min_alloc_size, empty_size, hint_byte,
2606                                      search_end, ins, data);
2607         BUG_ON(ret);
2608         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2609                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2610                                         root_objectid, ref_generation,
2611                                         owner_objectid, ins);
2612                 BUG_ON(ret);
2613
2614         } else {
2615                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2616         }
2617         return ret;
2618 }
2619
2620 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2621                                             struct btrfs_root *root,
2622                                             u64 bytenr, u32 blocksize)
2623 {
2624         struct extent_buffer *buf;
2625
2626         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2627         if (!buf)
2628                 return ERR_PTR(-ENOMEM);
2629         btrfs_set_header_generation(buf, trans->transid);
2630         btrfs_tree_lock(buf);
2631         clean_tree_block(trans, root, buf);
2632         btrfs_set_buffer_uptodate(buf);
2633         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2634                 set_extent_dirty(&root->dirty_log_pages, buf->start,
2635                          buf->start + buf->len - 1, GFP_NOFS);
2636         } else {
2637                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2638                          buf->start + buf->len - 1, GFP_NOFS);
2639         }
2640         trans->blocks_used++;
2641         return buf;
2642 }
2643
2644 /*
2645  * helper function to allocate a block for a given tree
2646  * returns the tree buffer or NULL.
2647  */
2648 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2649                                              struct btrfs_root *root,
2650                                              u32 blocksize, u64 parent,
2651                                              u64 root_objectid,
2652                                              u64 ref_generation,
2653                                              int level,
2654                                              u64 hint,
2655                                              u64 empty_size)
2656 {
2657         struct btrfs_key ins;
2658         int ret;
2659         struct extent_buffer *buf;
2660
2661         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
2662                                  root_objectid, ref_generation, level,
2663                                  empty_size, hint, (u64)-1, &ins, 0);
2664         if (ret) {
2665                 BUG_ON(ret > 0);
2666                 return ERR_PTR(ret);
2667         }
2668
2669         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2670         return buf;
2671 }
2672
2673 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2674                         struct btrfs_root *root, struct extent_buffer *leaf)
2675 {
2676         u64 leaf_owner;
2677         u64 leaf_generation;
2678         struct btrfs_key key;
2679         struct btrfs_file_extent_item *fi;
2680         int i;
2681         int nritems;
2682         int ret;
2683
2684         BUG_ON(!btrfs_is_leaf(leaf));
2685         nritems = btrfs_header_nritems(leaf);
2686         leaf_owner = btrfs_header_owner(leaf);
2687         leaf_generation = btrfs_header_generation(leaf);
2688
2689         for (i = 0; i < nritems; i++) {
2690                 u64 disk_bytenr;
2691                 cond_resched();
2692
2693                 btrfs_item_key_to_cpu(leaf, &key, i);
2694                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2695                         continue;
2696                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2697                 if (btrfs_file_extent_type(leaf, fi) ==
2698                     BTRFS_FILE_EXTENT_INLINE)
2699                         continue;
2700                 /*
2701                  * FIXME make sure to insert a trans record that
2702                  * repeats the snapshot del on crash
2703                  */
2704                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2705                 if (disk_bytenr == 0)
2706                         continue;
2707
2708                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2709                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2710                                 leaf->start, leaf_owner, leaf_generation,
2711                                 key.objectid, 0);
2712                 BUG_ON(ret);
2713
2714                 atomic_inc(&root->fs_info->throttle_gen);
2715                 wake_up(&root->fs_info->transaction_throttle);
2716                 cond_resched();
2717         }
2718         return 0;
2719 }
2720
2721 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2722                                         struct btrfs_root *root,
2723                                         struct btrfs_leaf_ref *ref)
2724 {
2725         int i;
2726         int ret;
2727         struct btrfs_extent_info *info = ref->extents;
2728
2729         for (i = 0; i < ref->nritems; i++) {
2730                 ret = __btrfs_free_extent(trans, root, info->bytenr,
2731                                           info->num_bytes, ref->bytenr,
2732                                           ref->owner, ref->generation,
2733                                           info->objectid, 0);
2734
2735                 atomic_inc(&root->fs_info->throttle_gen);
2736                 wake_up(&root->fs_info->transaction_throttle);
2737                 cond_resched();
2738
2739                 BUG_ON(ret);
2740                 info++;
2741         }
2742
2743         return 0;
2744 }
2745
2746 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2747                               u32 *refs)
2748 {
2749         int ret;
2750
2751         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
2752         BUG_ON(ret);
2753
2754 #if 0 // some debugging code in case we see problems here
2755         /* if the refs count is one, it won't get increased again.  But
2756          * if the ref count is > 1, someone may be decreasing it at
2757          * the same time we are.
2758          */
2759         if (*refs != 1) {
2760                 struct extent_buffer *eb = NULL;
2761                 eb = btrfs_find_create_tree_block(root, start, len);
2762                 if (eb)
2763                         btrfs_tree_lock(eb);
2764
2765                 mutex_lock(&root->fs_info->alloc_mutex);
2766                 ret = lookup_extent_ref(NULL, root, start, len, refs);
2767                 BUG_ON(ret);
2768                 mutex_unlock(&root->fs_info->alloc_mutex);
2769
2770                 if (eb) {
2771                         btrfs_tree_unlock(eb);
2772                         free_extent_buffer(eb);
2773                 }
2774                 if (*refs == 1) {
2775                         printk("block %llu went down to one during drop_snap\n",
2776                                (unsigned long long)start);
2777                 }
2778
2779         }
2780 #endif
2781
2782         cond_resched();
2783         return ret;
2784 }
2785
2786 /*
2787  * helper function for drop_snapshot, this walks down the tree dropping ref
2788  * counts as it goes.
2789  */
2790 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2791                                    struct btrfs_root *root,
2792                                    struct btrfs_path *path, int *level)
2793 {
2794         u64 root_owner;
2795         u64 root_gen;
2796         u64 bytenr;
2797         u64 ptr_gen;
2798         struct extent_buffer *next;
2799         struct extent_buffer *cur;
2800         struct extent_buffer *parent;
2801         struct btrfs_leaf_ref *ref;
2802         u32 blocksize;
2803         int ret;
2804         u32 refs;
2805
2806         WARN_ON(*level < 0);
2807         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2808         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2809                                 path->nodes[*level]->len, &refs);
2810         BUG_ON(ret);
2811         if (refs > 1)
2812                 goto out;
2813
2814         /*
2815          * walk down to the last node level and free all the leaves
2816          */
2817         while(*level >= 0) {
2818                 WARN_ON(*level < 0);
2819                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2820                 cur = path->nodes[*level];
2821
2822                 if (btrfs_header_level(cur) != *level)
2823                         WARN_ON(1);
2824
2825                 if (path->slots[*level] >=
2826                     btrfs_header_nritems(cur))
2827                         break;
2828                 if (*level == 0) {
2829                         ret = btrfs_drop_leaf_ref(trans, root, cur);
2830                         BUG_ON(ret);
2831                         break;
2832                 }
2833                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2834                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2835                 blocksize = btrfs_level_size(root, *level - 1);
2836
2837                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2838                 BUG_ON(ret);
2839                 if (refs != 1) {
2840                         parent = path->nodes[*level];
2841                         root_owner = btrfs_header_owner(parent);
2842                         root_gen = btrfs_header_generation(parent);
2843                         path->slots[*level]++;
2844
2845                         ret = __btrfs_free_extent(trans, root, bytenr,
2846                                                 blocksize, parent->start,
2847                                                 root_owner, root_gen,
2848                                                 *level - 1, 1);
2849                         BUG_ON(ret);
2850
2851                         atomic_inc(&root->fs_info->throttle_gen);
2852                         wake_up(&root->fs_info->transaction_throttle);
2853                         cond_resched();
2854
2855                         continue;
2856                 }
2857                 /*
2858                  * at this point, we have a single ref, and since the
2859                  * only place referencing this extent is a dead root
2860                  * the reference count should never go higher.
2861                  * So, we don't need to check it again
2862                  */
2863                 if (*level == 1) {
2864                         ref = btrfs_lookup_leaf_ref(root, bytenr);
2865                         if (ref && ref->generation != ptr_gen) {
2866                                 btrfs_free_leaf_ref(root, ref);
2867                                 ref = NULL;
2868                         }
2869                         if (ref) {
2870                                 ret = cache_drop_leaf_ref(trans, root, ref);
2871                                 BUG_ON(ret);
2872                                 btrfs_remove_leaf_ref(root, ref);
2873                                 btrfs_free_leaf_ref(root, ref);
2874                                 *level = 0;
2875                                 break;
2876                         }
2877                         if (printk_ratelimit()) {
2878                                 printk("leaf ref miss for bytenr %llu\n",
2879                                        (unsigned long long)bytenr);
2880                         }
2881                 }
2882                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2883                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2884                         free_extent_buffer(next);
2885
2886                         next = read_tree_block(root, bytenr, blocksize,
2887                                                ptr_gen);
2888                         cond_resched();
2889 #if 0
2890                         /*
2891                          * this is a debugging check and can go away
2892                          * the ref should never go all the way down to 1
2893                          * at this point
2894                          */
2895                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2896                                                 &refs);
2897                         BUG_ON(ret);
2898                         WARN_ON(refs != 1);
2899 #endif
2900                 }
2901                 WARN_ON(*level <= 0);
2902                 if (path->nodes[*level-1])
2903                         free_extent_buffer(path->nodes[*level-1]);
2904                 path->nodes[*level-1] = next;
2905                 *level = btrfs_header_level(next);
2906                 path->slots[*level] = 0;
2907                 cond_resched();
2908         }
2909 out:
2910         WARN_ON(*level < 0);
2911         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2912
2913         if (path->nodes[*level] == root->node) {
2914                 parent = path->nodes[*level];
2915                 bytenr = path->nodes[*level]->start;
2916         } else {
2917                 parent = path->nodes[*level + 1];
2918                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
2919         }
2920
2921         blocksize = btrfs_level_size(root, *level);
2922         root_owner = btrfs_header_owner(parent);
2923         root_gen = btrfs_header_generation(parent);
2924
2925         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
2926                                   parent->start, root_owner, root_gen,
2927                                   *level, 1);
2928         free_extent_buffer(path->nodes[*level]);
2929         path->nodes[*level] = NULL;
2930         *level += 1;
2931         BUG_ON(ret);
2932
2933         cond_resched();
2934         return 0;
2935 }
2936
2937 /*
2938  * helper function for drop_subtree, this function is similar to
2939  * walk_down_tree. The main difference is that it checks reference
2940  * counts while tree blocks are locked.
2941  */
2942 static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
2943                                       struct btrfs_root *root,
2944                                       struct btrfs_path *path, int *level)
2945 {
2946         struct extent_buffer *next;
2947         struct extent_buffer *cur;
2948         struct extent_buffer *parent;
2949         u64 bytenr;
2950         u64 ptr_gen;
2951         u32 blocksize;
2952         u32 refs;
2953         int ret;
2954
2955         cur = path->nodes[*level];
2956         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
2957                                       &refs);
2958         BUG_ON(ret);
2959         if (refs > 1)
2960                 goto out;
2961
2962         while (*level >= 0) {
2963                 cur = path->nodes[*level];
2964                 if (*level == 0) {
2965                         ret = btrfs_drop_leaf_ref(trans, root, cur);
2966                         BUG_ON(ret);
2967                         clean_tree_block(trans, root, cur);
2968                         break;
2969                 }
2970                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
2971                         clean_tree_block(trans, root, cur);
2972                         break;
2973                 }
2974
2975                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2976                 blocksize = btrfs_level_size(root, *level - 1);
2977                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2978
2979                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
2980                 btrfs_tree_lock(next);
2981
2982                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2983                                               &refs);
2984                 BUG_ON(ret);
2985                 if (refs > 1) {
2986                         parent = path->nodes[*level];
2987                         ret = btrfs_free_extent(trans, root, bytenr,
2988                                         blocksize, parent->start,
2989                                         btrfs_header_owner(parent),
2990                                         btrfs_header_generation(parent),
2991                                         *level - 1, 1);
2992                         BUG_ON(ret);
2993                         path->slots[*level]++;
2994                         btrfs_tree_unlock(next);
2995                         free_extent_buffer(next);
2996                         continue;
2997                 }
2998
2999                 *level = btrfs_header_level(next);
3000                 path->nodes[*level] = next;
3001                 path->slots[*level] = 0;
3002                 path->locks[*level] = 1;
3003                 cond_resched();
3004         }
3005 out:
3006         parent = path->nodes[*level + 1];
3007         bytenr = path->nodes[*level]->start;
3008         blocksize = path->nodes[*level]->len;
3009
3010         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3011                         parent->start, btrfs_header_owner(parent),
3012                         btrfs_header_generation(parent), *level, 1);
3013         BUG_ON(ret);
3014
3015         if (path->locks[*level]) {
3016                 btrfs_tree_unlock(path->nodes[*level]);
3017                 path->locks[*level] = 0;
3018         }
3019         free_extent_buffer(path->nodes[*level]);
3020         path->nodes[*level] = NULL;
3021         *level += 1;
3022         cond_resched();
3023         return 0;
3024 }
3025
3026 /*
3027  * helper for dropping snapshots.  This walks back up the tree in the path
3028  * to find the first node higher up where we haven't yet gone through
3029  * all the slots
3030  */
3031 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3032                                  struct btrfs_root *root,
3033                                  struct btrfs_path *path,
3034                                  int *level, int max_level)
3035 {
3036         u64 root_owner;
3037         u64 root_gen;
3038         struct btrfs_root_item *root_item = &root->root_item;
3039         int i;
3040         int slot;
3041         int ret;
3042
3043         for (i = *level; i < max_level && path->nodes[i]; i++) {
3044                 slot = path->slots[i];
3045                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3046                         struct extent_buffer *node;
3047                         struct btrfs_disk_key disk_key;
3048                         node = path->nodes[i];
3049                         path->slots[i]++;
3050                         *level = i;
3051                         WARN_ON(*level == 0);
3052                         btrfs_node_key(node, &disk_key, path->slots[i]);
3053                         memcpy(&root_item->drop_progress,
3054                                &disk_key, sizeof(disk_key));
3055                         root_item->drop_level = i;
3056                         return 0;
3057                 } else {
3058                         struct extent_buffer *parent;
3059                         if (path->nodes[*level] == root->node)
3060                                 parent = path->nodes[*level];
3061                         else
3062                                 parent = path->nodes[*level + 1];
3063
3064                         root_owner = btrfs_header_owner(parent);
3065                         root_gen = btrfs_header_generation(parent);
3066
3067                         clean_tree_block(trans, root, path->nodes[*level]);
3068                         ret = btrfs_free_extent(trans, root,
3069                                                 path->nodes[*level]->start,
3070                                                 path->nodes[*level]->len,
3071                                                 parent->start, root_owner,
3072                                                 root_gen, *level, 1);
3073                         BUG_ON(ret);
3074                         if (path->locks[*level]) {
3075                                 btrfs_tree_unlock(path->nodes[*level]);
3076                                 path->locks[*level] = 0;
3077                         }
3078                         free_extent_buffer(path->nodes[*level]);
3079                         path->nodes[*level] = NULL;
3080                         *level = i + 1;
3081                 }
3082         }
3083         return 1;
3084 }
3085
3086 /*
3087  * drop the reference count on the tree rooted at 'snap'.  This traverses
3088  * the tree freeing any blocks that have a ref count of zero after being
3089  * decremented.
3090  */
3091 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3092                         *root)
3093 {
3094         int ret = 0;
3095         int wret;
3096         int level;
3097         struct btrfs_path *path;
3098         int i;
3099         int orig_level;
3100         struct btrfs_root_item *root_item = &root->root_item;
3101
3102         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3103         path = btrfs_alloc_path();
3104         BUG_ON(!path);
3105
3106         level = btrfs_header_level(root->node);
3107         orig_level = level;
3108         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3109                 path->nodes[level] = root->node;
3110                 extent_buffer_get(root->node);
3111                 path->slots[level] = 0;
3112         } else {
3113                 struct btrfs_key key;
3114                 struct btrfs_disk_key found_key;
3115                 struct extent_buffer *node;
3116
3117                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3118                 level = root_item->drop_level;
3119                 path->lowest_level = level;
3120                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3121                 if (wret < 0) {
3122                         ret = wret;
3123                         goto out;
3124                 }
3125                 node = path->nodes[level];
3126                 btrfs_node_key(node, &found_key, path->slots[level]);
3127                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3128                                sizeof(found_key)));
3129                 /*
3130                  * unlock our path, this is safe because only this
3131                  * function is allowed to delete this snapshot
3132                  */
3133                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3134                         if (path->nodes[i] && path->locks[i]) {
3135                                 path->locks[i] = 0;
3136                                 btrfs_tree_unlock(path->nodes[i]);
3137                         }
3138                 }
3139         }
3140         while(1) {
3141                 wret = walk_down_tree(trans, root, path, &level);
3142                 if (wret > 0)
3143                         break;
3144                 if (wret < 0)
3145                         ret = wret;
3146
3147                 wret = walk_up_tree(trans, root, path, &level,
3148                                     BTRFS_MAX_LEVEL);
3149                 if (wret > 0)
3150                         break;
3151                 if (wret < 0)
3152                         ret = wret;
3153                 if (trans->transaction->in_commit) {
3154                         ret = -EAGAIN;
3155                         break;
3156                 }
3157                 atomic_inc(&root->fs_info->throttle_gen);
3158                 wake_up(&root->fs_info->transaction_throttle);
3159         }
3160         for (i = 0; i <= orig_level; i++) {
3161                 if (path->nodes[i]) {
3162                         free_extent_buffer(path->nodes[i]);
3163                         path->nodes[i] = NULL;
3164                 }
3165         }
3166 out:
3167         btrfs_free_path(path);
3168         return ret;
3169 }
3170
3171 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3172                         struct btrfs_root *root,
3173                         struct extent_buffer *node,
3174                         struct extent_buffer *parent)
3175 {
3176         struct btrfs_path *path;
3177         int level;
3178         int parent_level;
3179         int ret = 0;
3180         int wret;
3181
3182         path = btrfs_alloc_path();
3183         BUG_ON(!path);
3184
3185         BUG_ON(!btrfs_tree_locked(parent));
3186         parent_level = btrfs_header_level(parent);
3187         extent_buffer_get(parent);
3188         path->nodes[parent_level] = parent;
3189         path->slots[parent_level] = btrfs_header_nritems(parent);
3190
3191         BUG_ON(!btrfs_tree_locked(node));
3192         level = btrfs_header_level(node);
3193         extent_buffer_get(node);
3194         path->nodes[level] = node;
3195         path->slots[level] = 0;
3196
3197         while (1) {
3198                 wret = walk_down_subtree(trans, root, path, &level);
3199                 if (wret < 0)
3200                         ret = wret;
3201                 if (wret != 0)
3202                         break;
3203
3204                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3205                 if (wret < 0)
3206                         ret = wret;
3207                 if (wret != 0)
3208                         break;
3209         }
3210
3211         btrfs_free_path(path);
3212         return ret;
3213 }
3214
3215 static unsigned long calc_ra(unsigned long start, unsigned long last,
3216                              unsigned long nr)
3217 {
3218         return min(last, start + nr - 1);
3219 }
3220
3221 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3222                                          u64 len)
3223 {
3224         u64 page_start;
3225         u64 page_end;
3226         unsigned long first_index;
3227         unsigned long last_index;
3228         unsigned long i;
3229         struct page *page;
3230         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3231         struct file_ra_state *ra;
3232         struct btrfs_ordered_extent *ordered;
3233         unsigned int total_read = 0;
3234         unsigned int total_dirty = 0;
3235         int ret = 0;
3236
3237         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3238
3239         mutex_lock(&inode->i_mutex);
3240         first_index = start >> PAGE_CACHE_SHIFT;
3241         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3242
3243         /* make sure the dirty trick played by the caller work */
3244         ret = invalidate_inode_pages2_range(inode->i_mapping,
3245                                             first_index, last_index);
3246         if (ret)
3247                 goto out_unlock;
3248
3249         file_ra_state_init(ra, inode->i_mapping);
3250
3251         for (i = first_index ; i <= last_index; i++) {
3252                 if (total_read % ra->ra_pages == 0) {
3253                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3254                                        calc_ra(i, last_index, ra->ra_pages));
3255                 }
3256                 total_read++;
3257 again:
3258                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3259                         BUG_ON(1);
3260                 page = grab_cache_page(inode->i_mapping, i);
3261                 if (!page) {
3262                         ret = -ENOMEM;
3263                         goto out_unlock;
3264                 }
3265                 if (!PageUptodate(page)) {
3266                         btrfs_readpage(NULL, page);
3267                         lock_page(page);
3268                         if (!PageUptodate(page)) {
3269                                 unlock_page(page);
3270                                 page_cache_release(page);
3271                                 ret = -EIO;
3272                                 goto out_unlock;
3273                         }
3274                 }
3275                 wait_on_page_writeback(page);
3276
3277                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3278                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3279                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3280
3281                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3282                 if (ordered) {
3283                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3284                         unlock_page(page);
3285                         page_cache_release(page);
3286                         btrfs_start_ordered_extent(inode, ordered, 1);
3287                         btrfs_put_ordered_extent(ordered);
3288                         goto again;
3289                 }
3290                 set_page_extent_mapped(page);
3291
3292                 btrfs_set_extent_delalloc(inode, page_start, page_end);
3293                 if (i == first_index)
3294                         set_extent_bits(io_tree, page_start, page_end,
3295                                         EXTENT_BOUNDARY, GFP_NOFS);
3296
3297                 set_page_dirty(page);
3298                 total_dirty++;
3299
3300                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3301                 unlock_page(page);
3302                 page_cache_release(page);
3303         }
3304
3305 out_unlock:
3306         kfree(ra);
3307         mutex_unlock(&inode->i_mutex);
3308         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3309         return ret;
3310 }
3311
3312 static int noinline relocate_data_extent(struct inode *reloc_inode,
3313                                          struct btrfs_key *extent_key,
3314                                          u64 offset)
3315 {
3316         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3317         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3318         struct extent_map *em;
3319         u64 start = extent_key->objectid - offset;
3320         u64 end = start + extent_key->offset - 1;
3321
3322         em = alloc_extent_map(GFP_NOFS);
3323         BUG_ON(!em || IS_ERR(em));
3324
3325         em->start = start;
3326         em->len = extent_key->offset;
3327         em->block_len = extent_key->offset;
3328         em->block_start = extent_key->objectid;
3329         em->bdev = root->fs_info->fs_devices->latest_bdev;
3330         set_bit(EXTENT_FLAG_PINNED, &em->flags);
3331
3332         /* setup extent map to cheat btrfs_readpage */
3333         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3334         while (1) {
3335                 int ret;
3336                 spin_lock(&em_tree->lock);
3337                 ret = add_extent_mapping(em_tree, em);
3338                 spin_unlock(&em_tree->lock);
3339                 if (ret != -EEXIST) {
3340                         free_extent_map(em);
3341                         break;
3342                 }
3343                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
3344         }
3345         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3346
3347         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
3348 }
3349
3350 struct btrfs_ref_path {
3351         u64 extent_start;
3352         u64 nodes[BTRFS_MAX_LEVEL];
3353         u64 root_objectid;
3354         u64 root_generation;
3355         u64 owner_objectid;
3356         u32 num_refs;
3357         int lowest_level;
3358         int current_level;
3359         int shared_level;
3360
3361         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
3362         u64 new_nodes[BTRFS_MAX_LEVEL];
3363 };
3364
3365 struct disk_extent {
3366         u64 ram_bytes;
3367         u64 disk_bytenr;
3368         u64 disk_num_bytes;
3369         u64 offset;
3370         u64 num_bytes;
3371         u8 compression;
3372         u8 encryption;
3373         u16 other_encoding;
3374 };
3375
3376 static int is_cowonly_root(u64 root_objectid)
3377 {
3378         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3379             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3380             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3381             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3382             root_objectid == BTRFS_TREE_LOG_OBJECTID)
3383                 return 1;
3384         return 0;
3385 }
3386
3387 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
3388                                     struct btrfs_root *extent_root,
3389                                     struct btrfs_ref_path *ref_path,
3390                                     int first_time)
3391 {
3392         struct extent_buffer *leaf;
3393         struct btrfs_path *path;
3394         struct btrfs_extent_ref *ref;
3395         struct btrfs_key key;
3396         struct btrfs_key found_key;
3397         u64 bytenr;
3398         u32 nritems;
3399         int level;
3400         int ret = 1;
3401
3402         path = btrfs_alloc_path();
3403         if (!path)
3404                 return -ENOMEM;
3405
3406         if (first_time) {
3407                 ref_path->lowest_level = -1;
3408                 ref_path->current_level = -1;
3409                 ref_path->shared_level = -1;
3410                 goto walk_up;
3411         }
3412 walk_down:
3413         level = ref_path->current_level - 1;
3414         while (level >= -1) {
3415                 u64 parent;
3416                 if (level < ref_path->lowest_level)
3417                         break;
3418
3419                 if (level >= 0) {
3420                         bytenr = ref_path->nodes[level];
3421                 } else {
3422                         bytenr = ref_path->extent_start;
3423                 }
3424                 BUG_ON(bytenr == 0);
3425
3426                 parent = ref_path->nodes[level + 1];
3427                 ref_path->nodes[level + 1] = 0;
3428                 ref_path->current_level = level;
3429                 BUG_ON(parent == 0);
3430
3431                 key.objectid = bytenr;
3432                 key.offset = parent + 1;
3433                 key.type = BTRFS_EXTENT_REF_KEY;
3434
3435                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3436                 if (ret < 0)
3437                         goto out;
3438                 BUG_ON(ret == 0);
3439
3440                 leaf = path->nodes[0];
3441                 nritems = btrfs_header_nritems(leaf);
3442                 if (path->slots[0] >= nritems) {
3443                         ret = btrfs_next_leaf(extent_root, path);
3444                         if (ret < 0)
3445                                 goto out;
3446                         if (ret > 0)
3447                                 goto next;
3448                         leaf = path->nodes[0];
3449                 }
3450
3451                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3452                 if (found_key.objectid == bytenr &&
3453                     found_key.type == BTRFS_EXTENT_REF_KEY) {
3454                         if (level < ref_path->shared_level)
3455                                 ref_path->shared_level = level;
3456                         goto found;
3457                 }
3458 next:
3459                 level--;
3460                 btrfs_release_path(extent_root, path);
3461                 cond_resched();
3462         }
3463         /* reached lowest level */
3464         ret = 1;
3465         goto out;
3466 walk_up:
3467         level = ref_path->current_level;
3468         while (level < BTRFS_MAX_LEVEL - 1) {
3469                 u64 ref_objectid;
3470                 if (level >= 0) {
3471                         bytenr = ref_path->nodes[level];
3472                 } else {
3473                         bytenr = ref_path->extent_start;
3474                 }
3475                 BUG_ON(bytenr == 0);
3476
3477                 key.objectid = bytenr;
3478                 key.offset = 0;
3479                 key.type = BTRFS_EXTENT_REF_KEY;
3480
3481                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3482                 if (ret < 0)
3483                         goto out;
3484
3485                 leaf = path->nodes[0];
3486                 nritems = btrfs_header_nritems(leaf);
3487                 if (path->slots[0] >= nritems) {
3488                         ret = btrfs_next_leaf(extent_root, path);
3489                         if (ret < 0)
3490                                 goto out;
3491                         if (ret > 0) {
3492                                 /* the extent was freed by someone */
3493                                 if (ref_path->lowest_level == level)
3494                                         goto out;
3495                                 btrfs_release_path(extent_root, path);
3496                                 goto walk_down;
3497                         }
3498                         leaf = path->nodes[0];
3499                 }
3500
3501                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3502                 if (found_key.objectid != bytenr ||
3503                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
3504                         /* the extent was freed by someone */
3505                         if (ref_path->lowest_level == level) {
3506                                 ret = 1;
3507                                 goto out;
3508                         }
3509                         btrfs_release_path(extent_root, path);
3510                         goto walk_down;
3511                 }
3512 found:
3513                 ref = btrfs_item_ptr(leaf, path->slots[0],
3514                                 struct btrfs_extent_ref);
3515                 ref_objectid = btrfs_ref_objectid(leaf, ref);
3516                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3517                         if (first_time) {
3518                                 level = (int)ref_objectid;
3519                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
3520                                 ref_path->lowest_level = level;
3521                                 ref_path->current_level = level;
3522                                 ref_path->nodes[level] = bytenr;
3523                         } else {
3524                                 WARN_ON(ref_objectid != level);
3525                         }
3526                 } else {
3527                         WARN_ON(level != -1);
3528                 }
3529                 first_time = 0;
3530
3531                 if (ref_path->lowest_level == level) {
3532                         ref_path->owner_objectid = ref_objectid;
3533                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
3534                 }
3535
3536                 /*
3537                  * the block is tree root or the block isn't in reference
3538                  * counted tree.
3539                  */
3540                 if (found_key.objectid == found_key.offset ||
3541                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
3542                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3543                         ref_path->root_generation =
3544                                 btrfs_ref_generation(leaf, ref);
3545                         if (level < 0) {
3546                                 /* special reference from the tree log */
3547                                 ref_path->nodes[0] = found_key.offset;
3548                                 ref_path->current_level = 0;
3549                         }
3550                         ret = 0;
3551                         goto out;
3552                 }
3553
3554                 level++;
3555                 BUG_ON(ref_path->nodes[level] != 0);
3556                 ref_path->nodes[level] = found_key.offset;
3557                 ref_path->current_level = level;
3558
3559                 /*
3560                  * the reference was created in the running transaction,
3561                  * no need to continue walking up.
3562                  */
3563                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
3564                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3565                         ref_path->root_generation =
3566                                 btrfs_ref_generation(leaf, ref);
3567                         ret = 0;
3568                         goto out;
3569                 }
3570
3571                 btrfs_release_path(extent_root, path);
3572                 cond_resched();
3573         }
3574         /* reached max tree level, but no tree root found. */
3575         BUG();
3576 out:
3577         btrfs_free_path(path);
3578         return ret;
3579 }
3580
3581 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
3582                                 struct btrfs_root *extent_root,
3583                                 struct btrfs_ref_path *ref_path,
3584                                 u64 extent_start)
3585 {
3586         memset(ref_path, 0, sizeof(*ref_path));
3587         ref_path->extent_start = extent_start;
3588
3589         return __next_ref_path(trans, extent_root, ref_path, 1);
3590 }
3591
3592 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
3593                                struct btrfs_root *extent_root,
3594                                struct btrfs_ref_path *ref_path)
3595 {
3596         return __next_ref_path(trans, extent_root, ref_path, 0);
3597 }
3598
3599 static int noinline get_new_locations(struct inode *reloc_inode,
3600                                       struct btrfs_key *extent_key,
3601                                       u64 offset, int no_fragment,
3602                                       struct disk_extent **extents,
3603                                       int *nr_extents)
3604 {
3605         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3606         struct btrfs_path *path;
3607         struct btrfs_file_extent_item *fi;
3608         struct extent_buffer *leaf;
3609         struct disk_extent *exts = *extents;
3610         struct btrfs_key found_key;
3611         u64 cur_pos;
3612         u64 last_byte;
3613         u32 nritems;
3614         int nr = 0;
3615         int max = *nr_extents;
3616         int ret;
3617
3618         WARN_ON(!no_fragment && *extents);
3619         if (!exts) {
3620                 max = 1;
3621                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
3622                 if (!exts)
3623                         return -ENOMEM;
3624         }
3625
3626         path = btrfs_alloc_path();
3627         BUG_ON(!path);
3628
3629         cur_pos = extent_key->objectid - offset;
3630         last_byte = extent_key->objectid + extent_key->offset;
3631         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
3632                                        cur_pos, 0);
3633         if (ret < 0)
3634                 goto out;
3635         if (ret > 0) {
3636                 ret = -ENOENT;
3637                 goto out;
3638         }
3639
3640         while (1) {
3641                 leaf = path->nodes[0];
3642                 nritems = btrfs_header_nritems(leaf);
3643                 if (path->slots[0] >= nritems) {
3644                         ret = btrfs_next_leaf(root, path);
3645                         if (ret < 0)
3646                                 goto out;
3647                         if (ret > 0)
3648                                 break;
3649                         leaf = path->nodes[0];
3650                 }
3651
3652                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3653                 if (found_key.offset != cur_pos ||
3654                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
3655                     found_key.objectid != reloc_inode->i_ino)
3656                         break;
3657
3658                 fi = btrfs_item_ptr(leaf, path->slots[0],
3659                                     struct btrfs_file_extent_item);
3660                 if (btrfs_file_extent_type(leaf, fi) !=
3661                     BTRFS_FILE_EXTENT_REG ||
3662                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
3663                         break;
3664
3665                 if (nr == max) {
3666                         struct disk_extent *old = exts;
3667                         max *= 2;
3668                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
3669                         memcpy(exts, old, sizeof(*exts) * nr);
3670                         if (old != *extents)
3671                                 kfree(old);
3672                 }
3673
3674                 exts[nr].disk_bytenr =
3675                         btrfs_file_extent_disk_bytenr(leaf, fi);
3676                 exts[nr].disk_num_bytes =
3677                         btrfs_file_extent_disk_num_bytes(leaf, fi);
3678                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
3679                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3680                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
3681                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
3682                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
3683                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
3684                                                                            fi);
3685                 BUG_ON(exts[nr].offset > 0);
3686                 BUG_ON(exts[nr].compression || exts[nr].encryption);
3687                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
3688
3689                 cur_pos += exts[nr].num_bytes;
3690                 nr++;
3691
3692                 if (cur_pos + offset >= last_byte)
3693                         break;
3694
3695                 if (no_fragment) {
3696                         ret = 1;
3697                         goto out;
3698                 }
3699                 path->slots[0]++;
3700         }
3701
3702         WARN_ON(cur_pos + offset > last_byte);
3703         if (cur_pos + offset < last_byte) {
3704                 ret = -ENOENT;
3705                 goto out;
3706         }
3707         ret = 0;
3708 out:
3709         btrfs_free_path(path);
3710         if (ret) {
3711                 if (exts != *extents)
3712                         kfree(exts);
3713         } else {
3714                 *extents = exts;
3715                 *nr_extents = nr;
3716         }
3717         return ret;
3718 }
3719
3720 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
3721                                         struct btrfs_root *root,
3722                                         struct btrfs_path *path,
3723                                         struct btrfs_key *extent_key,
3724                                         struct btrfs_key *leaf_key,
3725                                         struct btrfs_ref_path *ref_path,
3726                                         struct disk_extent *new_extents,
3727                                         int nr_extents)
3728 {
3729         struct extent_buffer *leaf;
3730         struct btrfs_file_extent_item *fi;
3731         struct inode *inode = NULL;
3732         struct btrfs_key key;
3733         u64 lock_start = 0;
3734         u64 lock_end = 0;
3735         u64 num_bytes;
3736         u64 ext_offset;
3737         u64 first_pos;
3738         u32 nritems;
3739         int nr_scaned = 0;
3740         int extent_locked = 0;
3741         int extent_type;
3742         int ret;
3743
3744         memcpy(&key, leaf_key, sizeof(key));
3745         first_pos = INT_LIMIT(loff_t) - extent_key->offset;
3746         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3747                 if (key.objectid < ref_path->owner_objectid ||
3748                     (key.objectid == ref_path->owner_objectid &&
3749                      key.type < BTRFS_EXTENT_DATA_KEY)) {
3750                         key.objectid = ref_path->owner_objectid;
3751                         key.type = BTRFS_EXTENT_DATA_KEY;
3752                         key.offset = 0;
3753                 }
3754         }
3755
3756         while (1) {
3757                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3758                 if (ret < 0)
3759                         goto out;
3760
3761                 leaf = path->nodes[0];
3762                 nritems = btrfs_header_nritems(leaf);
3763 next:
3764                 if (extent_locked && ret > 0) {
3765                         /*
3766                          * the file extent item was modified by someone
3767                          * before the extent got locked.
3768                          */
3769                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3770                                       lock_end, GFP_NOFS);
3771                         extent_locked = 0;
3772                 }
3773
3774                 if (path->slots[0] >= nritems) {
3775                         if (++nr_scaned > 2)
3776                                 break;
3777
3778                         BUG_ON(extent_locked);
3779                         ret = btrfs_next_leaf(root, path);
3780                         if (ret < 0)
3781                                 goto out;
3782                         if (ret > 0)
3783                                 break;
3784                         leaf = path->nodes[0];
3785                         nritems = btrfs_header_nritems(leaf);
3786                 }
3787
3788                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3789
3790                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3791                         if ((key.objectid > ref_path->owner_objectid) ||
3792                             (key.objectid == ref_path->owner_objectid &&
3793                              key.type > BTRFS_EXTENT_DATA_KEY) ||
3794                             (key.offset >= first_pos + extent_key->offset))
3795                                 break;
3796                 }
3797
3798                 if (inode && key.objectid != inode->i_ino) {
3799                         BUG_ON(extent_locked);
3800                         btrfs_release_path(root, path);
3801                         mutex_unlock(&inode->i_mutex);
3802                         iput(inode);
3803                         inode = NULL;
3804                         continue;
3805                 }
3806
3807                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
3808                         path->slots[0]++;
3809                         ret = 1;
3810                         goto next;
3811                 }
3812                 fi = btrfs_item_ptr(leaf, path->slots[0],
3813                                     struct btrfs_file_extent_item);
3814                 extent_type = btrfs_file_extent_type(leaf, fi);
3815                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
3816                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
3817                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3818                      extent_key->objectid)) {
3819                         path->slots[0]++;
3820                         ret = 1;
3821                         goto next;
3822                 }
3823
3824                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3825                 ext_offset = btrfs_file_extent_offset(leaf, fi);
3826
3827                 if (first_pos > key.offset - ext_offset)
3828                         first_pos = key.offset - ext_offset;
3829
3830                 if (!extent_locked) {
3831                         lock_start = key.offset;
3832                         lock_end = lock_start + num_bytes - 1;
3833                 } else {
3834                         if (lock_start > key.offset ||
3835                             lock_end + 1 < key.offset + num_bytes) {
3836                                 unlock_extent(&BTRFS_I(inode)->io_tree,
3837                                               lock_start, lock_end, GFP_NOFS);
3838                                 extent_locked = 0;
3839                         }
3840                 }
3841
3842                 if (!inode) {
3843                         btrfs_release_path(root, path);
3844
3845                         inode = btrfs_iget_locked(root->fs_info->sb,
3846                                                   key.objectid, root);
3847                         if (inode->i_state & I_NEW) {
3848                                 BTRFS_I(inode)->root = root;
3849                                 BTRFS_I(inode)->location.objectid =
3850                                         key.objectid;
3851                                 BTRFS_I(inode)->location.type =
3852                                         BTRFS_INODE_ITEM_KEY;
3853                                 BTRFS_I(inode)->location.offset = 0;
3854                                 btrfs_read_locked_inode(inode);
3855                                 unlock_new_inode(inode);
3856                         }
3857                         /*
3858                          * some code call btrfs_commit_transaction while
3859                          * holding the i_mutex, so we can't use mutex_lock
3860                          * here.
3861                          */
3862                         if (is_bad_inode(inode) ||
3863                             !mutex_trylock(&inode->i_mutex)) {
3864                                 iput(inode);
3865                                 inode = NULL;
3866                                 key.offset = (u64)-1;
3867                                 goto skip;
3868                         }
3869                 }
3870
3871                 if (!extent_locked) {
3872                         struct btrfs_ordered_extent *ordered;
3873
3874                         btrfs_release_path(root, path);
3875
3876                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3877                                     lock_end, GFP_NOFS);
3878                         ordered = btrfs_lookup_first_ordered_extent(inode,
3879                                                                     lock_end);
3880                         if (ordered &&
3881                             ordered->file_offset <= lock_end &&
3882                             ordered->file_offset + ordered->len > lock_start) {
3883                                 unlock_extent(&BTRFS_I(inode)->io_tree,
3884                                               lock_start, lock_end, GFP_NOFS);
3885                                 btrfs_start_ordered_extent(inode, ordered, 1);
3886                                 btrfs_put_ordered_extent(ordered);
3887                                 key.offset += num_bytes;
3888                                 goto skip;
3889                         }
3890                         if (ordered)
3891                                 btrfs_put_ordered_extent(ordered);
3892
3893                         extent_locked = 1;
3894                         continue;
3895                 }
3896
3897                 if (nr_extents == 1) {
3898                         /* update extent pointer in place */
3899                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
3900                                                 new_extents[0].disk_bytenr);
3901                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3902                                                 new_extents[0].disk_num_bytes);
3903                         btrfs_mark_buffer_dirty(leaf);
3904
3905                         btrfs_drop_extent_cache(inode, key.offset,
3906                                                 key.offset + num_bytes - 1, 0);
3907
3908                         ret = btrfs_inc_extent_ref(trans, root,
3909                                                 new_extents[0].disk_bytenr,
3910                                                 new_extents[0].disk_num_bytes,
3911                                                 leaf->start,
3912                                                 root->root_key.objectid,
3913                                                 trans->transid,
3914                                                 key.objectid);
3915                         BUG_ON(ret);
3916
3917                         ret = btrfs_free_extent(trans, root,
3918                                                 extent_key->objectid,
3919                                                 extent_key->offset,
3920                                                 leaf->start,
3921                                                 btrfs_header_owner(leaf),
3922                                                 btrfs_header_generation(leaf),
3923                                                 key.objectid, 0);
3924                         BUG_ON(ret);
3925
3926                         btrfs_release_path(root, path);
3927                         key.offset += num_bytes;
3928                 } else {
3929                         BUG_ON(1);
3930 #if 0
3931                         u64 alloc_hint;
3932                         u64 extent_len;
3933                         int i;
3934                         /*
3935                          * drop old extent pointer at first, then insert the
3936                          * new pointers one bye one
3937                          */
3938                         btrfs_release_path(root, path);
3939                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
3940                                                  key.offset + num_bytes,
3941                                                  key.offset, &alloc_hint);
3942                         BUG_ON(ret);
3943
3944                         for (i = 0; i < nr_extents; i++) {
3945                                 if (ext_offset >= new_extents[i].num_bytes) {
3946                                         ext_offset -= new_extents[i].num_bytes;
3947                                         continue;
3948                                 }
3949                                 extent_len = min(new_extents[i].num_bytes -
3950                                                  ext_offset, num_bytes);
3951
3952                                 ret = btrfs_insert_empty_item(trans, root,
3953                                                               path, &key,
3954                                                               sizeof(*fi));
3955                                 BUG_ON(ret);
3956
3957                                 leaf = path->nodes[0];
3958                                 fi = btrfs_item_ptr(leaf, path->slots[0],
3959                                                 struct btrfs_file_extent_item);
3960                                 btrfs_set_file_extent_generation(leaf, fi,
3961                                                         trans->transid);
3962                                 btrfs_set_file_extent_type(leaf, fi,
3963                                                         BTRFS_FILE_EXTENT_REG);
3964                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3965                                                 new_extents[i].disk_bytenr);
3966                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3967                                                 new_extents[i].disk_num_bytes);
3968                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
3969                                                 new_extents[i].ram_bytes);
3970
3971                                 btrfs_set_file_extent_compression(leaf, fi,
3972                                                 new_extents[i].compression);
3973                                 btrfs_set_file_extent_encryption(leaf, fi,
3974                                                 new_extents[i].encryption);
3975                                 btrfs_set_file_extent_other_encoding(leaf, fi,
3976                                                 new_extents[i].other_encoding);
3977
3978                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3979                                                         extent_len);
3980                                 ext_offset += new_extents[i].offset;
3981                                 btrfs_set_file_extent_offset(leaf, fi,
3982                                                         ext_offset);
3983                                 btrfs_mark_buffer_dirty(leaf);
3984
3985                                 btrfs_drop_extent_cache(inode, key.offset,
3986                                                 key.offset + extent_len - 1, 0);
3987
3988                                 ret = btrfs_inc_extent_ref(trans, root,
3989                                                 new_extents[i].disk_bytenr,
3990                                                 new_extents[i].disk_num_bytes,
3991                                                 leaf->start,
3992                                                 root->root_key.objectid,
3993                                                 trans->transid, key.objectid);
3994                                 BUG_ON(ret);
3995                                 btrfs_release_path(root, path);
3996
3997                                 inode_add_bytes(inode, extent_len);
3998
3999                                 ext_offset = 0;
4000                                 num_bytes -= extent_len;
4001                                 key.offset += extent_len;
4002
4003                                 if (num_bytes == 0)
4004                                         break;
4005                         }
4006                         BUG_ON(i >= nr_extents);
4007 #endif
4008                 }
4009
4010                 if (extent_locked) {
4011                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4012                                       lock_end, GFP_NOFS);
4013                         extent_locked = 0;
4014                 }
4015 skip:
4016                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4017                     key.offset >= first_pos + extent_key->offset)
4018                         break;
4019
4020                 cond_resched();
4021         }
4022         ret = 0;
4023 out:
4024         btrfs_release_path(root, path);
4025         if (inode) {
4026                 mutex_unlock(&inode->i_mutex);
4027                 if (extent_locked) {
4028                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4029                                       lock_end, GFP_NOFS);
4030                 }
4031                 iput(inode);
4032         }
4033         return ret;
4034 }
4035
4036 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4037                                struct btrfs_root *root,
4038                                struct extent_buffer *buf, u64 orig_start)
4039 {
4040         int level;
4041         int ret;
4042
4043         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4044         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4045
4046         level = btrfs_header_level(buf);
4047         if (level == 0) {
4048                 struct btrfs_leaf_ref *ref;
4049                 struct btrfs_leaf_ref *orig_ref;
4050
4051                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4052                 if (!orig_ref)
4053                         return -ENOENT;
4054
4055                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4056                 if (!ref) {
4057                         btrfs_free_leaf_ref(root, orig_ref);
4058                         return -ENOMEM;
4059                 }
4060
4061                 ref->nritems = orig_ref->nritems;
4062                 memcpy(ref->extents, orig_ref->extents,
4063                         sizeof(ref->extents[0]) * ref->nritems);
4064
4065                 btrfs_free_leaf_ref(root, orig_ref);
4066
4067                 ref->root_gen = trans->transid;
4068                 ref->bytenr = buf->start;
4069                 ref->owner = btrfs_header_owner(buf);
4070                 ref->generation = btrfs_header_generation(buf);
4071                 ret = btrfs_add_leaf_ref(root, ref, 0);
4072                 WARN_ON(ret);
4073                 btrfs_free_leaf_ref(root, ref);
4074         }
4075         return 0;
4076 }
4077
4078 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4079                                         struct extent_buffer *leaf,
4080                                         struct btrfs_block_group_cache *group,
4081                                         struct btrfs_root *target_root)
4082 {
4083         struct btrfs_key key;
4084         struct inode *inode = NULL;
4085         struct btrfs_file_extent_item *fi;
4086         u64 num_bytes;
4087         u64 skip_objectid = 0;
4088         u32 nritems;
4089         u32 i;
4090
4091         nritems = btrfs_header_nritems(leaf);
4092         for (i = 0; i < nritems; i++) {
4093                 btrfs_item_key_to_cpu(leaf, &key, i);
4094                 if (key.objectid == skip_objectid ||
4095                     key.type != BTRFS_EXTENT_DATA_KEY)
4096                         continue;
4097                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4098                 if (btrfs_file_extent_type(leaf, fi) ==
4099                     BTRFS_FILE_EXTENT_INLINE)
4100                         continue;
4101                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4102                         continue;
4103                 if (!inode || inode->i_ino != key.objectid) {
4104                         iput(inode);
4105                         inode = btrfs_ilookup(target_root->fs_info->sb,
4106                                               key.objectid, target_root, 1);
4107                 }
4108                 if (!inode) {
4109                         skip_objectid = key.objectid;
4110                         continue;
4111                 }
4112                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4113
4114                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4115                             key.offset + num_bytes - 1, GFP_NOFS);
4116                 btrfs_drop_extent_cache(inode, key.offset,
4117                                         key.offset + num_bytes - 1, 1);
4118                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4119                               key.offset + num_bytes - 1, GFP_NOFS);
4120                 cond_resched();
4121         }
4122         iput(inode);
4123         return 0;
4124 }
4125
4126 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4127                                         struct btrfs_root *root,
4128                                         struct extent_buffer *leaf,
4129                                         struct btrfs_block_group_cache *group,
4130                                         struct inode *reloc_inode)
4131 {
4132         struct btrfs_key key;
4133         struct btrfs_key extent_key;
4134         struct btrfs_file_extent_item *fi;
4135         struct btrfs_leaf_ref *ref;
4136         struct disk_extent *new_extent;
4137         u64 bytenr;
4138         u64 num_bytes;
4139         u32 nritems;
4140         u32 i;
4141         int ext_index;
4142         int nr_extent;
4143         int ret;
4144
4145         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4146         BUG_ON(!new_extent);
4147
4148         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4149         BUG_ON(!ref);
4150
4151         ext_index = -1;
4152         nritems = btrfs_header_nritems(leaf);
4153         for (i = 0; i < nritems; i++) {
4154                 btrfs_item_key_to_cpu(leaf, &key, i);
4155                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4156                         continue;
4157                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4158                 if (btrfs_file_extent_type(leaf, fi) ==
4159                     BTRFS_FILE_EXTENT_INLINE)
4160                         continue;
4161                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4162                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4163                 if (bytenr == 0)
4164                         continue;
4165
4166                 ext_index++;
4167                 if (bytenr >= group->key.objectid + group->key.offset ||
4168                     bytenr + num_bytes <= group->key.objectid)
4169                         continue;
4170
4171                 extent_key.objectid = bytenr;
4172                 extent_key.offset = num_bytes;
4173                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4174                 nr_extent = 1;
4175                 ret = get_new_locations(reloc_inode, &extent_key,
4176                                         group->key.objectid, 1,
4177                                         &new_extent, &nr_extent);
4178                 if (ret > 0)
4179                         continue;
4180                 BUG_ON(ret < 0);
4181
4182                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4183                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4184                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4185                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4186
4187                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4188                                                 new_extent->disk_bytenr);
4189                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4190                                                 new_extent->disk_num_bytes);
4191                 btrfs_mark_buffer_dirty(leaf);
4192
4193                 ret = btrfs_inc_extent_ref(trans, root,
4194                                         new_extent->disk_bytenr,
4195                                         new_extent->disk_num_bytes,
4196                                         leaf->start,
4197                                         root->root_key.objectid,
4198                                         trans->transid, key.objectid);
4199                 BUG_ON(ret);
4200                 ret = btrfs_free_extent(trans, root,
4201                                         bytenr, num_bytes, leaf->start,
4202                                         btrfs_header_owner(leaf),
4203                                         btrfs_header_generation(leaf),
4204                                         key.objectid, 0);
4205                 BUG_ON(ret);
4206                 cond_resched();
4207         }
4208         kfree(new_extent);
4209         BUG_ON(ext_index + 1 != ref->nritems);
4210         btrfs_free_leaf_ref(root, ref);
4211         return 0;
4212 }
4213
4214 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4215                           struct btrfs_root *root)
4216 {
4217         struct btrfs_root *reloc_root;
4218         int ret;
4219
4220         if (root->reloc_root) {
4221                 reloc_root = root->reloc_root;
4222                 root->reloc_root = NULL;
4223                 list_add(&reloc_root->dead_list,
4224                          &root->fs_info->dead_reloc_roots);
4225
4226                 btrfs_set_root_bytenr(&reloc_root->root_item,
4227                                       reloc_root->node->start);
4228                 btrfs_set_root_level(&root->root_item,
4229                                      btrfs_header_level(reloc_root->node));
4230                 memset(&reloc_root->root_item.drop_progress, 0,
4231                         sizeof(struct btrfs_disk_key));
4232                 reloc_root->root_item.drop_level = 0;
4233
4234                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4235                                         &reloc_root->root_key,
4236                                         &reloc_root->root_item);
4237                 BUG_ON(ret);
4238         }
4239         return 0;
4240 }
4241
4242 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4243 {
4244         struct btrfs_trans_handle *trans;
4245         struct btrfs_root *reloc_root;
4246         struct btrfs_root *prev_root = NULL;
4247         struct list_head dead_roots;
4248         int ret;
4249         unsigned long nr;
4250
4251         INIT_LIST_HEAD(&dead_roots);
4252         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4253
4254         while (!list_empty(&dead_roots)) {
4255                 reloc_root = list_entry(dead_roots.prev,
4256                                         struct btrfs_root, dead_list);
4257                 list_del_init(&reloc_root->dead_list);
4258
4259                 BUG_ON(reloc_root->commit_root != NULL);
4260                 while (1) {
4261                         trans = btrfs_join_transaction(root, 1);
4262                         BUG_ON(!trans);
4263
4264                         mutex_lock(&root->fs_info->drop_mutex);
4265                         ret = btrfs_drop_snapshot(trans, reloc_root);
4266                         if (ret != -EAGAIN)
4267                                 break;
4268                         mutex_unlock(&root->fs_info->drop_mutex);
4269
4270                         nr = trans->blocks_used;
4271                         ret = btrfs_end_transaction(trans, root);
4272                         BUG_ON(ret);
4273                         btrfs_btree_balance_dirty(root, nr);
4274                 }
4275
4276                 free_extent_buffer(reloc_root->node);
4277
4278                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4279                                      &reloc_root->root_key);
4280                 BUG_ON(ret);
4281                 mutex_unlock(&root->fs_info->drop_mutex);
4282
4283                 nr = trans->blocks_used;
4284                 ret = btrfs_end_transaction(trans, root);
4285                 BUG_ON(ret);
4286                 btrfs_btree_balance_dirty(root, nr);
4287
4288                 kfree(prev_root);
4289                 prev_root = reloc_root;
4290         }
4291         if (prev_root) {
4292                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4293                 kfree(prev_root);
4294         }
4295         return 0;
4296 }
4297
4298 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4299 {
4300         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4301         return 0;
4302 }
4303
4304 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4305 {
4306         struct btrfs_root *reloc_root;
4307         struct btrfs_trans_handle *trans;
4308         struct btrfs_key location;
4309         int found;
4310         int ret;
4311
4312         mutex_lock(&root->fs_info->tree_reloc_mutex);
4313         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4314         BUG_ON(ret);
4315         found = !list_empty(&root->fs_info->dead_reloc_roots);
4316         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4317
4318         if (found) {
4319                 trans = btrfs_start_transaction(root, 1);
4320                 BUG_ON(!trans);
4321                 ret = btrfs_commit_transaction(trans, root);
4322                 BUG_ON(ret);
4323         }
4324
4325         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4326         location.offset = (u64)-1;
4327         location.type = BTRFS_ROOT_ITEM_KEY;
4328
4329         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4330         BUG_ON(!reloc_root);
4331         btrfs_orphan_cleanup(reloc_root);
4332         return 0;
4333 }
4334
4335 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
4336                                     struct btrfs_root *root)
4337 {
4338         struct btrfs_root *reloc_root;
4339         struct extent_buffer *eb;
4340         struct btrfs_root_item *root_item;
4341         struct btrfs_key root_key;
4342         int ret;
4343
4344         BUG_ON(!root->ref_cows);
4345         if (root->reloc_root)
4346                 return 0;
4347
4348         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4349         BUG_ON(!root_item);
4350
4351         ret = btrfs_copy_root(trans, root, root->commit_root,
4352                               &eb, BTRFS_TREE_RELOC_OBJECTID);
4353         BUG_ON(ret);
4354
4355         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4356         root_key.offset = root->root_key.objectid;
4357         root_key.type = BTRFS_ROOT_ITEM_KEY;
4358
4359         memcpy(root_item, &root->root_item, sizeof(root_item));
4360         btrfs_set_root_refs(root_item, 0);
4361         btrfs_set_root_bytenr(root_item, eb->start);
4362         btrfs_set_root_level(root_item, btrfs_header_level(eb));
4363         btrfs_set_root_generation(root_item, trans->transid);
4364
4365         btrfs_tree_unlock(eb);
4366         free_extent_buffer(eb);
4367
4368         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4369                                 &root_key, root_item);
4370         BUG_ON(ret);
4371         kfree(root_item);
4372
4373         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4374                                                  &root_key);
4375         BUG_ON(!reloc_root);
4376         reloc_root->last_trans = trans->transid;
4377         reloc_root->commit_root = NULL;
4378         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4379
4380         root->reloc_root = reloc_root;
4381         return 0;
4382 }
4383
4384 /*
4385  * Core function of space balance.
4386  *
4387  * The idea is using reloc trees to relocate tree blocks in reference
4388  * counted roots. There is one reloc tree for each subvol, and all
4389  * reloc trees share same root key objectid. Reloc trees are snapshots
4390  * of the latest committed roots of subvols (root->commit_root).
4391  *
4392  * To relocate a tree block referenced by a subvol, there are two steps.
4393  * COW the block through subvol's reloc tree, then update block pointer
4394  * in the subvol to point to the new block. Since all reloc trees share
4395  * same root key objectid, doing special handing for tree blocks owned
4396  * by them is easy. Once a tree block has been COWed in one reloc tree,
4397  * we can use the resulting new block directly when the same block is
4398  * required to COW again through other reloc trees. By this way, relocated
4399  * tree blocks are shared between reloc trees, so they are also shared
4400  * between subvols.
4401  */
4402 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
4403                                       struct btrfs_root *root,
4404                                       struct btrfs_path *path,
4405                                       struct btrfs_key *first_key,
4406                                       struct btrfs_ref_path *ref_path,
4407                                       struct btrfs_block_group_cache *group,
4408                                       struct inode *reloc_inode)
4409 {
4410         struct btrfs_root *reloc_root;
4411         struct extent_buffer *eb = NULL;
4412         struct btrfs_key *keys;
4413         u64 *nodes;
4414         int level;
4415         int shared_level;
4416         int lowest_level = 0;
4417         int ret;
4418
4419         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
4420                 lowest_level = ref_path->owner_objectid;
4421
4422         if (!root->ref_cows) {
4423                 path->lowest_level = lowest_level;
4424                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
4425                 BUG_ON(ret < 0);
4426                 path->lowest_level = 0;
4427                 btrfs_release_path(root, path);
4428                 return 0;
4429         }
4430
4431         mutex_lock(&root->fs_info->tree_reloc_mutex);
4432         ret = init_reloc_tree(trans, root);
4433         BUG_ON(ret);
4434         reloc_root = root->reloc_root;
4435
4436         shared_level = ref_path->shared_level;
4437         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
4438
4439         keys = ref_path->node_keys;
4440         nodes = ref_path->new_nodes;
4441         memset(&keys[shared_level + 1], 0,
4442                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
4443         memset(&nodes[shared_level + 1], 0,
4444                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
4445
4446         if (nodes[lowest_level] == 0) {
4447                 path->lowest_level = lowest_level;
4448                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4449                                         0, 1);
4450                 BUG_ON(ret);
4451                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
4452                         eb = path->nodes[level];
4453                         if (!eb || eb == reloc_root->node)
4454                                 break;
4455                         nodes[level] = eb->start;
4456                         if (level == 0)
4457                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4458                         else
4459                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4460                 }
4461                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4462                         eb = path->nodes[0];
4463                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
4464                                                       group, reloc_inode);
4465                         BUG_ON(ret);
4466                 }
4467                 btrfs_release_path(reloc_root, path);
4468         } else {
4469                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
4470                                        lowest_level);
4471                 BUG_ON(ret);
4472         }
4473
4474         /*
4475          * replace tree blocks in the fs tree with tree blocks in
4476          * the reloc tree.
4477          */
4478         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
4479         BUG_ON(ret < 0);
4480
4481         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4482                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4483                                         0, 0);
4484                 BUG_ON(ret);
4485                 extent_buffer_get(path->nodes[0]);
4486                 eb = path->nodes[0];
4487                 btrfs_release_path(reloc_root, path);
4488                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
4489                 BUG_ON(ret);
4490                 free_extent_buffer(eb);
4491         }
4492
4493         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4494         path->lowest_level = 0;
4495         return 0;
4496 }
4497
4498 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
4499                                         struct btrfs_root *root,
4500                                         struct btrfs_path *path,
4501                                         struct btrfs_key *first_key,
4502                                         struct btrfs_ref_path *ref_path)
4503 {
4504         int ret;
4505
4506         ret = relocate_one_path(trans, root, path, first_key,
4507                                 ref_path, NULL, NULL);
4508         BUG_ON(ret);
4509
4510         if (root == root->fs_info->extent_root)
4511                 btrfs_extent_post_op(trans, root);
4512
4513         return 0;
4514 }
4515
4516 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
4517                                     struct btrfs_root *extent_root,
4518                                     struct btrfs_path *path,
4519                                     struct btrfs_key *extent_key)
4520 {
4521         int ret;
4522
4523         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
4524         if (ret)
4525                 goto out;
4526         ret = btrfs_del_item(trans, extent_root, path);
4527 out:
4528         btrfs_release_path(extent_root, path);
4529         return ret;
4530 }
4531
4532 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
4533                                                 struct btrfs_ref_path *ref_path)
4534 {
4535         struct btrfs_key root_key;
4536
4537         root_key.objectid = ref_path->root_objectid;
4538         root_key.type = BTRFS_ROOT_ITEM_KEY;
4539         if (is_cowonly_root(ref_path->root_objectid))
4540                 root_key.offset = 0;
4541         else
4542                 root_key.offset = (u64)-1;
4543
4544         return btrfs_read_fs_root_no_name(fs_info, &root_key);
4545 }
4546
4547 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
4548                                         struct btrfs_path *path,
4549                                         struct btrfs_key *extent_key,
4550                                         struct btrfs_block_group_cache *group,
4551                                         struct inode *reloc_inode, int pass)
4552 {
4553         struct btrfs_trans_handle *trans;
4554         struct btrfs_root *found_root;
4555         struct btrfs_ref_path *ref_path = NULL;
4556         struct disk_extent *new_extents = NULL;
4557         int nr_extents = 0;
4558         int loops;
4559         int ret;
4560         int level;
4561         struct btrfs_key first_key;
4562         u64 prev_block = 0;
4563
4564
4565         trans = btrfs_start_transaction(extent_root, 1);
4566         BUG_ON(!trans);
4567
4568         if (extent_key->objectid == 0) {
4569                 ret = del_extent_zero(trans, extent_root, path, extent_key);
4570                 goto out;
4571         }
4572
4573         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
4574         if (!ref_path) {
4575                ret = -ENOMEM;
4576                goto out;
4577         }
4578
4579         for (loops = 0; ; loops++) {
4580                 if (loops == 0) {
4581                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
4582                                                    extent_key->objectid);
4583                 } else {
4584                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
4585                 }
4586                 if (ret < 0)
4587                         goto out;
4588                 if (ret > 0)
4589                         break;
4590
4591                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4592                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
4593                         continue;
4594
4595                 found_root = read_ref_root(extent_root->fs_info, ref_path);
4596                 BUG_ON(!found_root);
4597                 /*
4598                  * for reference counted tree, only process reference paths
4599                  * rooted at the latest committed root.
4600                  */
4601                 if (found_root->ref_cows &&
4602                     ref_path->root_generation != found_root->root_key.offset)
4603                         continue;
4604
4605                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4606                         if (pass == 0) {
4607                                 /*
4608                                  * copy data extents to new locations
4609                                  */
4610                                 u64 group_start = group->key.objectid;
4611                                 ret = relocate_data_extent(reloc_inode,
4612                                                            extent_key,
4613                                                            group_start);
4614                                 if (ret < 0)
4615                                         goto out;
4616                                 break;
4617                         }
4618                         level = 0;
4619                 } else {
4620                         level = ref_path->owner_objectid;
4621                 }
4622
4623                 if (prev_block != ref_path->nodes[level]) {
4624                         struct extent_buffer *eb;
4625                         u64 block_start = ref_path->nodes[level];
4626                         u64 block_size = btrfs_level_size(found_root, level);
4627
4628                         eb = read_tree_block(found_root, block_start,
4629                                              block_size, 0);
4630                         btrfs_tree_lock(eb);
4631                         BUG_ON(level != btrfs_header_level(eb));
4632
4633                         if (level == 0)
4634                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
4635                         else
4636                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
4637
4638                         btrfs_tree_unlock(eb);
4639                         free_extent_buffer(eb);
4640                         prev_block = block_start;
4641                 }
4642
4643                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4644                     pass >= 2) {
4645                         /*
4646                          * use fallback method to process the remaining
4647                          * references.
4648                          */
4649                         if (!new_extents) {
4650                                 u64 group_start = group->key.objectid;
4651                                 new_extents = kmalloc(sizeof(*new_extents),
4652                                                       GFP_NOFS);
4653                                 nr_extents = 1;
4654                                 ret = get_new_locations(reloc_inode,
4655                                                         extent_key,
4656                                                         group_start, 1,
4657                                                         &new_extents,
4658                                                         &nr_extents);
4659                                 if (ret)
4660                                         goto out;
4661                         }
4662                         btrfs_record_root_in_trans(found_root);
4663                         ret = replace_one_extent(trans, found_root,
4664                                                 path, extent_key,
4665                                                 &first_key, ref_path,
4666                                                 new_extents, nr_extents);
4667                         if (ret < 0)
4668                                 goto out;
4669                         continue;
4670                 }
4671
4672                 btrfs_record_root_in_trans(found_root);
4673                 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4674                         ret = relocate_tree_block(trans, found_root, path,
4675                                                   &first_key, ref_path);
4676                 } else {
4677                         /*
4678                          * try to update data extent references while
4679                          * keeping metadata shared between snapshots.
4680                          */
4681                         ret = relocate_one_path(trans, found_root, path,
4682                                                 &first_key, ref_path,
4683                                                 group, reloc_inode);
4684                 }
4685                 if (ret < 0)
4686                         goto out;
4687         }
4688         ret = 0;
4689 out:
4690         btrfs_end_transaction(trans, extent_root);
4691         kfree(new_extents);
4692         kfree(ref_path);
4693         return ret;
4694 }
4695
4696 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
4697 {
4698         u64 num_devices;
4699         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
4700                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
4701
4702         num_devices = root->fs_info->fs_devices->num_devices;
4703         if (num_devices == 1) {
4704                 stripped |= BTRFS_BLOCK_GROUP_DUP;
4705                 stripped = flags & ~stripped;
4706
4707                 /* turn raid0 into single device chunks */
4708                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
4709                         return stripped;
4710
4711                 /* turn mirroring into duplication */
4712                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
4713                              BTRFS_BLOCK_GROUP_RAID10))
4714                         return stripped | BTRFS_BLOCK_GROUP_DUP;
4715                 return flags;
4716         } else {
4717                 /* they already had raid on here, just return */
4718                 if (flags & stripped)
4719                         return flags;
4720
4721                 stripped |= BTRFS_BLOCK_GROUP_DUP;
4722                 stripped = flags & ~stripped;
4723
4724                 /* switch duplicated blocks with raid1 */
4725                 if (flags & BTRFS_BLOCK_GROUP_DUP)
4726                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
4727
4728                 /* turn single device chunks into raid0 */
4729                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
4730         }
4731         return flags;
4732 }
4733
4734 int __alloc_chunk_for_shrink(struct btrfs_root *root,
4735                      struct btrfs_block_group_cache *shrink_block_group,
4736                      int force)
4737 {
4738         struct btrfs_trans_handle *trans;
4739         u64 new_alloc_flags;
4740         u64 calc;
4741
4742         spin_lock(&shrink_block_group->lock);
4743         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
4744                 spin_unlock(&shrink_block_group->lock);
4745
4746                 trans = btrfs_start_transaction(root, 1);
4747                 spin_lock(&shrink_block_group->lock);
4748
4749                 new_alloc_flags = update_block_group_flags(root,
4750                                                    shrink_block_group->flags);
4751                 if (new_alloc_flags != shrink_block_group->flags) {
4752                         calc =
4753                              btrfs_block_group_used(&shrink_block_group->item);
4754                 } else {
4755                         calc = shrink_block_group->key.offset;
4756                 }
4757                 spin_unlock(&shrink_block_group->lock);
4758
4759                 do_chunk_alloc(trans, root->fs_info->extent_root,
4760                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
4761
4762                 btrfs_end_transaction(trans, root);
4763         } else
4764                 spin_unlock(&shrink_block_group->lock);
4765         return 0;
4766 }
4767
4768 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4769                                  struct btrfs_root *root,
4770                                  u64 objectid, u64 size)
4771 {
4772         struct btrfs_path *path;
4773         struct btrfs_inode_item *item;
4774         struct extent_buffer *leaf;
4775         int ret;
4776
4777         path = btrfs_alloc_path();
4778         if (!path)
4779                 return -ENOMEM;
4780
4781         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4782         if (ret)
4783                 goto out;
4784
4785         leaf = path->nodes[0];
4786         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4787         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4788         btrfs_set_inode_generation(leaf, item, 1);
4789         btrfs_set_inode_size(leaf, item, size);
4790         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4791         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM |
4792                                           BTRFS_INODE_NOCOMPRESS);
4793         btrfs_mark_buffer_dirty(leaf);
4794         btrfs_release_path(root, path);
4795 out:
4796         btrfs_free_path(path);
4797         return ret;
4798 }
4799
4800 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
4801                                         struct btrfs_block_group_cache *group)
4802 {
4803         struct inode *inode = NULL;
4804         struct btrfs_trans_handle *trans;
4805         struct btrfs_root *root;
4806         struct btrfs_key root_key;
4807         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
4808         int err = 0;
4809
4810         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4811         root_key.type = BTRFS_ROOT_ITEM_KEY;
4812         root_key.offset = (u64)-1;
4813         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
4814         if (IS_ERR(root))
4815                 return ERR_CAST(root);
4816
4817         trans = btrfs_start_transaction(root, 1);
4818         BUG_ON(!trans);
4819
4820         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
4821         if (err)
4822                 goto out;
4823
4824         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
4825         BUG_ON(err);
4826
4827         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
4828                                        group->key.offset, 0, group->key.offset,
4829                                        0, 0, 0);
4830         BUG_ON(err);
4831
4832         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
4833         if (inode->i_state & I_NEW) {
4834                 BTRFS_I(inode)->root = root;
4835                 BTRFS_I(inode)->location.objectid = objectid;
4836                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
4837                 BTRFS_I(inode)->location.offset = 0;
4838                 btrfs_read_locked_inode(inode);
4839                 unlock_new_inode(inode);
4840                 BUG_ON(is_bad_inode(inode));
4841         } else {
4842                 BUG_ON(1);
4843         }
4844
4845         err = btrfs_orphan_add(trans, inode);
4846 out:
4847         btrfs_end_transaction(trans, root);
4848         if (err) {
4849                 if (inode)
4850                         iput(inode);
4851                 inode = ERR_PTR(err);
4852         }
4853         return inode;
4854 }
4855
4856 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
4857 {
4858         struct btrfs_trans_handle *trans;
4859         struct btrfs_path *path;
4860         struct btrfs_fs_info *info = root->fs_info;
4861         struct extent_buffer *leaf;
4862         struct inode *reloc_inode;
4863         struct btrfs_block_group_cache *block_group;
4864         struct btrfs_key key;
4865         u64 skipped;
4866         u64 cur_byte;
4867         u64 total_found;
4868         u32 nritems;
4869         int ret;
4870         int progress;
4871         int pass = 0;
4872
4873         root = root->fs_info->extent_root;
4874
4875         block_group = btrfs_lookup_block_group(info, group_start);
4876         BUG_ON(!block_group);
4877
4878         printk("btrfs relocating block group %llu flags %llu\n",
4879                (unsigned long long)block_group->key.objectid,
4880                (unsigned long long)block_group->flags);
4881
4882         path = btrfs_alloc_path();
4883         BUG_ON(!path);
4884
4885         reloc_inode = create_reloc_inode(info, block_group);
4886         BUG_ON(IS_ERR(reloc_inode));
4887
4888         __alloc_chunk_for_shrink(root, block_group, 1);
4889         block_group->ro = 1;
4890         block_group->space_info->total_bytes -= block_group->key.offset;
4891
4892         btrfs_start_delalloc_inodes(info->tree_root);
4893         btrfs_wait_ordered_extents(info->tree_root, 0);
4894 again:
4895         skipped = 0;
4896         total_found = 0;
4897         progress = 0;
4898         key.objectid = block_group->key.objectid;
4899         key.offset = 0;
4900         key.type = 0;
4901         cur_byte = key.objectid;
4902
4903         trans = btrfs_start_transaction(info->tree_root, 1);
4904         btrfs_commit_transaction(trans, info->tree_root);
4905
4906         mutex_lock(&root->fs_info->cleaner_mutex);
4907         btrfs_clean_old_snapshots(info->tree_root);
4908         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
4909         mutex_unlock(&root->fs_info->cleaner_mutex);
4910
4911         while(1) {
4912                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4913                 if (ret < 0)
4914                         goto out;
4915 next:
4916                 leaf = path->nodes[0];
4917                 nritems = btrfs_header_nritems(leaf);
4918                 if (path->slots[0] >= nritems) {
4919                         ret = btrfs_next_leaf(root, path);
4920                         if (ret < 0)
4921                                 goto out;
4922                         if (ret == 1) {
4923                                 ret = 0;
4924                                 break;
4925                         }
4926                         leaf = path->nodes[0];
4927                         nritems = btrfs_header_nritems(leaf);
4928                 }
4929
4930                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4931
4932                 if (key.objectid >= block_group->key.objectid +
4933                     block_group->key.offset)
4934                         break;
4935
4936                 if (progress && need_resched()) {
4937                         btrfs_release_path(root, path);
4938                         cond_resched();
4939                         progress = 0;
4940                         continue;
4941                 }
4942                 progress = 1;
4943
4944                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
4945                     key.objectid + key.offset <= cur_byte) {
4946                         path->slots[0]++;
4947                         goto next;
4948                 }
4949
4950                 total_found++;
4951                 cur_byte = key.objectid + key.offset;
4952                 btrfs_release_path(root, path);
4953
4954                 __alloc_chunk_for_shrink(root, block_group, 0);
4955                 ret = relocate_one_extent(root, path, &key, block_group,
4956                                           reloc_inode, pass);
4957                 BUG_ON(ret < 0);
4958                 if (ret > 0)
4959                         skipped++;
4960
4961                 key.objectid = cur_byte;
4962                 key.type = 0;
4963                 key.offset = 0;
4964         }
4965
4966         btrfs_release_path(root, path);
4967
4968         if (pass == 0) {
4969                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
4970                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
4971                 WARN_ON(reloc_inode->i_mapping->nrpages);
4972         }
4973
4974         if (total_found > 0) {
4975                 printk("btrfs found %llu extents in pass %d\n",
4976                        (unsigned long long)total_found, pass);
4977                 pass++;
4978                 if (total_found == skipped && pass > 2) {
4979                         iput(reloc_inode);
4980                         reloc_inode = create_reloc_inode(info, block_group);
4981                         pass = 0;
4982                 }
4983                 goto again;
4984         }
4985
4986         /* delete reloc_inode */
4987         iput(reloc_inode);
4988
4989         /* unpin extents in this range */
4990         trans = btrfs_start_transaction(info->tree_root, 1);
4991         btrfs_commit_transaction(trans, info->tree_root);
4992
4993         spin_lock(&block_group->lock);
4994         WARN_ON(block_group->pinned > 0);
4995         WARN_ON(block_group->reserved > 0);
4996         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
4997         spin_unlock(&block_group->lock);
4998         ret = 0;
4999 out:
5000         btrfs_free_path(path);
5001         return ret;
5002 }
5003
5004 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5005                            struct btrfs_key *key)
5006 {
5007         int ret = 0;
5008         struct btrfs_key found_key;
5009         struct extent_buffer *leaf;
5010         int slot;
5011
5012         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5013         if (ret < 0)
5014                 goto out;
5015
5016         while(1) {
5017                 slot = path->slots[0];
5018                 leaf = path->nodes[0];
5019                 if (slot >= btrfs_header_nritems(leaf)) {
5020                         ret = btrfs_next_leaf(root, path);
5021                         if (ret == 0)
5022                                 continue;
5023                         if (ret < 0)
5024                                 goto out;
5025                         break;
5026                 }
5027                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5028
5029                 if (found_key.objectid >= key->objectid &&
5030                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5031                         ret = 0;
5032                         goto out;
5033                 }
5034                 path->slots[0]++;
5035         }
5036         ret = -ENOENT;
5037 out:
5038         return ret;
5039 }
5040
5041 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5042 {
5043         struct btrfs_block_group_cache *block_group;
5044         struct rb_node *n;
5045
5046         spin_lock(&info->block_group_cache_lock);
5047         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5048                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5049                                        cache_node);
5050                 rb_erase(&block_group->cache_node,
5051                          &info->block_group_cache_tree);
5052                 spin_unlock(&info->block_group_cache_lock);
5053
5054                 btrfs_remove_free_space_cache(block_group);
5055                 down_write(&block_group->space_info->groups_sem);
5056                 list_del(&block_group->list);
5057                 up_write(&block_group->space_info->groups_sem);
5058                 kfree(block_group);
5059
5060                 spin_lock(&info->block_group_cache_lock);
5061         }
5062         spin_unlock(&info->block_group_cache_lock);
5063         return 0;
5064 }
5065
5066 int btrfs_read_block_groups(struct btrfs_root *root)
5067 {
5068         struct btrfs_path *path;
5069         int ret;
5070         struct btrfs_block_group_cache *cache;
5071         struct btrfs_fs_info *info = root->fs_info;
5072         struct btrfs_space_info *space_info;
5073         struct btrfs_key key;
5074         struct btrfs_key found_key;
5075         struct extent_buffer *leaf;
5076
5077         root = info->extent_root;
5078         key.objectid = 0;
5079         key.offset = 0;
5080         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5081         path = btrfs_alloc_path();
5082         if (!path)
5083                 return -ENOMEM;
5084
5085         while(1) {
5086                 ret = find_first_block_group(root, path, &key);
5087                 if (ret > 0) {
5088                         ret = 0;
5089                         goto error;
5090                 }
5091                 if (ret != 0)
5092                         goto error;
5093
5094                 leaf = path->nodes[0];
5095                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5096                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5097                 if (!cache) {
5098                         ret = -ENOMEM;
5099                         break;
5100                 }
5101
5102                 spin_lock_init(&cache->lock);
5103                 mutex_init(&cache->alloc_mutex);
5104                 INIT_LIST_HEAD(&cache->list);
5105                 read_extent_buffer(leaf, &cache->item,
5106                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5107                                    sizeof(cache->item));
5108                 memcpy(&cache->key, &found_key, sizeof(found_key));
5109
5110                 key.objectid = found_key.objectid + found_key.offset;
5111                 btrfs_release_path(root, path);
5112                 cache->flags = btrfs_block_group_flags(&cache->item);
5113
5114                 ret = update_space_info(info, cache->flags, found_key.offset,
5115                                         btrfs_block_group_used(&cache->item),
5116                                         &space_info);
5117                 BUG_ON(ret);
5118                 cache->space_info = space_info;
5119                 down_write(&space_info->groups_sem);
5120                 list_add_tail(&cache->list, &space_info->block_groups);
5121                 up_write(&space_info->groups_sem);
5122
5123                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5124                 BUG_ON(ret);
5125
5126                 set_avail_alloc_bits(root->fs_info, cache->flags);
5127         }
5128         ret = 0;
5129 error:
5130         btrfs_free_path(path);
5131         return ret;
5132 }
5133
5134 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5135                            struct btrfs_root *root, u64 bytes_used,
5136                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5137                            u64 size)
5138 {
5139         int ret;
5140         struct btrfs_root *extent_root;
5141         struct btrfs_block_group_cache *cache;
5142
5143         extent_root = root->fs_info->extent_root;
5144
5145         root->fs_info->last_trans_new_blockgroup = trans->transid;
5146
5147         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5148         if (!cache)
5149                 return -ENOMEM;
5150
5151         cache->key.objectid = chunk_offset;
5152         cache->key.offset = size;
5153         spin_lock_init(&cache->lock);
5154         mutex_init(&cache->alloc_mutex);
5155         INIT_LIST_HEAD(&cache->list);
5156         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5157
5158         btrfs_set_block_group_used(&cache->item, bytes_used);
5159         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5160         cache->flags = type;
5161         btrfs_set_block_group_flags(&cache->item, type);
5162
5163         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5164                                 &cache->space_info);
5165         BUG_ON(ret);
5166         down_write(&cache->space_info->groups_sem);
5167         list_add_tail(&cache->list, &cache->space_info->block_groups);
5168         up_write(&cache->space_info->groups_sem);
5169
5170         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5171         BUG_ON(ret);
5172
5173         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5174                                 sizeof(cache->item));
5175         BUG_ON(ret);
5176
5177         finish_current_insert(trans, extent_root, 0);
5178         ret = del_pending_extents(trans, extent_root, 0);
5179         BUG_ON(ret);
5180         set_avail_alloc_bits(extent_root->fs_info, type);
5181
5182         return 0;
5183 }
5184
5185 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5186                              struct btrfs_root *root, u64 group_start)
5187 {
5188         struct btrfs_path *path;
5189         struct btrfs_block_group_cache *block_group;
5190         struct btrfs_key key;
5191         int ret;
5192
5193         root = root->fs_info->extent_root;
5194
5195         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5196         BUG_ON(!block_group);
5197
5198         memcpy(&key, &block_group->key, sizeof(key));
5199
5200         path = btrfs_alloc_path();
5201         BUG_ON(!path);
5202
5203         btrfs_remove_free_space_cache(block_group);
5204         rb_erase(&block_group->cache_node,
5205                  &root->fs_info->block_group_cache_tree);
5206         down_write(&block_group->space_info->groups_sem);
5207         list_del(&block_group->list);
5208         up_write(&block_group->space_info->groups_sem);
5209
5210         /*
5211         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5212         kfree(shrink_block_group);
5213         */
5214
5215         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5216         if (ret > 0)
5217                 ret = -EIO;
5218         if (ret < 0)
5219                 goto out;
5220
5221         ret = btrfs_del_item(trans, root, path);
5222 out:
5223         btrfs_free_path(path);
5224         return ret;
5225 }