git.oblomov.eu Git - linux-2.6/blob - fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include "compat.h"
  25 #include "hash.h"
  26 #include "crc32c.h"
  27 #include "ctree.h"
  28 #include "disk-io.h"
  29 #include "print-tree.h"
  30 #include "transaction.h"
  31 #include "volumes.h"
  32 #include "locking.h"
  33 #include "ref-cache.h"
  34
  35 #define PENDING_EXTENT_INSERT 0
  36 #define PENDING_EXTENT_DELETE 1
  37 #define PENDING_BACKREF_UPDATE 2
  38
  39 struct pending_extent_op {
  40         int type;
  41         u64 bytenr;
  42         u64 num_bytes;
  43         u64 parent;
  44         u64 orig_parent;
  45         u64 generation;
  46         u64 orig_generation;
  47         int level;
  48         struct list_head list;
  49         int del;
  50 };
  51
  52 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
  53                                          struct btrfs_root *root, u64 parent,
  54                                          u64 root_objectid, u64 ref_generation,
  55                                          u64 owner, struct btrfs_key *ins,
  56                                          int ref_mod);
  57 static int update_reserved_extents(struct btrfs_root *root,
  58                                    u64 bytenr, u64 num, int reserve);
  59 static int pin_down_bytes(struct btrfs_trans_handle *trans,
  60                           struct btrfs_root *root,
  61                           u64 bytenr, u64 num_bytes, int is_data);
  62 static int update_block_group(struct btrfs_trans_handle *trans,
  63                               struct btrfs_root *root,
  64                               u64 bytenr, u64 num_bytes, int alloc,
  65                               int mark_free);
  66 static noinline int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  67                                         struct btrfs_root *root,
  68                                         u64 bytenr, u64 num_bytes, u64 parent,
  69                                         u64 root_objectid, u64 ref_generation,
  70                                         u64 owner_objectid, int pin,
  71                                         int ref_to_drop);
  72
  73 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  74                           struct btrfs_root *extent_root, u64 alloc_bytes,
  75                           u64 flags, int force);
  76
  77 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  78 {
  79         return (cache->flags & bits) == bits;
  80 }
  81
  82 /*
  83  * this adds the block group to the fs_info rb tree for the block group
  84  * cache
  85  */
  86 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  87                                 struct btrfs_block_group_cache *block_group)
  88 {
  89         struct rb_node **p;
  90         struct rb_node *parent = NULL;
  91         struct btrfs_block_group_cache *cache;
  92
  93         spin_lock(&info->block_group_cache_lock);
  94         p = &info->block_group_cache_tree.rb_node;
  95
  96         while (*p) {
  97                 parent = *p;
  98                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  99                                  cache_node);
 100                 if (block_group->key.objectid < cache->key.objectid) {
 101                         p = &(*p)->rb_left;
 102                 } else if (block_group->key.objectid > cache->key.objectid) {
 103                         p = &(*p)->rb_right;
 104                 } else {
 105                         spin_unlock(&info->block_group_cache_lock);
 106                         return -EEXIST;
 107                 }
 108         }
 109
 110         rb_link_node(&block_group->cache_node, parent, p);
 111         rb_insert_color(&block_group->cache_node,
 112                         &info->block_group_cache_tree);
 113         spin_unlock(&info->block_group_cache_lock);
 114
 115         return 0;
 116 }
 117
 118 /*
 119  * This will return the block group at or after bytenr if contains is 0, else
 120  * it will return the block group that contains the bytenr
 121  */
 122 static struct btrfs_block_group_cache *
 123 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 124                               int contains)
 125 {
 126         struct btrfs_block_group_cache *cache, *ret = NULL;
 127         struct rb_node *n;
 128         u64 end, start;
 129
 130         spin_lock(&info->block_group_cache_lock);
 131         n = info->block_group_cache_tree.rb_node;
 132
 133         while (n) {
 134                 cache = rb_entry(n, struct btrfs_block_group_cache,
 135                                  cache_node);
 136                 end = cache->key.objectid + cache->key.offset - 1;
 137                 start = cache->key.objectid;
 138
 139                 if (bytenr < start) {
 140                         if (!contains && (!ret || start < ret->key.objectid))
 141                                 ret = cache;
 142                         n = n->rb_left;
 143                 } else if (bytenr > start) {
 144                         if (contains && bytenr <= end) {
 145                                 ret = cache;
 146                                 break;
 147                         }
 148                         n = n->rb_right;
 149                 } else {
 150                         ret = cache;
 151                         break;
 152                 }
 153         }
 154         if (ret)
 155                 atomic_inc(&ret->count);
 156         spin_unlock(&info->block_group_cache_lock);
 157
 158         return ret;
 159 }
 160
 161 /*
 162  * this is only called by cache_block_group, since we could have freed extents
 163  * we need to check the pinned_extents for any extents that can't be used yet
 164  * since their free space will be released as soon as the transaction commits.
 165  */
 166 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 167                               struct btrfs_fs_info *info, u64 start, u64 end)
 168 {
 169         u64 extent_start, extent_end, size;
 170         int ret;
 171
 172         mutex_lock(&info->pinned_mutex);
 173         while (start < end) {
 174                 ret = find_first_extent_bit(&info->pinned_extents, start,
 175                                             &extent_start, &extent_end,
 176                                             EXTENT_DIRTY);
 177                 if (ret)
 178                         break;
 179
 180                 if (extent_start == start) {
 181                         start = extent_end + 1;
 182                 } else if (extent_start > start && extent_start < end) {
 183                         size = extent_start - start;
 184                         ret = btrfs_add_free_space(block_group, start,
 185                                                    size);
 186                         BUG_ON(ret);
 187                         start = extent_end + 1;
 188                 } else {
 189                         break;
 190                 }
 191         }
 192
 193         if (start < end) {
 194                 size = end - start;
 195                 ret = btrfs_add_free_space(block_group, start, size);
 196                 BUG_ON(ret);
 197         }
 198         mutex_unlock(&info->pinned_mutex);
 199
 200         return 0;
 201 }
 202
 203 static int remove_sb_from_cache(struct btrfs_root *root,
 204                                 struct btrfs_block_group_cache *cache)
 205 {
 206         u64 bytenr;
 207         u64 *logical;
 208         int stripe_len;
 209         int i, nr, ret;
 210
 211         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 212                 bytenr = btrfs_sb_offset(i);
 213                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 214                                        cache->key.objectid, bytenr, 0,
 215                                        &logical, &nr, &stripe_len);
 216                 BUG_ON(ret);
 217                 while (nr--) {
 218                         btrfs_remove_free_space(cache, logical[nr],
 219                                                 stripe_len);
 220                 }
 221                 kfree(logical);
 222         }
 223         return 0;
 224 }
 225
 226 static int cache_block_group(struct btrfs_root *root,
 227                              struct btrfs_block_group_cache *block_group)
 228 {
 229         struct btrfs_path *path;
 230         int ret = 0;
 231         struct btrfs_key key;
 232         struct extent_buffer *leaf;
 233         int slot;
 234         u64 last;
 235
 236         if (!block_group)
 237                 return 0;
 238
 239         root = root->fs_info->extent_root;
 240
 241         if (block_group->cached)
 242                 return 0;
 243
 244         path = btrfs_alloc_path();
 245         if (!path)
 246                 return -ENOMEM;
 247
 248         path->reada = 2;
 249         /*
 250          * we get into deadlocks with paths held by callers of this function.
 251          * since the alloc_mutex is protecting things right now, just
 252          * skip the locking here
 253          */
 254         path->skip_locking = 1;
 255         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 256         key.objectid = last;
 257         key.offset = 0;
 258         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 259         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 260         if (ret < 0)
 261                 goto err;
 262
 263         while (1) {
 264                 leaf = path->nodes[0];
 265                 slot = path->slots[0];
 266                 if (slot >= btrfs_header_nritems(leaf)) {
 267                         ret = btrfs_next_leaf(root, path);
 268                         if (ret < 0)
 269                                 goto err;
 270                         if (ret == 0)
 271                                 continue;
 272                         else
 273                                 break;
 274                 }
 275                 btrfs_item_key_to_cpu(leaf, &key, slot);
 276                 if (key.objectid < block_group->key.objectid)
 277                         goto next;
 278
 279                 if (key.objectid >= block_group->key.objectid +
 280                     block_group->key.offset)
 281                         break;
 282
 283                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 284                         add_new_free_space(block_group, root->fs_info, last,
 285                                            key.objectid);
 286
 287                         last = key.objectid + key.offset;
 288                 }
 289 next:
 290                 path->slots[0]++;
 291         }
 292
 293         add_new_free_space(block_group, root->fs_info, last,
 294                            block_group->key.objectid +
 295                            block_group->key.offset);
 296
 297         remove_sb_from_cache(root, block_group);
 298         block_group->cached = 1;
 299         ret = 0;
 300 err:
 301         btrfs_free_path(path);
 302         return ret;
 303 }
 304
 305 /*
 306  * return the block group that starts at or after bytenr
 307  */
 308 static struct btrfs_block_group_cache *
 309 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 310 {
 311         struct btrfs_block_group_cache *cache;
 312
 313         cache = block_group_cache_tree_search(info, bytenr, 0);
 314
 315         return cache;
 316 }
 317
 318 /*
 319  * return the block group that contains teh given bytenr
 320  */
 321 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 322                                                  struct btrfs_fs_info *info,
 323                                                  u64 bytenr)
 324 {
 325         struct btrfs_block_group_cache *cache;
 326
 327         cache = block_group_cache_tree_search(info, bytenr, 1);
 328
 329         return cache;
 330 }
 331
 332 static inline void put_block_group(struct btrfs_block_group_cache *cache)
 333 {
 334         if (atomic_dec_and_test(&cache->count))
 335                 kfree(cache);
 336 }
 337
 338 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 339                                                   u64 flags)
 340 {
 341         struct list_head *head = &info->space_info;
 342         struct btrfs_space_info *found;
 343
 344         rcu_read_lock();
 345         list_for_each_entry_rcu(found, head, list) {
 346                 if (found->flags == flags) {
 347                         rcu_read_unlock();
 348                         return found;
 349                 }
 350         }
 351         rcu_read_unlock();
 352         return NULL;
 353 }
 354
 355 /*
 356  * after adding space to the filesystem, we need to clear the full flags
 357  * on all the space infos.
 358  */
 359 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 360 {
 361         struct list_head *head = &info->space_info;
 362         struct btrfs_space_info *found;
 363
 364         rcu_read_lock();
 365         list_for_each_entry_rcu(found, head, list)
 366                 found->full = 0;
 367         rcu_read_unlock();
 368 }
 369
 370 static u64 div_factor(u64 num, int factor)
 371 {
 372         if (factor == 10)
 373                 return num;
 374         num *= factor;
 375         do_div(num, 10);
 376         return num;
 377 }
 378
 379 u64 btrfs_find_block_group(struct btrfs_root *root,
 380                            u64 search_start, u64 search_hint, int owner)
 381 {
 382         struct btrfs_block_group_cache *cache;
 383         u64 used;
 384         u64 last = max(search_hint, search_start);
 385         u64 group_start = 0;
 386         int full_search = 0;
 387         int factor = 9;
 388         int wrapped = 0;
 389 again:
 390         while (1) {
 391                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 392                 if (!cache)
 393                         break;
 394
 395                 spin_lock(&cache->lock);
 396                 last = cache->key.objectid + cache->key.offset;
 397                 used = btrfs_block_group_used(&cache->item);
 398
 399                 if ((full_search || !cache->ro) &&
 400                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 401                         if (used + cache->pinned + cache->reserved <
 402                             div_factor(cache->key.offset, factor)) {
 403                                 group_start = cache->key.objectid;
 404                                 spin_unlock(&cache->lock);
 405                                 put_block_group(cache);
 406                                 goto found;
 407                         }
 408                 }
 409                 spin_unlock(&cache->lock);
 410                 put_block_group(cache);
 411                 cond_resched();
 412         }
 413         if (!wrapped) {
 414                 last = search_start;
 415                 wrapped = 1;
 416                 goto again;
 417         }
 418         if (!full_search && factor < 10) {
 419                 last = search_start;
 420                 full_search = 1;
 421                 factor = 10;
 422                 goto again;
 423         }
 424 found:
 425         return group_start;
 426 }
 427
 428 /* simple helper to search for an existing extent at a given offset */
 429 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 430 {
 431         int ret;
 432         struct btrfs_key key;
 433         struct btrfs_path *path;
 434
 435         path = btrfs_alloc_path();
 436         BUG_ON(!path);
 437         key.objectid = start;
 438         key.offset = len;
 439         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 440         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 441                                 0, 0);
 442         btrfs_free_path(path);
 443         return ret;
 444 }
 445
 446 /*
 447  * Back reference rules.  Back refs have three main goals:
 448  *
 449  * 1) differentiate between all holders of references to an extent so that
 450  *    when a reference is dropped we can make sure it was a valid reference
 451  *    before freeing the extent.
 452  *
 453  * 2) Provide enough information to quickly find the holders of an extent
 454  *    if we notice a given block is corrupted or bad.
 455  *
 456  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 457  *    maintenance.  This is actually the same as #2, but with a slightly
 458  *    different use case.
 459  *
 460  * File extents can be referenced by:
 461  *
 462  * - multiple snapshots, subvolumes, or different generations in one subvol
 463  * - different files inside a single subvolume
 464  * - different offsets inside a file (bookend extents in file.c)
 465  *
 466  * The extent ref structure has fields for:
 467  *
 468  * - Objectid of the subvolume root
 469  * - Generation number of the tree holding the reference
 470  * - objectid of the file holding the reference
 471  * - number of references holding by parent node (alway 1 for tree blocks)
 472  *
 473  * Btree leaf may hold multiple references to a file extent. In most cases,
 474  * these references are from same file and the corresponding offsets inside
 475  * the file are close together.
 476  *
 477  * When a file extent is allocated the fields are filled in:
 478  *     (root_key.objectid, trans->transid, inode objectid, 1)
 479  *
 480  * When a leaf is cow'd new references are added for every file extent found
 481  * in the leaf.  It looks similar to the create case, but trans->transid will
 482  * be different when the block is cow'd.
 483  *
 484  *     (root_key.objectid, trans->transid, inode objectid,
 485  *      number of references in the leaf)
 486  *
 487  * When a file extent is removed either during snapshot deletion or
 488  * file truncation, we find the corresponding back reference and check
 489  * the following fields:
 490  *
 491  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 492  *      inode objectid)
 493  *
 494  * Btree extents can be referenced by:
 495  *
 496  * - Different subvolumes
 497  * - Different generations of the same subvolume
 498  *
 499  * When a tree block is created, back references are inserted:
 500  *
 501  * (root->root_key.objectid, trans->transid, level, 1)
 502  *
 503  * When a tree block is cow'd, new back references are added for all the
 504  * blocks it points to. If the tree block isn't in reference counted root,
 505  * the old back references are removed. These new back references are of
 506  * the form (trans->transid will have increased since creation):
 507  *
 508  * (root->root_key.objectid, trans->transid, level, 1)
 509  *
 510  * When a backref is in deleting, the following fields are checked:
 511  *
 512  * if backref was for a tree root:
 513  *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
 514  * else
 515  *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
 516  *
 517  * Back Reference Key composing:
 518  *
 519  * The key objectid corresponds to the first byte in the extent, the key
 520  * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
 521  * byte of parent extent. If a extent is tree root, the key offset is set
 522  * to the key objectid.
 523  */
 524
 525 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
 526                                           struct btrfs_root *root,
 527                                           struct btrfs_path *path,
 528                                           u64 bytenr, u64 parent,
 529                                           u64 ref_root, u64 ref_generation,
 530                                           u64 owner_objectid, int del)
 531 {
 532         struct btrfs_key key;
 533         struct btrfs_extent_ref *ref;
 534         struct extent_buffer *leaf;
 535         u64 ref_objectid;
 536         int ret;
 537
 538         key.objectid = bytenr;
 539         key.type = BTRFS_EXTENT_REF_KEY;
 540         key.offset = parent;
 541
 542         ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
 543         if (ret < 0)
 544                 goto out;
 545         if (ret > 0) {
 546                 ret = -ENOENT;
 547                 goto out;
 548         }
 549
 550         leaf = path->nodes[0];
 551         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 552         ref_objectid = btrfs_ref_objectid(leaf, ref);
 553         if (btrfs_ref_root(leaf, ref) != ref_root ||
 554             btrfs_ref_generation(leaf, ref) != ref_generation ||
 555             (ref_objectid != owner_objectid &&
 556              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 557                 ret = -EIO;
 558                 WARN_ON(1);
 559                 goto out;
 560         }
 561         ret = 0;
 562 out:
 563         return ret;
 564 }
 565
 566 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
 567                                           struct btrfs_root *root,
 568                                           struct btrfs_path *path,
 569                                           u64 bytenr, u64 parent,
 570                                           u64 ref_root, u64 ref_generation,
 571                                           u64 owner_objectid,
 572                                           int refs_to_add)
 573 {
 574         struct btrfs_key key;
 575         struct extent_buffer *leaf;
 576         struct btrfs_extent_ref *ref;
 577         u32 num_refs;
 578         int ret;
 579
 580         key.objectid = bytenr;
 581         key.type = BTRFS_EXTENT_REF_KEY;
 582         key.offset = parent;
 583
 584         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
 585         if (ret == 0) {
 586                 leaf = path->nodes[0];
 587                 ref = btrfs_item_ptr(leaf, path->slots[0],
 588                                      struct btrfs_extent_ref);
 589                 btrfs_set_ref_root(leaf, ref, ref_root);
 590                 btrfs_set_ref_generation(leaf, ref, ref_generation);
 591                 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
 592                 btrfs_set_ref_num_refs(leaf, ref, refs_to_add);
 593         } else if (ret == -EEXIST) {
 594                 u64 existing_owner;
 595
 596                 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
 597                 leaf = path->nodes[0];
 598                 ref = btrfs_item_ptr(leaf, path->slots[0],
 599                                      struct btrfs_extent_ref);
 600                 if (btrfs_ref_root(leaf, ref) != ref_root ||
 601                     btrfs_ref_generation(leaf, ref) != ref_generation) {
 602                         ret = -EIO;
 603                         WARN_ON(1);
 604                         goto out;
 605                 }
 606
 607                 num_refs = btrfs_ref_num_refs(leaf, ref);
 608                 BUG_ON(num_refs == 0);
 609                 btrfs_set_ref_num_refs(leaf, ref, num_refs + refs_to_add);
 610
 611                 existing_owner = btrfs_ref_objectid(leaf, ref);
 612                 if (existing_owner != owner_objectid &&
 613                     existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
 614                         btrfs_set_ref_objectid(leaf, ref,
 615                                         BTRFS_MULTIPLE_OBJECTIDS);
 616                 }
 617                 ret = 0;
 618         } else {
 619                 goto out;
 620         }
 621         btrfs_mark_buffer_dirty(path->nodes[0]);
 622 out:
 623         btrfs_release_path(root, path);
 624         return ret;
 625 }
 626
 627 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
 628                                           struct btrfs_root *root,
 629                                           struct btrfs_path *path,
 630                                           int refs_to_drop)
 631 {
 632         struct extent_buffer *leaf;
 633         struct btrfs_extent_ref *ref;
 634         u32 num_refs;
 635         int ret = 0;
 636
 637         leaf = path->nodes[0];
 638         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 639         num_refs = btrfs_ref_num_refs(leaf, ref);
 640         BUG_ON(num_refs < refs_to_drop);
 641         num_refs -= refs_to_drop;
 642         if (num_refs == 0) {
 643                 ret = btrfs_del_item(trans, root, path);
 644         } else {
 645                 btrfs_set_ref_num_refs(leaf, ref, num_refs);
 646                 btrfs_mark_buffer_dirty(leaf);
 647         }
 648         btrfs_release_path(root, path);
 649         return ret;
 650 }
 651
 652 #ifdef BIO_RW_DISCARD
 653 static void btrfs_issue_discard(struct block_device *bdev,
 654                                 u64 start, u64 len)
 655 {
 656         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
 657 }
 658 #endif
 659
 660 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
 661                                 u64 num_bytes)
 662 {
 663 #ifdef BIO_RW_DISCARD
 664         int ret;
 665         u64 map_length = num_bytes;
 666         struct btrfs_multi_bio *multi = NULL;
 667
 668         /* Tell the block device(s) that the sectors can be discarded */
 669         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
 670                               bytenr, &map_length, &multi, 0);
 671         if (!ret) {
 672                 struct btrfs_bio_stripe *stripe = multi->stripes;
 673                 int i;
 674
 675                 if (map_length > num_bytes)
 676                         map_length = num_bytes;
 677
 678                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
 679                         btrfs_issue_discard(stripe->dev->bdev,
 680                                             stripe->physical,
 681                                             map_length);
 682                 }
 683                 kfree(multi);
 684         }
 685
 686         return ret;
 687 #else
 688         return 0;
 689 #endif
 690 }
 691
 692 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
 693                                      struct btrfs_root *root, u64 bytenr,
 694                                      u64 num_bytes,
 695                                      u64 orig_parent, u64 parent,
 696                                      u64 orig_root, u64 ref_root,
 697                                      u64 orig_generation, u64 ref_generation,
 698                                      u64 owner_objectid)
 699 {
 700         int ret;
 701         int pin = owner_objectid < BTRFS_FIRST_FREE_OBJECTID;
 702
 703         ret = btrfs_update_delayed_ref(trans, bytenr, num_bytes,
 704                                        orig_parent, parent, orig_root,
 705                                        ref_root, orig_generation,
 706                                        ref_generation, owner_objectid, pin);
 707         BUG_ON(ret);
 708         return ret;
 709 }
 710
 711 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
 712                             struct btrfs_root *root, u64 bytenr,
 713                             u64 num_bytes, u64 orig_parent, u64 parent,
 714                             u64 ref_root, u64 ref_generation,
 715                             u64 owner_objectid)
 716 {
 717         int ret;
 718         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
 719             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
 720                 return 0;
 721
 722         ret = __btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
 723                                         orig_parent, parent, ref_root,
 724                                         ref_root, ref_generation,
 725                                         ref_generation, owner_objectid);
 726         return ret;
 727 }
 728 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 729                                   struct btrfs_root *root, u64 bytenr,
 730                                   u64 num_bytes,
 731                                   u64 orig_parent, u64 parent,
 732                                   u64 orig_root, u64 ref_root,
 733                                   u64 orig_generation, u64 ref_generation,
 734                                   u64 owner_objectid)
 735 {
 736         int ret;
 737
 738         ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, ref_root,
 739                                     ref_generation, owner_objectid,
 740                                     BTRFS_ADD_DELAYED_REF, 0);
 741         BUG_ON(ret);
 742         return ret;
 743 }
 744
 745 static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans,
 746                           struct btrfs_root *root, u64 bytenr,
 747                           u64 num_bytes, u64 parent, u64 ref_root,
 748                           u64 ref_generation, u64 owner_objectid,
 749                           int refs_to_add)
 750 {
 751         struct btrfs_path *path;
 752         int ret;
 753         struct btrfs_key key;
 754         struct extent_buffer *l;
 755         struct btrfs_extent_item *item;
 756         u32 refs;
 757
 758         path = btrfs_alloc_path();
 759         if (!path)
 760                 return -ENOMEM;
 761
 762         path->reada = 1;
 763         key.objectid = bytenr;
 764         key.type = BTRFS_EXTENT_ITEM_KEY;
 765         key.offset = num_bytes;
 766
 767         /* first find the extent item and update its reference count */
 768         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
 769                                 path, 0, 1);
 770         if (ret < 0)
 771                 return ret;
 772
 773         if (ret > 0) {
 774                 WARN_ON(1);
 775                 btrfs_free_path(path);
 776                 return -EIO;
 777         }
 778         l = path->nodes[0];
 779
 780         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 781         if (key.objectid != bytenr) {
 782                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
 783                 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
 784                        (unsigned long long)bytenr,
 785                        (unsigned long long)key.objectid);
 786                 BUG();
 787         }
 788         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
 789
 790         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 791
 792         refs = btrfs_extent_refs(l, item);
 793         btrfs_set_extent_refs(l, item, refs + refs_to_add);
 794         btrfs_mark_buffer_dirty(path->nodes[0]);
 795
 796         btrfs_release_path(root->fs_info->extent_root, path);
 797
 798         path->reada = 1;
 799         /* now insert the actual backref */
 800         ret = insert_extent_backref(trans, root->fs_info->extent_root,
 801                                     path, bytenr, parent,
 802                                     ref_root, ref_generation,
 803                                     owner_objectid, refs_to_add);
 804         BUG_ON(ret);
 805         btrfs_free_path(path);
 806         return 0;
 807 }
 808
 809 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 810                          struct btrfs_root *root,
 811                          u64 bytenr, u64 num_bytes, u64 parent,
 812                          u64 ref_root, u64 ref_generation,
 813                          u64 owner_objectid)
 814 {
 815         int ret;
 816         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
 817             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
 818                 return 0;
 819
 820         ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, parent,
 821                                      0, ref_root, 0, ref_generation,
 822                                      owner_objectid);
 823         return ret;
 824 }
 825
 826 static int drop_delayed_ref(struct btrfs_trans_handle *trans,
 827                                         struct btrfs_root *root,
 828                                         struct btrfs_delayed_ref_node *node)
 829 {
 830         int ret = 0;
 831         struct btrfs_delayed_ref *ref = btrfs_delayed_node_to_ref(node);
 832
 833         BUG_ON(node->ref_mod == 0);
 834         ret = __btrfs_free_extent(trans, root, node->bytenr, node->num_bytes,
 835                                   node->parent, ref->root, ref->generation,
 836                                   ref->owner_objectid, ref->pin, node->ref_mod);
 837
 838         return ret;
 839 }
 840
 841 /* helper function to actually process a single delayed ref entry */
 842 static noinline int run_one_delayed_ref(struct btrfs_trans_handle *trans,
 843                                         struct btrfs_root *root,
 844                                         struct btrfs_delayed_ref_node *node,
 845                                         int insert_reserved)
 846 {
 847         int ret;
 848         struct btrfs_delayed_ref *ref;
 849
 850         if (node->parent == (u64)-1) {
 851                 struct btrfs_delayed_ref_head *head;
 852                 /*
 853                  * we've hit the end of the chain and we were supposed
 854                  * to insert this extent into the tree.  But, it got
 855                  * deleted before we ever needed to insert it, so all
 856                  * we have to do is clean up the accounting
 857                  */
 858                 if (insert_reserved) {
 859                         update_reserved_extents(root, node->bytenr,
 860                                                 node->num_bytes, 0);
 861                 }
 862                 head = btrfs_delayed_node_to_head(node);
 863                 mutex_unlock(&head->mutex);
 864                 return 0;
 865         }
 866
 867         ref = btrfs_delayed_node_to_ref(node);
 868         if (ref->action == BTRFS_ADD_DELAYED_REF) {
 869                 if (insert_reserved) {
 870                         struct btrfs_key ins;
 871
 872                         ins.objectid = node->bytenr;
 873                         ins.offset = node->num_bytes;
 874                         ins.type = BTRFS_EXTENT_ITEM_KEY;
 875
 876                         /* record the full extent allocation */
 877                         ret = __btrfs_alloc_reserved_extent(trans, root,
 878                                         node->parent, ref->root,
 879                                         ref->generation, ref->owner_objectid,
 880                                         &ins, node->ref_mod);
 881                         update_reserved_extents(root, node->bytenr,
 882                                                 node->num_bytes, 0);
 883                 } else {
 884                         /* just add one backref */
 885                         ret = add_extent_ref(trans, root, node->bytenr,
 886                                      node->num_bytes,
 887                                      node->parent, ref->root, ref->generation,
 888                                      ref->owner_objectid, node->ref_mod);
 889                 }
 890                 BUG_ON(ret);
 891         } else if (ref->action == BTRFS_DROP_DELAYED_REF) {
 892                 WARN_ON(insert_reserved);
 893                 ret = drop_delayed_ref(trans, root, node);
 894         }
 895         return 0;
 896 }
 897
 898 static noinline struct btrfs_delayed_ref_node *
 899 select_delayed_ref(struct btrfs_delayed_ref_head *head)
 900 {
 901         struct rb_node *node;
 902         struct btrfs_delayed_ref_node *ref;
 903         int action = BTRFS_ADD_DELAYED_REF;
 904 again:
 905         /*
 906          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
 907          * this prevents ref count from going down to zero when
 908          * there still are pending delayed ref.
 909          */
 910         node = rb_prev(&head->node.rb_node);
 911         while (1) {
 912                 if (!node)
 913                         break;
 914                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
 915                                 rb_node);
 916                 if (ref->bytenr != head->node.bytenr)
 917                         break;
 918                 if (btrfs_delayed_node_to_ref(ref)->action == action)
 919                         return ref;
 920                 node = rb_prev(node);
 921         }
 922         if (action == BTRFS_ADD_DELAYED_REF) {
 923                 action = BTRFS_DROP_DELAYED_REF;
 924                 goto again;
 925         }
 926         return NULL;
 927 }
 928
 929 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 930                                        struct btrfs_root *root,
 931                                        struct list_head *cluster)
 932 {
 933         struct btrfs_delayed_ref_root *delayed_refs;
 934         struct btrfs_delayed_ref_node *ref;
 935         struct btrfs_delayed_ref_head *locked_ref = NULL;
 936         int ret;
 937         int count = 0;
 938         int must_insert_reserved = 0;
 939
 940         delayed_refs = &trans->transaction->delayed_refs;
 941         while (1) {
 942                 if (!locked_ref) {
 943                         /* pick a new head ref from the cluster list */
 944                         if (list_empty(cluster))
 945                                 break;
 946
 947                         locked_ref = list_entry(cluster->next,
 948                                      struct btrfs_delayed_ref_head, cluster);
 949
 950                         /* grab the lock that says we are going to process
 951                          * all the refs for this head */
 952                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
 953
 954                         /*
 955                          * we may have dropped the spin lock to get the head
 956                          * mutex lock, and that might have given someone else
 957                          * time to free the head.  If that's true, it has been
 958                          * removed from our list and we can move on.
 959                          */
 960                         if (ret == -EAGAIN) {
 961                                 locked_ref = NULL;
 962                                 count++;
 963                                 continue;
 964                         }
 965                 }
 966
 967                 /*
 968                  * record the must insert reserved flag before we
 969                  * drop the spin lock.
 970                  */
 971                 must_insert_reserved = locked_ref->must_insert_reserved;
 972                 locked_ref->must_insert_reserved = 0;
 973
 974                 /*
 975                  * locked_ref is the head node, so we have to go one
 976                  * node back for any delayed ref updates
 977                  */
 978                 ref = select_delayed_ref(locked_ref);
 979                 if (!ref) {
 980                         /* All delayed refs have been processed, Go ahead
 981                          * and send the head node to run_one_delayed_ref,
 982                          * so that any accounting fixes can happen
 983                          */
 984                         ref = &locked_ref->node;
 985                         list_del_init(&locked_ref->cluster);
 986                         locked_ref = NULL;
 987                 }
 988
 989                 ref->in_tree = 0;
 990                 rb_erase(&ref->rb_node, &delayed_refs->root);
 991                 delayed_refs->num_entries--;
 992                 spin_unlock(&delayed_refs->lock);
 993
 994                 ret = run_one_delayed_ref(trans, root, ref,
 995                                           must_insert_reserved);
 996                 BUG_ON(ret);
 997                 btrfs_put_delayed_ref(ref);
 998
 999                 count++;
1000                 cond_resched();
1001                 spin_lock(&delayed_refs->lock);
1002         }
1003         return count;
1004 }
1005
1006 /*
1007  * this starts processing the delayed reference count updates and
1008  * extent insertions we have queued up so far.  count can be
1009  * 0, which means to process everything in the tree at the start
1010  * of the run (but not newly added entries), or it can be some target
1011  * number you'd like to process.
1012  */
1013 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1014                            struct btrfs_root *root, unsigned long count)
1015 {
1016         struct rb_node *node;
1017         struct btrfs_delayed_ref_root *delayed_refs;
1018         struct btrfs_delayed_ref_node *ref;
1019         struct list_head cluster;
1020         int ret;
1021         int run_all = count == (unsigned long)-1;
1022         int run_most = 0;
1023
1024         if (root == root->fs_info->extent_root)
1025                 root = root->fs_info->tree_root;
1026
1027         delayed_refs = &trans->transaction->delayed_refs;
1028         INIT_LIST_HEAD(&cluster);
1029 again:
1030         spin_lock(&delayed_refs->lock);
1031         if (count == 0) {
1032                 count = delayed_refs->num_entries * 2;
1033                 run_most = 1;
1034         }
1035         while (1) {
1036                 if (!(run_all || run_most) &&
1037                     delayed_refs->num_heads_ready < 64)
1038                         break;
1039
1040                 /*
1041                  * go find something we can process in the rbtree.  We start at
1042                  * the beginning of the tree, and then build a cluster
1043                  * of refs to process starting at the first one we are able to
1044                  * lock
1045                  */
1046                 ret = btrfs_find_ref_cluster(trans, &cluster,
1047                                              delayed_refs->run_delayed_start);
1048                 if (ret)
1049                         break;
1050
1051                 ret = run_clustered_refs(trans, root, &cluster);
1052                 BUG_ON(ret < 0);
1053
1054                 count -= min_t(unsigned long, ret, count);
1055
1056                 if (count == 0)
1057                         break;
1058         }
1059
1060         if (run_all) {
1061                 node = rb_first(&delayed_refs->root);
1062                 if (!node)
1063                         goto out;
1064                 count = (unsigned long)-1;
1065
1066                 while (node) {
1067                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
1068                                        rb_node);
1069                         if (btrfs_delayed_ref_is_head(ref)) {
1070                                 struct btrfs_delayed_ref_head *head;
1071
1072                                 head = btrfs_delayed_node_to_head(ref);
1073                                 atomic_inc(&ref->refs);
1074
1075                                 spin_unlock(&delayed_refs->lock);
1076                                 mutex_lock(&head->mutex);
1077                                 mutex_unlock(&head->mutex);
1078
1079                                 btrfs_put_delayed_ref(ref);
1080                                 cond_resched();
1081                                 goto again;
1082                         }
1083                         node = rb_next(node);
1084                 }
1085                 spin_unlock(&delayed_refs->lock);
1086                 schedule_timeout(1);
1087                 goto again;
1088         }
1089 out:
1090         spin_unlock(&delayed_refs->lock);
1091         return 0;
1092 }
1093
1094 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1095                           struct btrfs_root *root, u64 objectid, u64 bytenr)
1096 {
1097         struct btrfs_root *extent_root = root->fs_info->extent_root;
1098         struct btrfs_path *path;
1099         struct extent_buffer *leaf;
1100         struct btrfs_extent_ref *ref_item;
1101         struct btrfs_key key;
1102         struct btrfs_key found_key;
1103         u64 ref_root;
1104         u64 last_snapshot;
1105         u32 nritems;
1106         int ret;
1107
1108         key.objectid = bytenr;
1109         key.offset = (u64)-1;
1110         key.type = BTRFS_EXTENT_ITEM_KEY;
1111
1112         path = btrfs_alloc_path();
1113         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1114         if (ret < 0)
1115                 goto out;
1116         BUG_ON(ret == 0);
1117
1118         ret = -ENOENT;
1119         if (path->slots[0] == 0)
1120                 goto out;
1121
1122         path->slots[0]--;
1123         leaf = path->nodes[0];
1124         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1125
1126         if (found_key.objectid != bytenr ||
1127             found_key.type != BTRFS_EXTENT_ITEM_KEY)
1128                 goto out;
1129
1130         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1131         while (1) {
1132                 leaf = path->nodes[0];
1133                 nritems = btrfs_header_nritems(leaf);
1134                 if (path->slots[0] >= nritems) {
1135                         ret = btrfs_next_leaf(extent_root, path);
1136                         if (ret < 0)
1137                                 goto out;
1138                         if (ret == 0)
1139                                 continue;
1140                         break;
1141                 }
1142                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1143                 if (found_key.objectid != bytenr)
1144                         break;
1145
1146                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1147                         path->slots[0]++;
1148                         continue;
1149                 }
1150
1151                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1152                                           struct btrfs_extent_ref);
1153                 ref_root = btrfs_ref_root(leaf, ref_item);
1154                 if ((ref_root != root->root_key.objectid &&
1155                      ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1156                      objectid != btrfs_ref_objectid(leaf, ref_item)) {
1157                         ret = 1;
1158                         goto out;
1159                 }
1160                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1161                         ret = 1;
1162                         goto out;
1163                 }
1164
1165                 path->slots[0]++;
1166         }
1167         ret = 0;
1168 out:
1169         btrfs_free_path(path);
1170         return ret;
1171 }
1172
1173 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1174                     struct extent_buffer *buf, u32 nr_extents)
1175 {
1176         struct btrfs_key key;
1177         struct btrfs_file_extent_item *fi;
1178         u64 root_gen;
1179         u32 nritems;
1180         int i;
1181         int level;
1182         int ret = 0;
1183         int shared = 0;
1184
1185         if (!root->ref_cows)
1186                 return 0;
1187
1188         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1189                 shared = 0;
1190                 root_gen = root->root_key.offset;
1191         } else {
1192                 shared = 1;
1193                 root_gen = trans->transid - 1;
1194         }
1195
1196         level = btrfs_header_level(buf);
1197         nritems = btrfs_header_nritems(buf);
1198
1199         if (level == 0) {
1200                 struct btrfs_leaf_ref *ref;
1201                 struct btrfs_extent_info *info;
1202
1203                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1204                 if (!ref) {
1205                         ret = -ENOMEM;
1206                         goto out;
1207                 }
1208
1209                 ref->root_gen = root_gen;
1210                 ref->bytenr = buf->start;
1211                 ref->owner = btrfs_header_owner(buf);
1212                 ref->generation = btrfs_header_generation(buf);
1213                 ref->nritems = nr_extents;
1214                 info = ref->extents;
1215
1216                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1217                         u64 disk_bytenr;
1218                         btrfs_item_key_to_cpu(buf, &key, i);
1219                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1220                                 continue;
1221                         fi = btrfs_item_ptr(buf, i,
1222                                             struct btrfs_file_extent_item);
1223                         if (btrfs_file_extent_type(buf, fi) ==
1224                             BTRFS_FILE_EXTENT_INLINE)
1225                                 continue;
1226                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1227                         if (disk_bytenr == 0)
1228                                 continue;
1229
1230                         info->bytenr = disk_bytenr;
1231                         info->num_bytes =
1232                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1233                         info->objectid = key.objectid;
1234                         info->offset = key.offset;
1235                         info++;
1236                 }
1237
1238                 ret = btrfs_add_leaf_ref(root, ref, shared);
1239                 if (ret == -EEXIST && shared) {
1240                         struct btrfs_leaf_ref *old;
1241                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1242                         BUG_ON(!old);
1243                         btrfs_remove_leaf_ref(root, old);
1244                         btrfs_free_leaf_ref(root, old);
1245                         ret = btrfs_add_leaf_ref(root, ref, shared);
1246                 }
1247                 WARN_ON(ret);
1248                 btrfs_free_leaf_ref(root, ref);
1249         }
1250 out:
1251         return ret;
1252 }
1253
1254 /* when a block goes through cow, we update the reference counts of
1255  * everything that block points to.  The internal pointers of the block
1256  * can be in just about any order, and it is likely to have clusters of
1257  * things that are close together and clusters of things that are not.
1258  *
1259  * To help reduce the seeks that come with updating all of these reference
1260  * counts, sort them by byte number before actual updates are done.
1261  *
1262  * struct refsort is used to match byte number to slot in the btree block.
1263  * we sort based on the byte number and then use the slot to actually
1264  * find the item.
1265  *
1266  * struct refsort is smaller than strcut btrfs_item and smaller than
1267  * struct btrfs_key_ptr.  Since we're currently limited to the page size
1268  * for a btree block, there's no way for a kmalloc of refsorts for a
1269  * single node to be bigger than a page.
1270  */
1271 struct refsort {
1272         u64 bytenr;
1273         u32 slot;
1274 };
1275
1276 /*
1277  * for passing into sort()
1278  */
1279 static int refsort_cmp(const void *a_void, const void *b_void)
1280 {
1281         const struct refsort *a = a_void;
1282         const struct refsort *b = b_void;
1283
1284         if (a->bytenr < b->bytenr)
1285                 return -1;
1286         if (a->bytenr > b->bytenr)
1287                 return 1;
1288         return 0;
1289 }
1290
1291
1292 noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1293                            struct btrfs_root *root,
1294                            struct extent_buffer *orig_buf,
1295                            struct extent_buffer *buf, u32 *nr_extents)
1296 {
1297         u64 bytenr;
1298         u64 ref_root;
1299         u64 orig_root;
1300         u64 ref_generation;
1301         u64 orig_generation;
1302         struct refsort *sorted;
1303         u32 nritems;
1304         u32 nr_file_extents = 0;
1305         struct btrfs_key key;
1306         struct btrfs_file_extent_item *fi;
1307         int i;
1308         int level;
1309         int ret = 0;
1310         int faili = 0;
1311         int refi = 0;
1312         int slot;
1313         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1314                             u64, u64, u64, u64, u64, u64, u64, u64, u64);
1315
1316         ref_root = btrfs_header_owner(buf);
1317         ref_generation = btrfs_header_generation(buf);
1318         orig_root = btrfs_header_owner(orig_buf);
1319         orig_generation = btrfs_header_generation(orig_buf);
1320
1321         nritems = btrfs_header_nritems(buf);
1322         level = btrfs_header_level(buf);
1323
1324         sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS);
1325         BUG_ON(!sorted);
1326
1327         if (root->ref_cows) {
1328                 process_func = __btrfs_inc_extent_ref;
1329         } else {
1330                 if (level == 0 &&
1331                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1332                         goto out;
1333                 if (level != 0 &&
1334                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1335                         goto out;
1336                 process_func = __btrfs_update_extent_ref;
1337         }
1338
1339         /*
1340          * we make two passes through the items.  In the first pass we
1341          * only record the byte number and slot.  Then we sort based on
1342          * byte number and do the actual work based on the sorted results
1343          */
1344         for (i = 0; i < nritems; i++) {
1345                 cond_resched();
1346                 if (level == 0) {
1347                         btrfs_item_key_to_cpu(buf, &key, i);
1348                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1349                                 continue;
1350                         fi = btrfs_item_ptr(buf, i,
1351                                             struct btrfs_file_extent_item);
1352                         if (btrfs_file_extent_type(buf, fi) ==
1353                             BTRFS_FILE_EXTENT_INLINE)
1354                                 continue;
1355                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1356                         if (bytenr == 0)
1357                                 continue;
1358
1359                         nr_file_extents++;
1360                         sorted[refi].bytenr = bytenr;
1361                         sorted[refi].slot = i;
1362                         refi++;
1363                 } else {
1364                         bytenr = btrfs_node_blockptr(buf, i);
1365                         sorted[refi].bytenr = bytenr;
1366                         sorted[refi].slot = i;
1367                         refi++;
1368                 }
1369         }
1370         /*
1371          * if refi == 0, we didn't actually put anything into the sorted
1372          * array and we're done
1373          */
1374         if (refi == 0)
1375                 goto out;
1376
1377         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
1378
1379         for (i = 0; i < refi; i++) {
1380                 cond_resched();
1381                 slot = sorted[i].slot;
1382                 bytenr = sorted[i].bytenr;
1383
1384                 if (level == 0) {
1385                         btrfs_item_key_to_cpu(buf, &key, slot);
1386                         fi = btrfs_item_ptr(buf, slot,
1387                                             struct btrfs_file_extent_item);
1388
1389                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1390                         if (bytenr == 0)
1391                                 continue;
1392
1393                         ret = process_func(trans, root, bytenr,
1394                                    btrfs_file_extent_disk_num_bytes(buf, fi),
1395                                    orig_buf->start, buf->start,
1396                                    orig_root, ref_root,
1397                                    orig_generation, ref_generation,
1398                                    key.objectid);
1399
1400                         if (ret) {
1401                                 faili = slot;
1402                                 WARN_ON(1);
1403                                 goto fail;
1404                         }
1405                 } else {
1406                         ret = process_func(trans, root, bytenr, buf->len,
1407                                            orig_buf->start, buf->start,
1408                                            orig_root, ref_root,
1409                                            orig_generation, ref_generation,
1410                                            level - 1);
1411                         if (ret) {
1412                                 faili = slot;
1413                                 WARN_ON(1);
1414                                 goto fail;
1415                         }
1416                 }
1417         }
1418 out:
1419         kfree(sorted);
1420         if (nr_extents) {
1421                 if (level == 0)
1422                         *nr_extents = nr_file_extents;
1423                 else
1424                         *nr_extents = nritems;
1425         }
1426         return 0;
1427 fail:
1428         kfree(sorted);
1429         WARN_ON(1);
1430         return ret;
1431 }
1432
1433 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1434                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1435                      struct extent_buffer *buf, int start_slot, int nr)
1436
1437 {
1438         u64 bytenr;
1439         u64 ref_root;
1440         u64 orig_root;
1441         u64 ref_generation;
1442         u64 orig_generation;
1443         struct btrfs_key key;
1444         struct btrfs_file_extent_item *fi;
1445         int i;
1446         int ret;
1447         int slot;
1448         int level;
1449
1450         BUG_ON(start_slot < 0);
1451         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1452
1453         ref_root = btrfs_header_owner(buf);
1454         ref_generation = btrfs_header_generation(buf);
1455         orig_root = btrfs_header_owner(orig_buf);
1456         orig_generation = btrfs_header_generation(orig_buf);
1457         level = btrfs_header_level(buf);
1458
1459         if (!root->ref_cows) {
1460                 if (level == 0 &&
1461                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1462                         return 0;
1463                 if (level != 0 &&
1464                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1465                         return 0;
1466         }
1467
1468         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1469                 cond_resched();
1470                 if (level == 0) {
1471                         btrfs_item_key_to_cpu(buf, &key, slot);
1472                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1473                                 continue;
1474                         fi = btrfs_item_ptr(buf, slot,
1475                                             struct btrfs_file_extent_item);
1476                         if (btrfs_file_extent_type(buf, fi) ==
1477                             BTRFS_FILE_EXTENT_INLINE)
1478                                 continue;
1479                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1480                         if (bytenr == 0)
1481                                 continue;
1482                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1483                                     btrfs_file_extent_disk_num_bytes(buf, fi),
1484                                     orig_buf->start, buf->start,
1485                                     orig_root, ref_root, orig_generation,
1486                                     ref_generation, key.objectid);
1487                         if (ret)
1488                                 goto fail;
1489                 } else {
1490                         bytenr = btrfs_node_blockptr(buf, slot);
1491                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1492                                             buf->len, orig_buf->start,
1493                                             buf->start, orig_root, ref_root,
1494                                             orig_generation, ref_generation,
1495                                             level - 1);
1496                         if (ret)
1497                                 goto fail;
1498                 }
1499         }
1500         return 0;
1501 fail:
1502         WARN_ON(1);
1503         return -1;
1504 }
1505
1506 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1507                                  struct btrfs_root *root,
1508                                  struct btrfs_path *path,
1509                                  struct btrfs_block_group_cache *cache)
1510 {
1511         int ret;
1512         struct btrfs_root *extent_root = root->fs_info->extent_root;
1513         unsigned long bi;
1514         struct extent_buffer *leaf;
1515
1516         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1517         if (ret < 0)
1518                 goto fail;
1519         BUG_ON(ret);
1520
1521         leaf = path->nodes[0];
1522         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1523         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1524         btrfs_mark_buffer_dirty(leaf);
1525         btrfs_release_path(extent_root, path);
1526 fail:
1527         if (ret)
1528                 return ret;
1529         return 0;
1530
1531 }
1532
1533 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1534                                    struct btrfs_root *root)
1535 {
1536         struct btrfs_block_group_cache *cache, *entry;
1537         struct rb_node *n;
1538         int err = 0;
1539         int werr = 0;
1540         struct btrfs_path *path;
1541         u64 last = 0;
1542
1543         path = btrfs_alloc_path();
1544         if (!path)
1545                 return -ENOMEM;
1546
1547         while (1) {
1548                 cache = NULL;
1549                 spin_lock(&root->fs_info->block_group_cache_lock);
1550                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1551                      n; n = rb_next(n)) {
1552                         entry = rb_entry(n, struct btrfs_block_group_cache,
1553                                          cache_node);
1554                         if (entry->dirty) {
1555                                 cache = entry;
1556                                 break;
1557                         }
1558                 }
1559                 spin_unlock(&root->fs_info->block_group_cache_lock);
1560
1561                 if (!cache)
1562                         break;
1563
1564                 cache->dirty = 0;
1565                 last += cache->key.offset;
1566
1567                 err = write_one_cache_group(trans, root,
1568                                             path, cache);
1569                 /*
1570                  * if we fail to write the cache group, we want
1571                  * to keep it marked dirty in hopes that a later
1572                  * write will work
1573                  */
1574                 if (err) {
1575                         werr = err;
1576                         continue;
1577                 }
1578         }
1579         btrfs_free_path(path);
1580         return werr;
1581 }
1582
1583 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1584 {
1585         struct btrfs_block_group_cache *block_group;
1586         int readonly = 0;
1587
1588         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1589         if (!block_group || block_group->ro)
1590                 readonly = 1;
1591         if (block_group)
1592                 put_block_group(block_group);
1593         return readonly;
1594 }
1595
1596 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1597                              u64 total_bytes, u64 bytes_used,
1598                              struct btrfs_space_info **space_info)
1599 {
1600         struct btrfs_space_info *found;
1601
1602         found = __find_space_info(info, flags);
1603         if (found) {
1604                 spin_lock(&found->lock);
1605                 found->total_bytes += total_bytes;
1606                 found->bytes_used += bytes_used;
1607                 found->full = 0;
1608                 spin_unlock(&found->lock);
1609                 *space_info = found;
1610                 return 0;
1611         }
1612         found = kzalloc(sizeof(*found), GFP_NOFS);
1613         if (!found)
1614                 return -ENOMEM;
1615
1616         INIT_LIST_HEAD(&found->block_groups);
1617         init_rwsem(&found->groups_sem);
1618         spin_lock_init(&found->lock);
1619         found->flags = flags;
1620         found->total_bytes = total_bytes;
1621         found->bytes_used = bytes_used;
1622         found->bytes_pinned = 0;
1623         found->bytes_reserved = 0;
1624         found->bytes_readonly = 0;
1625         found->bytes_delalloc = 0;
1626         found->full = 0;
1627         found->force_alloc = 0;
1628         *space_info = found;
1629         list_add_rcu(&found->list, &info->space_info);
1630         return 0;
1631 }
1632
1633 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1634 {
1635         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1636                                    BTRFS_BLOCK_GROUP_RAID1 |
1637                                    BTRFS_BLOCK_GROUP_RAID10 |
1638                                    BTRFS_BLOCK_GROUP_DUP);
1639         if (extra_flags) {
1640                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1641                         fs_info->avail_data_alloc_bits |= extra_flags;
1642                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1643                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1644                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1645                         fs_info->avail_system_alloc_bits |= extra_flags;
1646         }
1647 }
1648
1649 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1650 {
1651         spin_lock(&cache->space_info->lock);
1652         spin_lock(&cache->lock);
1653         if (!cache->ro) {
1654                 cache->space_info->bytes_readonly += cache->key.offset -
1655                                         btrfs_block_group_used(&cache->item);
1656                 cache->ro = 1;
1657         }
1658         spin_unlock(&cache->lock);
1659         spin_unlock(&cache->space_info->lock);
1660 }
1661
1662 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1663 {
1664         u64 num_devices = root->fs_info->fs_devices->rw_devices;
1665
1666         if (num_devices == 1)
1667                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1668         if (num_devices < 4)
1669                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1670
1671         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1672             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1673                       BTRFS_BLOCK_GROUP_RAID10))) {
1674                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1675         }
1676
1677         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1678             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1679                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1680         }
1681
1682         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1683             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1684              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1685              (flags & BTRFS_BLOCK_GROUP_DUP)))
1686                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1687         return flags;
1688 }
1689
1690 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
1691 {
1692         struct btrfs_fs_info *info = root->fs_info;
1693         u64 alloc_profile;
1694
1695         if (data) {
1696                 alloc_profile = info->avail_data_alloc_bits &
1697                         info->data_alloc_profile;
1698                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1699         } else if (root == root->fs_info->chunk_root) {
1700                 alloc_profile = info->avail_system_alloc_bits &
1701                         info->system_alloc_profile;
1702                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1703         } else {
1704                 alloc_profile = info->avail_metadata_alloc_bits &
1705                         info->metadata_alloc_profile;
1706                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1707         }
1708
1709         return btrfs_reduce_alloc_profile(root, data);
1710 }
1711
1712 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
1713 {
1714         u64 alloc_target;
1715
1716         alloc_target = btrfs_get_alloc_profile(root, 1);
1717         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
1718                                                        alloc_target);
1719 }
1720
1721 /*
1722  * for now this just makes sure we have at least 5% of our metadata space free
1723  * for use.
1724  */
1725 int btrfs_check_metadata_free_space(struct btrfs_root *root)
1726 {
1727         struct btrfs_fs_info *info = root->fs_info;
1728         struct btrfs_space_info *meta_sinfo;
1729         u64 alloc_target, thresh;
1730         int committed = 0, ret;
1731
1732         /* get the space info for where the metadata will live */
1733         alloc_target = btrfs_get_alloc_profile(root, 0);
1734         meta_sinfo = __find_space_info(info, alloc_target);
1735
1736 again:
1737         spin_lock(&meta_sinfo->lock);
1738         if (!meta_sinfo->full)
1739                 thresh = meta_sinfo->total_bytes * 80;
1740         else
1741                 thresh = meta_sinfo->total_bytes * 95;
1742
1743         do_div(thresh, 100);
1744
1745         if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
1746             meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
1747                 struct btrfs_trans_handle *trans;
1748                 if (!meta_sinfo->full) {
1749                         meta_sinfo->force_alloc = 1;
1750                         spin_unlock(&meta_sinfo->lock);
1751
1752                         trans = btrfs_start_transaction(root, 1);
1753                         if (!trans)
1754                                 return -ENOMEM;
1755
1756                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1757                                              2 * 1024 * 1024, alloc_target, 0);
1758                         btrfs_end_transaction(trans, root);
1759                         goto again;
1760                 }
1761                 spin_unlock(&meta_sinfo->lock);
1762
1763                 if (!committed) {
1764                         committed = 1;
1765                         trans = btrfs_join_transaction(root, 1);
1766                         if (!trans)
1767                                 return -ENOMEM;
1768                         ret = btrfs_commit_transaction(trans, root);
1769                         if (ret)
1770                                 return ret;
1771                         goto again;
1772                 }
1773                 return -ENOSPC;
1774         }
1775         spin_unlock(&meta_sinfo->lock);
1776
1777         return 0;
1778 }
1779
1780 /*
1781  * This will check the space that the inode allocates from to make sure we have
1782  * enough space for bytes.
1783  */
1784 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
1785                                 u64 bytes)
1786 {
1787         struct btrfs_space_info *data_sinfo;
1788         int ret = 0, committed = 0;
1789
1790         /* make sure bytes are sectorsize aligned */
1791         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
1792
1793         data_sinfo = BTRFS_I(inode)->space_info;
1794 again:
1795         /* make sure we have enough space to handle the data first */
1796         spin_lock(&data_sinfo->lock);
1797         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
1798             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
1799             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
1800             data_sinfo->bytes_may_use < bytes) {
1801                 struct btrfs_trans_handle *trans;
1802
1803                 /*
1804                  * if we don't have enough free bytes in this space then we need
1805                  * to alloc a new chunk.
1806                  */
1807                 if (!data_sinfo->full) {
1808                         u64 alloc_target;
1809
1810                         data_sinfo->force_alloc = 1;
1811                         spin_unlock(&data_sinfo->lock);
1812
1813                         alloc_target = btrfs_get_alloc_profile(root, 1);
1814                         trans = btrfs_start_transaction(root, 1);
1815                         if (!trans)
1816                                 return -ENOMEM;
1817
1818                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1819                                              bytes + 2 * 1024 * 1024,
1820                                              alloc_target, 0);
1821                         btrfs_end_transaction(trans, root);
1822                         if (ret)
1823                                 return ret;
1824                         goto again;
1825                 }
1826                 spin_unlock(&data_sinfo->lock);
1827
1828                 /* commit the current transaction and try again */
1829                 if (!committed) {
1830                         committed = 1;
1831                         trans = btrfs_join_transaction(root, 1);
1832                         if (!trans)
1833                                 return -ENOMEM;
1834                         ret = btrfs_commit_transaction(trans, root);
1835                         if (ret)
1836                                 return ret;
1837                         goto again;
1838                 }
1839
1840                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
1841                        ", %llu bytes_used, %llu bytes_reserved, "
1842                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use"
1843                        "%llu total\n", bytes, data_sinfo->bytes_delalloc,
1844                        data_sinfo->bytes_used, data_sinfo->bytes_reserved,
1845                        data_sinfo->bytes_pinned, data_sinfo->bytes_readonly,
1846                        data_sinfo->bytes_may_use, data_sinfo->total_bytes);
1847                 return -ENOSPC;
1848         }
1849         data_sinfo->bytes_may_use += bytes;
1850         BTRFS_I(inode)->reserved_bytes += bytes;
1851         spin_unlock(&data_sinfo->lock);
1852
1853         return btrfs_check_metadata_free_space(root);
1854 }
1855
1856 /*
1857  * if there was an error for whatever reason after calling
1858  * btrfs_check_data_free_space, call this so we can cleanup the counters.
1859  */
1860 void btrfs_free_reserved_data_space(struct btrfs_root *root,
1861                                     struct inode *inode, u64 bytes)
1862 {
1863         struct btrfs_space_info *data_sinfo;
1864
1865         /* make sure bytes are sectorsize aligned */
1866         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
1867
1868         data_sinfo = BTRFS_I(inode)->space_info;
1869         spin_lock(&data_sinfo->lock);
1870         data_sinfo->bytes_may_use -= bytes;
1871         BTRFS_I(inode)->reserved_bytes -= bytes;
1872         spin_unlock(&data_sinfo->lock);
1873 }
1874
1875 /* called when we are adding a delalloc extent to the inode's io_tree */
1876 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
1877                                   u64 bytes)
1878 {
1879         struct btrfs_space_info *data_sinfo;
1880
1881         /* get the space info for where this inode will be storing its data */
1882         data_sinfo = BTRFS_I(inode)->space_info;
1883
1884         /* make sure we have enough space to handle the data first */
1885         spin_lock(&data_sinfo->lock);
1886         data_sinfo->bytes_delalloc += bytes;
1887
1888         /*
1889          * we are adding a delalloc extent without calling
1890          * btrfs_check_data_free_space first.  This happens on a weird
1891          * writepage condition, but shouldn't hurt our accounting
1892          */
1893         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
1894                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
1895                 BTRFS_I(inode)->reserved_bytes = 0;
1896         } else {
1897                 data_sinfo->bytes_may_use -= bytes;
1898                 BTRFS_I(inode)->reserved_bytes -= bytes;
1899         }
1900
1901         spin_unlock(&data_sinfo->lock);
1902 }
1903
1904 /* called when we are clearing an delalloc extent from the inode's io_tree */
1905 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
1906                               u64 bytes)
1907 {
1908         struct btrfs_space_info *info;
1909
1910         info = BTRFS_I(inode)->space_info;
1911
1912         spin_lock(&info->lock);
1913         info->bytes_delalloc -= bytes;
1914         spin_unlock(&info->lock);
1915 }
1916
1917 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1918                           struct btrfs_root *extent_root, u64 alloc_bytes,
1919                           u64 flags, int force)
1920 {
1921         struct btrfs_space_info *space_info;
1922         u64 thresh;
1923         int ret = 0;
1924
1925         mutex_lock(&extent_root->fs_info->chunk_mutex);
1926
1927         flags = btrfs_reduce_alloc_profile(extent_root, flags);
1928
1929         space_info = __find_space_info(extent_root->fs_info, flags);
1930         if (!space_info) {
1931                 ret = update_space_info(extent_root->fs_info, flags,
1932                                         0, 0, &space_info);
1933                 BUG_ON(ret);
1934         }
1935         BUG_ON(!space_info);
1936
1937         spin_lock(&space_info->lock);
1938         if (space_info->force_alloc) {
1939                 force = 1;
1940                 space_info->force_alloc = 0;
1941         }
1942         if (space_info->full) {
1943                 spin_unlock(&space_info->lock);
1944                 goto out;
1945         }
1946
1947         thresh = space_info->total_bytes - space_info->bytes_readonly;
1948         thresh = div_factor(thresh, 6);
1949         if (!force &&
1950            (space_info->bytes_used + space_info->bytes_pinned +
1951             space_info->bytes_reserved + alloc_bytes) < thresh) {
1952                 spin_unlock(&space_info->lock);
1953                 goto out;
1954         }
1955         spin_unlock(&space_info->lock);
1956
1957         ret = btrfs_alloc_chunk(trans, extent_root, flags);
1958         if (ret)
1959                 space_info->full = 1;
1960 out:
1961         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1962         return ret;
1963 }
1964
1965 static int update_block_group(struct btrfs_trans_handle *trans,
1966                               struct btrfs_root *root,
1967                               u64 bytenr, u64 num_bytes, int alloc,
1968                               int mark_free)
1969 {
1970         struct btrfs_block_group_cache *cache;
1971         struct btrfs_fs_info *info = root->fs_info;
1972         u64 total = num_bytes;
1973         u64 old_val;
1974         u64 byte_in_group;
1975
1976         while (total) {
1977                 cache = btrfs_lookup_block_group(info, bytenr);
1978                 if (!cache)
1979                         return -1;
1980                 byte_in_group = bytenr - cache->key.objectid;
1981                 WARN_ON(byte_in_group > cache->key.offset);
1982
1983                 spin_lock(&cache->space_info->lock);
1984                 spin_lock(&cache->lock);
1985                 cache->dirty = 1;
1986                 old_val = btrfs_block_group_used(&cache->item);
1987                 num_bytes = min(total, cache->key.offset - byte_in_group);
1988                 if (alloc) {
1989                         old_val += num_bytes;
1990                         cache->space_info->bytes_used += num_bytes;
1991                         if (cache->ro)
1992                                 cache->space_info->bytes_readonly -= num_bytes;
1993                         btrfs_set_block_group_used(&cache->item, old_val);
1994                         spin_unlock(&cache->lock);
1995                         spin_unlock(&cache->space_info->lock);
1996                 } else {
1997                         old_val -= num_bytes;
1998                         cache->space_info->bytes_used -= num_bytes;
1999                         if (cache->ro)
2000                                 cache->space_info->bytes_readonly += num_bytes;
2001                         btrfs_set_block_group_used(&cache->item, old_val);
2002                         spin_unlock(&cache->lock);
2003                         spin_unlock(&cache->space_info->lock);
2004                         if (mark_free) {
2005                                 int ret;
2006
2007                                 ret = btrfs_discard_extent(root, bytenr,
2008                                                            num_bytes);
2009                                 WARN_ON(ret);
2010
2011                                 ret = btrfs_add_free_space(cache, bytenr,
2012                                                            num_bytes);
2013                                 WARN_ON(ret);
2014                         }
2015                 }
2016                 put_block_group(cache);
2017                 total -= num_bytes;
2018                 bytenr += num_bytes;
2019         }
2020         return 0;
2021 }
2022
2023 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2024 {
2025         struct btrfs_block_group_cache *cache;
2026         u64 bytenr;
2027
2028         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2029         if (!cache)
2030                 return 0;
2031
2032         bytenr = cache->key.objectid;
2033         put_block_group(cache);
2034
2035         return bytenr;
2036 }
2037
2038 int btrfs_update_pinned_extents(struct btrfs_root *root,
2039                                 u64 bytenr, u64 num, int pin)
2040 {
2041         u64 len;
2042         struct btrfs_block_group_cache *cache;
2043         struct btrfs_fs_info *fs_info = root->fs_info;
2044
2045         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2046         if (pin) {
2047                 set_extent_dirty(&fs_info->pinned_extents,
2048                                 bytenr, bytenr + num - 1, GFP_NOFS);
2049         } else {
2050                 clear_extent_dirty(&fs_info->pinned_extents,
2051                                 bytenr, bytenr + num - 1, GFP_NOFS);
2052         }
2053         while (num > 0) {
2054                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2055                 BUG_ON(!cache);
2056                 len = min(num, cache->key.offset -
2057                           (bytenr - cache->key.objectid));
2058                 if (pin) {
2059                         spin_lock(&cache->space_info->lock);
2060                         spin_lock(&cache->lock);
2061                         cache->pinned += len;
2062                         cache->space_info->bytes_pinned += len;
2063                         spin_unlock(&cache->lock);
2064                         spin_unlock(&cache->space_info->lock);
2065                         fs_info->total_pinned += len;
2066                 } else {
2067                         spin_lock(&cache->space_info->lock);
2068                         spin_lock(&cache->lock);
2069                         cache->pinned -= len;
2070                         cache->space_info->bytes_pinned -= len;
2071                         spin_unlock(&cache->lock);
2072                         spin_unlock(&cache->space_info->lock);
2073                         fs_info->total_pinned -= len;
2074                         if (cache->cached)
2075                                 btrfs_add_free_space(cache, bytenr, len);
2076                 }
2077                 put_block_group(cache);
2078                 bytenr += len;
2079                 num -= len;
2080         }
2081         return 0;
2082 }
2083
2084 static int update_reserved_extents(struct btrfs_root *root,
2085                                    u64 bytenr, u64 num, int reserve)
2086 {
2087         u64 len;
2088         struct btrfs_block_group_cache *cache;
2089         struct btrfs_fs_info *fs_info = root->fs_info;
2090
2091         while (num > 0) {
2092                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2093                 BUG_ON(!cache);
2094                 len = min(num, cache->key.offset -
2095                           (bytenr - cache->key.objectid));
2096
2097                 spin_lock(&cache->space_info->lock);
2098                 spin_lock(&cache->lock);
2099                 if (reserve) {
2100                         cache->reserved += len;
2101                         cache->space_info->bytes_reserved += len;
2102                 } else {
2103                         cache->reserved -= len;
2104                         cache->space_info->bytes_reserved -= len;
2105                 }
2106                 spin_unlock(&cache->lock);
2107                 spin_unlock(&cache->space_info->lock);
2108                 put_block_group(cache);
2109                 bytenr += len;
2110                 num -= len;
2111         }
2112         return 0;
2113 }
2114
2115 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2116 {
2117         u64 last = 0;
2118         u64 start;
2119         u64 end;
2120         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2121         int ret;
2122
2123         mutex_lock(&root->fs_info->pinned_mutex);
2124         while (1) {
2125                 ret = find_first_extent_bit(pinned_extents, last,
2126                                             &start, &end, EXTENT_DIRTY);
2127                 if (ret)
2128                         break;
2129                 set_extent_dirty(copy, start, end, GFP_NOFS);
2130                 last = end + 1;
2131         }
2132         mutex_unlock(&root->fs_info->pinned_mutex);
2133         return 0;
2134 }
2135
2136 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2137                                struct btrfs_root *root,
2138                                struct extent_io_tree *unpin)
2139 {
2140         u64 start;
2141         u64 end;
2142         int ret;
2143
2144         mutex_lock(&root->fs_info->pinned_mutex);
2145         while (1) {
2146                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2147                                             EXTENT_DIRTY);
2148                 if (ret)
2149                         break;
2150
2151                 ret = btrfs_discard_extent(root, start, end + 1 - start);
2152
2153                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2154                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2155
2156                 if (need_resched()) {
2157                         mutex_unlock(&root->fs_info->pinned_mutex);
2158                         cond_resched();
2159                         mutex_lock(&root->fs_info->pinned_mutex);
2160                 }
2161         }
2162         mutex_unlock(&root->fs_info->pinned_mutex);
2163         return ret;
2164 }
2165
2166 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2167                           struct btrfs_root *root,
2168                           u64 bytenr, u64 num_bytes, int is_data)
2169 {
2170         int err = 0;
2171         struct extent_buffer *buf;
2172
2173         if (is_data)
2174                 goto pinit;
2175
2176         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2177         if (!buf)
2178                 goto pinit;
2179
2180         /* we can reuse a block if it hasn't been written
2181          * and it is from this transaction.  We can't
2182          * reuse anything from the tree log root because
2183          * it has tiny sub-transactions.
2184          */
2185         if (btrfs_buffer_uptodate(buf, 0) &&
2186             btrfs_try_tree_lock(buf)) {
2187                 u64 header_owner = btrfs_header_owner(buf);
2188                 u64 header_transid = btrfs_header_generation(buf);
2189                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2190                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2191                     header_owner != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2192                     header_transid == trans->transid &&
2193                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2194                         clean_tree_block(NULL, root, buf);
2195                         btrfs_tree_unlock(buf);
2196                         free_extent_buffer(buf);
2197                         return 1;
2198                 }
2199                 btrfs_tree_unlock(buf);
2200         }
2201         free_extent_buffer(buf);
2202 pinit:
2203         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2204
2205         BUG_ON(err < 0);
2206         return 0;
2207 }
2208
2209 /*
2210  * remove an extent from the root, returns 0 on success
2211  */
2212 static int __free_extent(struct btrfs_trans_handle *trans,
2213                          struct btrfs_root *root,
2214                          u64 bytenr, u64 num_bytes, u64 parent,
2215                          u64 root_objectid, u64 ref_generation,
2216                          u64 owner_objectid, int pin, int mark_free,
2217                          int refs_to_drop)
2218 {
2219         struct btrfs_path *path;
2220         struct btrfs_key key;
2221         struct btrfs_fs_info *info = root->fs_info;
2222         struct btrfs_root *extent_root = info->extent_root;
2223         struct extent_buffer *leaf;
2224         int ret;
2225         int extent_slot = 0;
2226         int found_extent = 0;
2227         int num_to_del = 1;
2228         struct btrfs_extent_item *ei;
2229         u32 refs;
2230
2231         key.objectid = bytenr;
2232         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2233         key.offset = num_bytes;
2234         path = btrfs_alloc_path();
2235         if (!path)
2236                 return -ENOMEM;
2237
2238         path->reada = 1;
2239         ret = lookup_extent_backref(trans, extent_root, path,
2240                                     bytenr, parent, root_objectid,
2241                                     ref_generation, owner_objectid, 1);
2242         if (ret == 0) {
2243                 struct btrfs_key found_key;
2244                 extent_slot = path->slots[0];
2245                 while (extent_slot > 0) {
2246                         extent_slot--;
2247                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2248                                               extent_slot);
2249                         if (found_key.objectid != bytenr)
2250                                 break;
2251                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2252                             found_key.offset == num_bytes) {
2253                                 found_extent = 1;
2254                                 break;
2255                         }
2256                         if (path->slots[0] - extent_slot > 5)
2257                                 break;
2258                 }
2259                 if (!found_extent) {
2260                         ret = remove_extent_backref(trans, extent_root, path,
2261                                                     refs_to_drop);
2262                         BUG_ON(ret);
2263                         btrfs_release_path(extent_root, path);
2264                         ret = btrfs_search_slot(trans, extent_root,
2265                                                 &key, path, -1, 1);
2266                         if (ret) {
2267                                 printk(KERN_ERR "umm, got %d back from search"
2268                                        ", was looking for %llu\n", ret,
2269                                        (unsigned long long)bytenr);
2270                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2271                         }
2272                         BUG_ON(ret);
2273                         extent_slot = path->slots[0];
2274                 }
2275         } else {
2276                 btrfs_print_leaf(extent_root, path->nodes[0]);
2277                 WARN_ON(1);
2278                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2279                        "parent %llu root %llu gen %llu owner %llu\n",
2280                        (unsigned long long)bytenr,
2281                        (unsigned long long)parent,
2282                        (unsigned long long)root_objectid,
2283                        (unsigned long long)ref_generation,
2284                        (unsigned long long)owner_objectid);
2285         }
2286
2287         leaf = path->nodes[0];
2288         ei = btrfs_item_ptr(leaf, extent_slot,
2289                             struct btrfs_extent_item);
2290         refs = btrfs_extent_refs(leaf, ei);
2291
2292         /*
2293          * we're not allowed to delete the extent item if there
2294          * are other delayed ref updates pending
2295          */
2296
2297         BUG_ON(refs < refs_to_drop);
2298         refs -= refs_to_drop;
2299         btrfs_set_extent_refs(leaf, ei, refs);
2300         btrfs_mark_buffer_dirty(leaf);
2301
2302         if (refs == 0 && found_extent &&
2303             path->slots[0] == extent_slot + 1) {
2304                 struct btrfs_extent_ref *ref;
2305                 ref = btrfs_item_ptr(leaf, path->slots[0],
2306                                      struct btrfs_extent_ref);
2307                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != refs_to_drop);
2308                 /* if the back ref and the extent are next to each other
2309                  * they get deleted below in one shot
2310                  */
2311                 path->slots[0] = extent_slot;
2312                 num_to_del = 2;
2313         } else if (found_extent) {
2314                 /* otherwise delete the extent back ref */
2315                 ret = remove_extent_backref(trans, extent_root, path,
2316                                             refs_to_drop);
2317                 BUG_ON(ret);
2318                 /* if refs are 0, we need to setup the path for deletion */
2319                 if (refs == 0) {
2320                         btrfs_release_path(extent_root, path);
2321                         ret = btrfs_search_slot(trans, extent_root, &key, path,
2322                                                 -1, 1);
2323                         BUG_ON(ret);
2324                 }
2325         }
2326
2327         if (refs == 0) {
2328                 u64 super_used;
2329                 u64 root_used;
2330
2331                 if (pin) {
2332                         mutex_lock(&root->fs_info->pinned_mutex);
2333                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2334                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2335                         mutex_unlock(&root->fs_info->pinned_mutex);
2336                         if (ret > 0)
2337                                 mark_free = 1;
2338                         BUG_ON(ret < 0);
2339                 }
2340                 /* block accounting for super block */
2341                 spin_lock(&info->delalloc_lock);
2342                 super_used = btrfs_super_bytes_used(&info->super_copy);
2343                 btrfs_set_super_bytes_used(&info->super_copy,
2344                                            super_used - num_bytes);
2345
2346                 /* block accounting for root item */
2347                 root_used = btrfs_root_used(&root->root_item);
2348                 btrfs_set_root_used(&root->root_item,
2349                                            root_used - num_bytes);
2350                 spin_unlock(&info->delalloc_lock);
2351                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2352                                       num_to_del);
2353                 BUG_ON(ret);
2354                 btrfs_release_path(extent_root, path);
2355
2356                 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2357                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2358                         BUG_ON(ret);
2359                 }
2360
2361                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2362                                          mark_free);
2363                 BUG_ON(ret);
2364         }
2365         btrfs_free_path(path);
2366         return ret;
2367 }
2368
2369 /*
2370  * remove an extent from the root, returns 0 on success
2371  */
2372 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2373                                         struct btrfs_root *root,
2374                                         u64 bytenr, u64 num_bytes, u64 parent,
2375                                         u64 root_objectid, u64 ref_generation,
2376                                         u64 owner_objectid, int pin,
2377                                         int refs_to_drop)
2378 {
2379         WARN_ON(num_bytes < root->sectorsize);
2380
2381         /*
2382          * if metadata always pin
2383          * if data pin when any transaction has committed this
2384          */
2385         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID ||
2386             ref_generation != trans->transid)
2387                 pin = 1;
2388
2389         if (ref_generation != trans->transid)
2390                 pin = 1;
2391
2392         return __free_extent(trans, root, bytenr, num_bytes, parent,
2393                             root_objectid, ref_generation,
2394                             owner_objectid, pin, pin == 0, refs_to_drop);
2395 }
2396
2397 /*
2398  * when we free an extent, it is possible (and likely) that we free the last
2399  * delayed ref for that extent as well.  This searches the delayed ref tree for
2400  * a given extent, and if there are no other delayed refs to be processed, it
2401  * removes it from the tree.
2402  */
2403 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
2404                                       struct btrfs_root *root, u64 bytenr)
2405 {
2406         struct btrfs_delayed_ref_head *head;
2407         struct btrfs_delayed_ref_root *delayed_refs;
2408         struct btrfs_delayed_ref_node *ref;
2409         struct rb_node *node;
2410         int ret;
2411
2412         delayed_refs = &trans->transaction->delayed_refs;
2413         spin_lock(&delayed_refs->lock);
2414         head = btrfs_find_delayed_ref_head(trans, bytenr);
2415         if (!head)
2416                 goto out;
2417
2418         node = rb_prev(&head->node.rb_node);
2419         if (!node)
2420                 goto out;
2421
2422         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2423
2424         /* there are still entries for this ref, we can't drop it */
2425         if (ref->bytenr == bytenr)
2426                 goto out;
2427
2428         /*
2429          * waiting for the lock here would deadlock.  If someone else has it
2430          * locked they are already in the process of dropping it anyway
2431          */
2432         if (!mutex_trylock(&head->mutex))
2433                 goto out;
2434
2435         /*
2436          * at this point we have a head with no other entries.  Go
2437          * ahead and process it.
2438          */
2439         head->node.in_tree = 0;
2440         rb_erase(&head->node.rb_node, &delayed_refs->root);
2441
2442         delayed_refs->num_entries--;
2443
2444         /*
2445          * we don't take a ref on the node because we're removing it from the
2446          * tree, so we just steal the ref the tree was holding.
2447          */
2448         delayed_refs->num_heads--;
2449         if (list_empty(&head->cluster))
2450                 delayed_refs->num_heads_ready--;
2451
2452         list_del_init(&head->cluster);
2453         spin_unlock(&delayed_refs->lock);
2454
2455         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
2456                                   &head->node, head->must_insert_reserved);
2457         BUG_ON(ret);
2458         btrfs_put_delayed_ref(&head->node);
2459         return 0;
2460 out:
2461         spin_unlock(&delayed_refs->lock);
2462         return 0;
2463 }
2464
2465 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2466                       struct btrfs_root *root,
2467                       u64 bytenr, u64 num_bytes, u64 parent,
2468                       u64 root_objectid, u64 ref_generation,
2469                       u64 owner_objectid, int pin)
2470 {
2471         int ret;
2472
2473         /*
2474          * tree log blocks never actually go into the extent allocation
2475          * tree, just update pinning info and exit early.
2476          *
2477          * data extents referenced by the tree log do need to have
2478          * their reference counts bumped.
2479          */
2480         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID &&
2481             owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2482                 mutex_lock(&root->fs_info->pinned_mutex);
2483                 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2484                 mutex_unlock(&root->fs_info->pinned_mutex);
2485                 update_reserved_extents(root, bytenr, num_bytes, 0);
2486                 ret = 0;
2487         } else {
2488                 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent,
2489                                        root_objectid, ref_generation,
2490                                        owner_objectid,
2491                                        BTRFS_DROP_DELAYED_REF, 1);
2492                 BUG_ON(ret);
2493                 ret = check_ref_cleanup(trans, root, bytenr);
2494                 BUG_ON(ret);
2495         }
2496         return ret;
2497 }
2498
2499 static u64 stripe_align(struct btrfs_root *root, u64 val)
2500 {
2501         u64 mask = ((u64)root->stripesize - 1);
2502         u64 ret = (val + mask) & ~mask;
2503         return ret;
2504 }
2505
2506 /*
2507  * walks the btree of allocated extents and find a hole of a given size.
2508  * The key ins is changed to record the hole:
2509  * ins->objectid == block start
2510  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2511  * ins->offset == number of blocks
2512  * Any available blocks before search_start are skipped.
2513  */
2514 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2515                                      struct btrfs_root *orig_root,
2516                                      u64 num_bytes, u64 empty_size,
2517                                      u64 search_start, u64 search_end,
2518                                      u64 hint_byte, struct btrfs_key *ins,
2519                                      u64 exclude_start, u64 exclude_nr,
2520                                      int data)
2521 {
2522         int ret = 0;
2523         struct btrfs_root *root = orig_root->fs_info->extent_root;
2524         u64 total_needed = num_bytes;
2525         u64 *last_ptr = NULL;
2526         u64 last_wanted = 0;
2527         struct btrfs_block_group_cache *block_group = NULL;
2528         int chunk_alloc_done = 0;
2529         int empty_cluster = 2 * 1024 * 1024;
2530         int allowed_chunk_alloc = 0;
2531         struct list_head *head = NULL, *cur = NULL;
2532         int loop = 0;
2533         int extra_loop = 0;
2534         struct btrfs_space_info *space_info;
2535
2536         WARN_ON(num_bytes < root->sectorsize);
2537         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2538         ins->objectid = 0;
2539         ins->offset = 0;
2540
2541         if (orig_root->ref_cows || empty_size)
2542                 allowed_chunk_alloc = 1;
2543
2544         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2545                 last_ptr = &root->fs_info->last_alloc;
2546                 if (!btrfs_test_opt(root, SSD))
2547                         empty_cluster = 64 * 1024;
2548         }
2549
2550         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2551                 last_ptr = &root->fs_info->last_data_alloc;
2552
2553         if (last_ptr) {
2554                 if (*last_ptr) {
2555                         hint_byte = *last_ptr;
2556                         last_wanted = *last_ptr;
2557                 } else
2558                         empty_size += empty_cluster;
2559         } else {
2560                 empty_cluster = 0;
2561         }
2562         search_start = max(search_start, first_logical_byte(root, 0));
2563         search_start = max(search_start, hint_byte);
2564
2565         if (last_wanted && search_start != last_wanted) {
2566                 last_wanted = 0;
2567                 empty_size += empty_cluster;
2568         }
2569
2570         total_needed += empty_size;
2571         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2572         if (!block_group)
2573                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2574                                                              search_start);
2575         space_info = __find_space_info(root->fs_info, data);
2576
2577         down_read(&space_info->groups_sem);
2578         while (1) {
2579                 struct btrfs_free_space *free_space;
2580                 /*
2581                  * the only way this happens if our hint points to a block
2582                  * group thats not of the proper type, while looping this
2583                  * should never happen
2584                  */
2585                 if (empty_size)
2586                         extra_loop = 1;
2587
2588                 if (!block_group)
2589                         goto new_group_no_lock;
2590
2591                 if (unlikely(!block_group->cached)) {
2592                         mutex_lock(&block_group->cache_mutex);
2593                         ret = cache_block_group(root, block_group);
2594                         mutex_unlock(&block_group->cache_mutex);
2595                         if (ret)
2596                                 break;
2597                 }
2598
2599                 mutex_lock(&block_group->alloc_mutex);
2600                 if (unlikely(!block_group_bits(block_group, data)))
2601                         goto new_group;
2602
2603                 if (unlikely(block_group->ro))
2604                         goto new_group;
2605
2606                 free_space = btrfs_find_free_space(block_group, search_start,
2607                                                    total_needed);
2608                 if (free_space) {
2609                         u64 start = block_group->key.objectid;
2610                         u64 end = block_group->key.objectid +
2611                                 block_group->key.offset;
2612
2613                         search_start = stripe_align(root, free_space->offset);
2614
2615                         /* move on to the next group */
2616                         if (search_start + num_bytes >= search_end)
2617                                 goto new_group;
2618
2619                         /* move on to the next group */
2620                         if (search_start + num_bytes > end)
2621                                 goto new_group;
2622
2623                         if (last_wanted && search_start != last_wanted) {
2624                                 total_needed += empty_cluster;
2625                                 empty_size += empty_cluster;
2626                                 last_wanted = 0;
2627                                 /*
2628                                  * if search_start is still in this block group
2629                                  * then we just re-search this block group
2630                                  */
2631                                 if (search_start >= start &&
2632                                     search_start < end) {
2633                                         mutex_unlock(&block_group->alloc_mutex);
2634                                         continue;
2635                                 }
2636
2637                                 /* else we go to the next block group */
2638                                 goto new_group;
2639                         }
2640
2641                         if (exclude_nr > 0 &&
2642                             (search_start + num_bytes > exclude_start &&
2643                              search_start < exclude_start + exclude_nr)) {
2644                                 search_start = exclude_start + exclude_nr;
2645                                 /*
2646                                  * if search_start is still in this block group
2647                                  * then we just re-search this block group
2648                                  */
2649                                 if (search_start >= start &&
2650                                     search_start < end) {
2651                                         mutex_unlock(&block_group->alloc_mutex);
2652                                         last_wanted = 0;
2653                                         continue;
2654                                 }
2655
2656                                 /* else we go to the next block group */
2657                                 goto new_group;
2658                         }
2659
2660                         ins->objectid = search_start;
2661                         ins->offset = num_bytes;
2662
2663                         btrfs_remove_free_space_lock(block_group, search_start,
2664                                                      num_bytes);
2665                         /* we are all good, lets return */
2666                         mutex_unlock(&block_group->alloc_mutex);
2667                         break;
2668                 }
2669 new_group:
2670                 mutex_unlock(&block_group->alloc_mutex);
2671                 put_block_group(block_group);
2672                 block_group = NULL;
2673 new_group_no_lock:
2674                 /* don't try to compare new allocations against the
2675                  * last allocation any more
2676                  */
2677                 last_wanted = 0;
2678
2679                 /*
2680                  * Here's how this works.
2681                  * loop == 0: we were searching a block group via a hint
2682                  *              and didn't find anything, so we start at
2683                  *              the head of the block groups and keep searching
2684                  * loop == 1: we're searching through all of the block groups
2685                  *              if we hit the head again we have searched
2686                  *              all of the block groups for this space and we
2687                  *              need to try and allocate, if we cant error out.
2688                  * loop == 2: we allocated more space and are looping through
2689                  *              all of the block groups again.
2690                  */
2691                 if (loop == 0) {
2692                         head = &space_info->block_groups;
2693                         cur = head->next;
2694                         loop++;
2695                 } else if (loop == 1 && cur == head) {
2696                         int keep_going;
2697
2698                         /* at this point we give up on the empty_size
2699                          * allocations and just try to allocate the min
2700                          * space.
2701                          *
2702                          * The extra_loop field was set if an empty_size
2703                          * allocation was attempted above, and if this
2704                          * is try we need to try the loop again without
2705                          * the additional empty_size.
2706                          */
2707                         total_needed -= empty_size;
2708                         empty_size = 0;
2709                         keep_going = extra_loop;
2710                         loop++;
2711
2712                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2713                                 up_read(&space_info->groups_sem);
2714                                 ret = do_chunk_alloc(trans, root, num_bytes +
2715                                                      2 * 1024 * 1024, data, 1);
2716                                 down_read(&space_info->groups_sem);
2717                                 if (ret < 0)
2718                                         goto loop_check;
2719                                 head = &space_info->block_groups;
2720                                 /*
2721                                  * we've allocated a new chunk, keep
2722                                  * trying
2723                                  */
2724                                 keep_going = 1;
2725                                 chunk_alloc_done = 1;
2726                         } else if (!allowed_chunk_alloc) {
2727                                 space_info->force_alloc = 1;
2728                         }
2729 loop_check:
2730                         if (keep_going) {
2731                                 cur = head->next;
2732                                 extra_loop = 0;
2733                         } else {
2734                                 break;
2735                         }
2736                 } else if (cur == head) {
2737                         break;
2738                 }
2739
2740                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2741                                          list);
2742                 atomic_inc(&block_group->count);
2743
2744                 search_start = block_group->key.objectid;
2745                 cur = cur->next;
2746         }
2747
2748         /* we found what we needed */
2749         if (ins->objectid) {
2750                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2751                         trans->block_group = block_group->key.objectid;
2752
2753                 if (last_ptr)
2754                         *last_ptr = ins->objectid + ins->offset;
2755                 ret = 0;
2756         } else if (!ret) {
2757                 printk(KERN_ERR "btrfs searching for %llu bytes, "
2758                        "num_bytes %llu, loop %d, allowed_alloc %d\n",
2759                        (unsigned long long)total_needed,
2760                        (unsigned long long)num_bytes,
2761                        loop, allowed_chunk_alloc);
2762                 ret = -ENOSPC;
2763         }
2764         if (block_group)
2765                 put_block_group(block_group);
2766
2767         up_read(&space_info->groups_sem);
2768         return ret;
2769 }
2770
2771 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2772 {
2773         struct btrfs_block_group_cache *cache;
2774
2775         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
2776                (unsigned long long)(info->total_bytes - info->bytes_used -
2777                                     info->bytes_pinned - info->bytes_reserved),
2778                (info->full) ? "" : "not ");
2779         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
2780                " may_use=%llu, used=%llu\n", info->total_bytes,
2781                info->bytes_pinned, info->bytes_delalloc, info->bytes_may_use,
2782                info->bytes_used);
2783
2784         down_read(&info->groups_sem);
2785         list_for_each_entry(cache, &info->block_groups, list) {
2786                 spin_lock(&cache->lock);
2787                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
2788                        "%llu pinned %llu reserved\n",
2789                        (unsigned long long)cache->key.objectid,
2790                        (unsigned long long)cache->key.offset,
2791                        (unsigned long long)btrfs_block_group_used(&cache->item),
2792                        (unsigned long long)cache->pinned,
2793                        (unsigned long long)cache->reserved);
2794                 btrfs_dump_free_space(cache, bytes);
2795                 spin_unlock(&cache->lock);
2796         }
2797         up_read(&info->groups_sem);
2798 }
2799
2800 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2801                                   struct btrfs_root *root,
2802                                   u64 num_bytes, u64 min_alloc_size,
2803                                   u64 empty_size, u64 hint_byte,
2804                                   u64 search_end, struct btrfs_key *ins,
2805                                   u64 data)
2806 {
2807         int ret;
2808         u64 search_start = 0;
2809         struct btrfs_fs_info *info = root->fs_info;
2810
2811         data = btrfs_get_alloc_profile(root, data);
2812 again:
2813         /*
2814          * the only place that sets empty_size is btrfs_realloc_node, which
2815          * is not called recursively on allocations
2816          */
2817         if (empty_size || root->ref_cows) {
2818                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2819                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2820                                      2 * 1024 * 1024,
2821                                      BTRFS_BLOCK_GROUP_METADATA |
2822                                      (info->metadata_alloc_profile &
2823                                       info->avail_metadata_alloc_bits), 0);
2824                 }
2825                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2826                                      num_bytes + 2 * 1024 * 1024, data, 0);
2827         }
2828
2829         WARN_ON(num_bytes < root->sectorsize);
2830         ret = find_free_extent(trans, root, num_bytes, empty_size,
2831                                search_start, search_end, hint_byte, ins,
2832                                trans->alloc_exclude_start,
2833                                trans->alloc_exclude_nr, data);
2834
2835         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2836                 num_bytes = num_bytes >> 1;
2837                 num_bytes = num_bytes & ~(root->sectorsize - 1);
2838                 num_bytes = max(num_bytes, min_alloc_size);
2839                 do_chunk_alloc(trans, root->fs_info->extent_root,
2840                                num_bytes, data, 1);
2841                 goto again;
2842         }
2843         if (ret) {
2844                 struct btrfs_space_info *sinfo;
2845
2846                 sinfo = __find_space_info(root->fs_info, data);
2847                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
2848                        "wanted %llu\n", (unsigned long long)data,
2849                        (unsigned long long)num_bytes);
2850                 dump_space_info(sinfo, num_bytes);
2851                 BUG();
2852         }
2853
2854         return ret;
2855 }
2856
2857 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2858 {
2859         struct btrfs_block_group_cache *cache;
2860         int ret = 0;
2861
2862         cache = btrfs_lookup_block_group(root->fs_info, start);
2863         if (!cache) {
2864                 printk(KERN_ERR "Unable to find block group for %llu\n",
2865                        (unsigned long long)start);
2866                 return -ENOSPC;
2867         }
2868
2869         ret = btrfs_discard_extent(root, start, len);
2870
2871         btrfs_add_free_space(cache, start, len);
2872         put_block_group(cache);
2873         update_reserved_extents(root, start, len, 0);
2874
2875         return ret;
2876 }
2877
2878 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2879                                   struct btrfs_root *root,
2880                                   u64 num_bytes, u64 min_alloc_size,
2881                                   u64 empty_size, u64 hint_byte,
2882                                   u64 search_end, struct btrfs_key *ins,
2883                                   u64 data)
2884 {
2885         int ret;
2886         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2887                                      empty_size, hint_byte, search_end, ins,
2888                                      data);
2889         update_reserved_extents(root, ins->objectid, ins->offset, 1);
2890         return ret;
2891 }
2892
2893 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2894                                          struct btrfs_root *root, u64 parent,
2895                                          u64 root_objectid, u64 ref_generation,
2896                                          u64 owner, struct btrfs_key *ins,
2897                                          int ref_mod)
2898 {
2899         int ret;
2900         u64 super_used;
2901         u64 root_used;
2902         u64 num_bytes = ins->offset;
2903         u32 sizes[2];
2904         struct btrfs_fs_info *info = root->fs_info;
2905         struct btrfs_root *extent_root = info->extent_root;
2906         struct btrfs_extent_item *extent_item;
2907         struct btrfs_extent_ref *ref;
2908         struct btrfs_path *path;
2909         struct btrfs_key keys[2];
2910
2911         if (parent == 0)
2912                 parent = ins->objectid;
2913
2914         /* block accounting for super block */
2915         spin_lock(&info->delalloc_lock);
2916         super_used = btrfs_super_bytes_used(&info->super_copy);
2917         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2918
2919         /* block accounting for root item */
2920         root_used = btrfs_root_used(&root->root_item);
2921         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2922         spin_unlock(&info->delalloc_lock);
2923
2924         memcpy(&keys[0], ins, sizeof(*ins));
2925         keys[1].objectid = ins->objectid;
2926         keys[1].type = BTRFS_EXTENT_REF_KEY;
2927         keys[1].offset = parent;
2928         sizes[0] = sizeof(*extent_item);
2929         sizes[1] = sizeof(*ref);
2930
2931         path = btrfs_alloc_path();
2932         BUG_ON(!path);
2933
2934         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2935                                        sizes, 2);
2936         BUG_ON(ret);
2937
2938         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2939                                      struct btrfs_extent_item);
2940         btrfs_set_extent_refs(path->nodes[0], extent_item, ref_mod);
2941         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2942                              struct btrfs_extent_ref);
2943
2944         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2945         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2946         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2947         btrfs_set_ref_num_refs(path->nodes[0], ref, ref_mod);
2948
2949         btrfs_mark_buffer_dirty(path->nodes[0]);
2950
2951         trans->alloc_exclude_start = 0;
2952         trans->alloc_exclude_nr = 0;
2953         btrfs_free_path(path);
2954
2955         if (ret)
2956                 goto out;
2957
2958         ret = update_block_group(trans, root, ins->objectid,
2959                                  ins->offset, 1, 0);
2960         if (ret) {
2961                 printk(KERN_ERR "btrfs update block group failed for %llu "
2962                        "%llu\n", (unsigned long long)ins->objectid,
2963                        (unsigned long long)ins->offset);
2964                 BUG();
2965         }
2966 out:
2967         return ret;
2968 }
2969
2970 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2971                                 struct btrfs_root *root, u64 parent,
2972                                 u64 root_objectid, u64 ref_generation,
2973                                 u64 owner, struct btrfs_key *ins)
2974 {
2975         int ret;
2976
2977         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2978                 return 0;
2979
2980         ret = btrfs_add_delayed_ref(trans, ins->objectid,
2981                                     ins->offset, parent, root_objectid,
2982                                     ref_generation, owner,
2983                                     BTRFS_ADD_DELAYED_EXTENT, 0);
2984         BUG_ON(ret);
2985         return ret;
2986 }
2987
2988 /*
2989  * this is used by the tree logging recovery code.  It records that
2990  * an extent has been allocated and makes sure to clear the free
2991  * space cache bits as well
2992  */
2993 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2994                                 struct btrfs_root *root, u64 parent,
2995                                 u64 root_objectid, u64 ref_generation,
2996                                 u64 owner, struct btrfs_key *ins)
2997 {
2998         int ret;
2999         struct btrfs_block_group_cache *block_group;
3000
3001         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3002         mutex_lock(&block_group->cache_mutex);
3003         cache_block_group(root, block_group);
3004         mutex_unlock(&block_group->cache_mutex);
3005
3006         ret = btrfs_remove_free_space(block_group, ins->objectid,
3007                                       ins->offset);
3008         BUG_ON(ret);
3009         put_block_group(block_group);
3010         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3011                                             ref_generation, owner, ins, 1);
3012         return ret;
3013 }
3014
3015 /*
3016  * finds a free extent and does all the dirty work required for allocation
3017  * returns the key for the extent through ins, and a tree buffer for
3018  * the first block of the extent through buf.
3019  *
3020  * returns 0 if everything worked, non-zero otherwise.
3021  */
3022 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3023                        struct btrfs_root *root,
3024                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3025                        u64 root_objectid, u64 ref_generation,
3026                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3027                        u64 search_end, struct btrfs_key *ins, u64 data)
3028 {
3029         int ret;
3030         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3031                                      min_alloc_size, empty_size, hint_byte,
3032                                      search_end, ins, data);
3033         BUG_ON(ret);
3034         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3035                 ret = btrfs_add_delayed_ref(trans, ins->objectid,
3036                                             ins->offset, parent, root_objectid,
3037                                             ref_generation, owner_objectid,
3038                                             BTRFS_ADD_DELAYED_EXTENT, 0);
3039                 BUG_ON(ret);
3040         }
3041         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3042         return ret;
3043 }
3044
3045 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3046                                             struct btrfs_root *root,
3047                                             u64 bytenr, u32 blocksize,
3048                                             int level)
3049 {
3050         struct extent_buffer *buf;
3051
3052         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3053         if (!buf)
3054                 return ERR_PTR(-ENOMEM);
3055         btrfs_set_header_generation(buf, trans->transid);
3056         btrfs_set_buffer_lockdep_class(buf, level);
3057         btrfs_tree_lock(buf);
3058         clean_tree_block(trans, root, buf);
3059
3060         btrfs_set_lock_blocking(buf);
3061         btrfs_set_buffer_uptodate(buf);
3062
3063         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3064                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3065                          buf->start + buf->len - 1, GFP_NOFS);
3066         } else {
3067                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3068                          buf->start + buf->len - 1, GFP_NOFS);
3069         }
3070         trans->blocks_used++;
3071         /* this returns a buffer locked for blocking */
3072         return buf;
3073 }
3074
3075 /*
3076  * helper function to allocate a block for a given tree
3077  * returns the tree buffer or NULL.
3078  */
3079 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3080                                              struct btrfs_root *root,
3081                                              u32 blocksize, u64 parent,
3082                                              u64 root_objectid,
3083                                              u64 ref_generation,
3084                                              int level,
3085                                              u64 hint,
3086                                              u64 empty_size)
3087 {
3088         struct btrfs_key ins;
3089         int ret;
3090         struct extent_buffer *buf;
3091
3092         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3093                                  root_objectid, ref_generation, level,
3094                                  empty_size, hint, (u64)-1, &ins, 0);
3095         if (ret) {
3096                 BUG_ON(ret > 0);
3097                 return ERR_PTR(ret);
3098         }
3099
3100         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
3101                                     blocksize, level);
3102         return buf;
3103 }
3104
3105 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3106                         struct btrfs_root *root, struct extent_buffer *leaf)
3107 {
3108         u64 leaf_owner;
3109         u64 leaf_generation;
3110         struct refsort *sorted;
3111         struct btrfs_key key;
3112         struct btrfs_file_extent_item *fi;
3113         int i;
3114         int nritems;
3115         int ret;
3116         int refi = 0;
3117         int slot;
3118
3119         BUG_ON(!btrfs_is_leaf(leaf));
3120         nritems = btrfs_header_nritems(leaf);
3121         leaf_owner = btrfs_header_owner(leaf);
3122         leaf_generation = btrfs_header_generation(leaf);
3123
3124         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3125         /* we do this loop twice.  The first time we build a list
3126          * of the extents we have a reference on, then we sort the list
3127          * by bytenr.  The second time around we actually do the
3128          * extent freeing.
3129          */
3130         for (i = 0; i < nritems; i++) {
3131                 u64 disk_bytenr;
3132                 cond_resched();
3133
3134                 btrfs_item_key_to_cpu(leaf, &key, i);
3135
3136                 /* only extents have references, skip everything else */
3137                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3138                         continue;
3139
3140                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3141
3142                 /* inline extents live in the btree, they don't have refs */
3143                 if (btrfs_file_extent_type(leaf, fi) ==
3144                     BTRFS_FILE_EXTENT_INLINE)
3145                         continue;
3146
3147                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3148
3149                 /* holes don't have refs */
3150                 if (disk_bytenr == 0)
3151                         continue;
3152
3153                 sorted[refi].bytenr = disk_bytenr;
3154                 sorted[refi].slot = i;
3155                 refi++;
3156         }
3157
3158         if (refi == 0)
3159                 goto out;
3160
3161         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3162
3163         for (i = 0; i < refi; i++) {
3164                 u64 disk_bytenr;
3165
3166                 disk_bytenr = sorted[i].bytenr;
3167                 slot = sorted[i].slot;
3168
3169                 cond_resched();
3170
3171                 btrfs_item_key_to_cpu(leaf, &key, slot);
3172                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3173                         continue;
3174
3175                 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3176
3177                 ret = btrfs_free_extent(trans, root, disk_bytenr,
3178                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3179                                 leaf->start, leaf_owner, leaf_generation,
3180                                 key.objectid, 0);
3181                 BUG_ON(ret);
3182
3183                 atomic_inc(&root->fs_info->throttle_gen);
3184                 wake_up(&root->fs_info->transaction_throttle);
3185                 cond_resched();
3186         }
3187 out:
3188         kfree(sorted);
3189         return 0;
3190 }
3191
3192 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3193                                         struct btrfs_root *root,
3194                                         struct btrfs_leaf_ref *ref)
3195 {
3196         int i;
3197         int ret;
3198         struct btrfs_extent_info *info;
3199         struct refsort *sorted;
3200
3201         if (ref->nritems == 0)
3202                 return 0;
3203
3204         sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
3205         for (i = 0; i < ref->nritems; i++) {
3206                 sorted[i].bytenr = ref->extents[i].bytenr;
3207                 sorted[i].slot = i;
3208         }
3209         sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
3210
3211         /*
3212          * the items in the ref were sorted when the ref was inserted
3213          * into the ref cache, so this is already in order
3214          */
3215         for (i = 0; i < ref->nritems; i++) {
3216                 info = ref->extents + sorted[i].slot;
3217                 ret = btrfs_free_extent(trans, root, info->bytenr,
3218                                           info->num_bytes, ref->bytenr,
3219                                           ref->owner, ref->generation,
3220                                           info->objectid, 0);
3221
3222                 atomic_inc(&root->fs_info->throttle_gen);
3223                 wake_up(&root->fs_info->transaction_throttle);
3224                 cond_resched();
3225
3226                 BUG_ON(ret);
3227                 info++;
3228         }
3229
3230         kfree(sorted);
3231         return 0;
3232 }
3233
3234 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
3235                                      struct btrfs_root *root, u64 start,
3236                                      u64 len, u32 *refs)
3237 {
3238         int ret;
3239
3240         ret = btrfs_lookup_extent_ref(trans, root, start, len, refs);
3241         BUG_ON(ret);
3242
3243 #if 0 /* some debugging code in case we see problems here */
3244         /* if the refs count is one, it won't get increased again.  But
3245          * if the ref count is > 1, someone may be decreasing it at
3246          * the same time we are.
3247          */
3248         if (*refs != 1) {
3249                 struct extent_buffer *eb = NULL;
3250                 eb = btrfs_find_create_tree_block(root, start, len);
3251                 if (eb)
3252                         btrfs_tree_lock(eb);
3253
3254                 mutex_lock(&root->fs_info->alloc_mutex);
3255                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3256                 BUG_ON(ret);
3257                 mutex_unlock(&root->fs_info->alloc_mutex);
3258
3259                 if (eb) {
3260                         btrfs_tree_unlock(eb);
3261                         free_extent_buffer(eb);
3262                 }
3263                 if (*refs == 1) {
3264                         printk(KERN_ERR "btrfs block %llu went down to one "
3265                                "during drop_snap\n", (unsigned long long)start);
3266                 }
3267
3268         }
3269 #endif
3270
3271         cond_resched();
3272         return ret;
3273 }
3274
3275 /*
3276  * this is used while deleting old snapshots, and it drops the refs
3277  * on a whole subtree starting from a level 1 node.
3278  *
3279  * The idea is to sort all the leaf pointers, and then drop the
3280  * ref on all the leaves in order.  Most of the time the leaves
3281  * will have ref cache entries, so no leaf IOs will be required to
3282  * find the extents they have references on.
3283  *
3284  * For each leaf, any references it has are also dropped in order
3285  *
3286  * This ends up dropping the references in something close to optimal
3287  * order for reading and modifying the extent allocation tree.
3288  */
3289 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3290                                         struct btrfs_root *root,
3291                                         struct btrfs_path *path)
3292 {
3293         u64 bytenr;
3294         u64 root_owner;
3295         u64 root_gen;
3296         struct extent_buffer *eb = path->nodes[1];
3297         struct extent_buffer *leaf;
3298         struct btrfs_leaf_ref *ref;
3299         struct refsort *sorted = NULL;
3300         int nritems = btrfs_header_nritems(eb);
3301         int ret;
3302         int i;
3303         int refi = 0;
3304         int slot = path->slots[1];
3305         u32 blocksize = btrfs_level_size(root, 0);
3306         u32 refs;
3307
3308         if (nritems == 0)
3309                 goto out;
3310
3311         root_owner = btrfs_header_owner(eb);
3312         root_gen = btrfs_header_generation(eb);
3313         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3314
3315         /*
3316          * step one, sort all the leaf pointers so we don't scribble
3317          * randomly into the extent allocation tree
3318          */
3319         for (i = slot; i < nritems; i++) {
3320                 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
3321                 sorted[refi].slot = i;
3322                 refi++;
3323         }
3324
3325         /*
3326          * nritems won't be zero, but if we're picking up drop_snapshot
3327          * after a crash, slot might be > 0, so double check things
3328          * just in case.
3329          */
3330         if (refi == 0)
3331                 goto out;
3332
3333         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3334
3335         /*
3336          * the first loop frees everything the leaves point to
3337          */
3338         for (i = 0; i < refi; i++) {
3339                 u64 ptr_gen;
3340
3341                 bytenr = sorted[i].bytenr;
3342
3343                 /*
3344                  * check the reference count on this leaf.  If it is > 1
3345                  * we just decrement it below and don't update any
3346                  * of the refs the leaf points to.
3347                  */
3348                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3349                                                 blocksize, &refs);
3350                 BUG_ON(ret);
3351                 if (refs != 1)
3352                         continue;
3353
3354                 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
3355
3356                 /*
3357                  * the leaf only had one reference, which means the
3358                  * only thing pointing to this leaf is the snapshot
3359                  * we're deleting.  It isn't possible for the reference
3360                  * count to increase again later
3361                  *
3362                  * The reference cache is checked for the leaf,
3363                  * and if found we'll be able to drop any refs held by
3364                  * the leaf without needing to read it in.
3365                  */
3366                 ref = btrfs_lookup_leaf_ref(root, bytenr);
3367                 if (ref && ref->generation != ptr_gen) {
3368                         btrfs_free_leaf_ref(root, ref);
3369                         ref = NULL;
3370                 }
3371                 if (ref) {
3372                         ret = cache_drop_leaf_ref(trans, root, ref);
3373                         BUG_ON(ret);
3374                         btrfs_remove_leaf_ref(root, ref);
3375                         btrfs_free_leaf_ref(root, ref);
3376                 } else {
3377                         /*
3378                          * the leaf wasn't in the reference cache, so
3379                          * we have to read it.
3380                          */
3381                         leaf = read_tree_block(root, bytenr, blocksize,
3382                                                ptr_gen);
3383                         ret = btrfs_drop_leaf_ref(trans, root, leaf);
3384                         BUG_ON(ret);
3385                         free_extent_buffer(leaf);
3386                 }
3387                 atomic_inc(&root->fs_info->throttle_gen);
3388                 wake_up(&root->fs_info->transaction_throttle);
3389                 cond_resched();
3390         }
3391
3392         /*
3393          * run through the loop again to free the refs on the leaves.
3394          * This is faster than doing it in the loop above because
3395          * the leaves are likely to be clustered together.  We end up
3396          * working in nice chunks on the extent allocation tree.
3397          */
3398         for (i = 0; i < refi; i++) {
3399                 bytenr = sorted[i].bytenr;
3400                 ret = btrfs_free_extent(trans, root, bytenr,
3401                                         blocksize, eb->start,
3402                                         root_owner, root_gen, 0, 1);
3403                 BUG_ON(ret);
3404
3405                 atomic_inc(&root->fs_info->throttle_gen);
3406                 wake_up(&root->fs_info->transaction_throttle);
3407                 cond_resched();
3408         }
3409 out:
3410         kfree(sorted);
3411
3412         /*
3413          * update the path to show we've processed the entire level 1
3414          * node.  This will get saved into the root's drop_snapshot_progress
3415          * field so these drops are not repeated again if this transaction
3416          * commits.
3417          */
3418         path->slots[1] = nritems;
3419         return 0;
3420 }
3421
3422 /*
3423  * helper function for drop_snapshot, this walks down the tree dropping ref
3424  * counts as it goes.
3425  */
3426 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3427                                    struct btrfs_root *root,
3428                                    struct btrfs_path *path, int *level)
3429 {
3430         u64 root_owner;
3431         u64 root_gen;
3432         u64 bytenr;
3433         u64 ptr_gen;
3434         struct extent_buffer *next;
3435         struct extent_buffer *cur;
3436         struct extent_buffer *parent;
3437         u32 blocksize;
3438         int ret;
3439         u32 refs;
3440
3441         WARN_ON(*level < 0);
3442         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3443         ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
3444                                 path->nodes[*level]->len, &refs);
3445         BUG_ON(ret);
3446         if (refs > 1)
3447                 goto out;
3448
3449         /*
3450          * walk down to the last node level and free all the leaves
3451          */
3452         while (*level >= 0) {
3453                 WARN_ON(*level < 0);
3454                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3455                 cur = path->nodes[*level];
3456
3457                 if (btrfs_header_level(cur) != *level)
3458                         WARN_ON(1);
3459
3460                 if (path->slots[*level] >=
3461                     btrfs_header_nritems(cur))
3462                         break;
3463
3464                 /* the new code goes down to level 1 and does all the
3465                  * leaves pointed to that node in bulk.  So, this check
3466                  * for level 0 will always be false.
3467                  *
3468                  * But, the disk format allows the drop_snapshot_progress
3469                  * field in the root to leave things in a state where
3470                  * a leaf will need cleaning up here.  If someone crashes
3471                  * with the old code and then boots with the new code,
3472                  * we might find a leaf here.
3473                  */
3474                 if (*level == 0) {
3475                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3476                         BUG_ON(ret);
3477                         break;
3478                 }
3479
3480                 /*
3481                  * once we get to level one, process the whole node
3482                  * at once, including everything below it.
3483                  */
3484                 if (*level == 1) {
3485                         ret = drop_level_one_refs(trans, root, path);
3486                         BUG_ON(ret);
3487                         break;
3488                 }
3489
3490                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3491                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3492                 blocksize = btrfs_level_size(root, *level - 1);
3493
3494                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3495                                                 blocksize, &refs);
3496                 BUG_ON(ret);
3497
3498                 /*
3499                  * if there is more than one reference, we don't need
3500                  * to read that node to drop any references it has.  We
3501                  * just drop the ref we hold on that node and move on to the
3502                  * next slot in this level.
3503                  */
3504                 if (refs != 1) {
3505                         parent = path->nodes[*level];
3506                         root_owner = btrfs_header_owner(parent);
3507                         root_gen = btrfs_header_generation(parent);
3508                         path->slots[*level]++;
3509
3510                         ret = btrfs_free_extent(trans, root, bytenr,
3511                                                 blocksize, parent->start,
3512                                                 root_owner, root_gen,
3513                                                 *level - 1, 1);
3514                         BUG_ON(ret);
3515
3516                         atomic_inc(&root->fs_info->throttle_gen);
3517                         wake_up(&root->fs_info->transaction_throttle);
3518                         cond_resched();
3519
3520                         continue;
3521                 }
3522
3523                 /*
3524                  * we need to keep freeing things in the next level down.
3525                  * read the block and loop around to process it
3526                  */
3527                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3528                 WARN_ON(*level <= 0);
3529                 if (path->nodes[*level-1])
3530                         free_extent_buffer(path->nodes[*level-1]);
3531                 path->nodes[*level-1] = next;
3532                 *level = btrfs_header_level(next);
3533                 path->slots[*level] = 0;
3534                 cond_resched();
3535         }
3536 out:
3537         WARN_ON(*level < 0);
3538         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3539
3540         if (path->nodes[*level] == root->node) {
3541                 parent = path->nodes[*level];
3542                 bytenr = path->nodes[*level]->start;
3543         } else {
3544                 parent = path->nodes[*level + 1];
3545                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3546         }
3547
3548         blocksize = btrfs_level_size(root, *level);
3549         root_owner = btrfs_header_owner(parent);
3550         root_gen = btrfs_header_generation(parent);
3551
3552         /*
3553          * cleanup and free the reference on the last node
3554          * we processed
3555          */
3556         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3557                                   parent->start, root_owner, root_gen,
3558                                   *level, 1);
3559         free_extent_buffer(path->nodes[*level]);
3560         path->nodes[*level] = NULL;
3561
3562         *level += 1;
3563         BUG_ON(ret);
3564
3565         cond_resched();
3566         return 0;
3567 }
3568
3569 /*
3570  * helper function for drop_subtree, this function is similar to
3571  * walk_down_tree. The main difference is that it checks reference
3572  * counts while tree blocks are locked.
3573  */
3574 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3575                                       struct btrfs_root *root,
3576                                       struct btrfs_path *path, int *level)
3577 {
3578         struct extent_buffer *next;
3579         struct extent_buffer *cur;
3580         struct extent_buffer *parent;
3581         u64 bytenr;
3582         u64 ptr_gen;
3583         u32 blocksize;
3584         u32 refs;
3585         int ret;
3586
3587         cur = path->nodes[*level];
3588         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3589                                       &refs);
3590         BUG_ON(ret);
3591         if (refs > 1)
3592                 goto out;
3593
3594         while (*level >= 0) {
3595                 cur = path->nodes[*level];
3596                 if (*level == 0) {
3597                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3598                         BUG_ON(ret);
3599                         clean_tree_block(trans, root, cur);
3600                         break;
3601                 }
3602                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3603                         clean_tree_block(trans, root, cur);
3604                         break;
3605                 }
3606
3607                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3608                 blocksize = btrfs_level_size(root, *level - 1);
3609                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3610
3611                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3612                 btrfs_tree_lock(next);
3613                 btrfs_set_lock_blocking(next);
3614
3615                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3616                                               &refs);
3617                 BUG_ON(ret);
3618                 if (refs > 1) {
3619                         parent = path->nodes[*level];
3620                         ret = btrfs_free_extent(trans, root, bytenr,
3621                                         blocksize, parent->start,
3622                                         btrfs_header_owner(parent),
3623                                         btrfs_header_generation(parent),
3624                                         *level - 1, 1);
3625                         BUG_ON(ret);
3626                         path->slots[*level]++;
3627                         btrfs_tree_unlock(next);
3628                         free_extent_buffer(next);
3629                         continue;
3630                 }
3631
3632                 *level = btrfs_header_level(next);
3633                 path->nodes[*level] = next;
3634                 path->slots[*level] = 0;
3635                 path->locks[*level] = 1;
3636                 cond_resched();
3637         }
3638 out:
3639         parent = path->nodes[*level + 1];
3640         bytenr = path->nodes[*level]->start;
3641         blocksize = path->nodes[*level]->len;
3642
3643         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3644                         parent->start, btrfs_header_owner(parent),
3645                         btrfs_header_generation(parent), *level, 1);
3646         BUG_ON(ret);
3647
3648         if (path->locks[*level]) {
3649                 btrfs_tree_unlock(path->nodes[*level]);
3650                 path->locks[*level] = 0;
3651         }
3652         free_extent_buffer(path->nodes[*level]);
3653         path->nodes[*level] = NULL;
3654         *level += 1;
3655         cond_resched();
3656         return 0;
3657 }
3658
3659 /*
3660  * helper for dropping snapshots.  This walks back up the tree in the path
3661  * to find the first node higher up where we haven't yet gone through
3662  * all the slots
3663  */
3664 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3665                                  struct btrfs_root *root,
3666                                  struct btrfs_path *path,
3667                                  int *level, int max_level)
3668 {
3669         u64 root_owner;
3670         u64 root_gen;
3671         struct btrfs_root_item *root_item = &root->root_item;
3672         int i;
3673         int slot;
3674         int ret;
3675
3676         for (i = *level; i < max_level && path->nodes[i]; i++) {
3677                 slot = path->slots[i];
3678                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3679                         struct extent_buffer *node;
3680                         struct btrfs_disk_key disk_key;
3681
3682                         /*
3683                          * there is more work to do in this level.
3684                          * Update the drop_progress marker to reflect
3685                          * the work we've done so far, and then bump
3686                          * the slot number
3687                          */
3688                         node = path->nodes[i];
3689                         path->slots[i]++;
3690                         *level = i;
3691                         WARN_ON(*level == 0);
3692                         btrfs_node_key(node, &disk_key, path->slots[i]);
3693                         memcpy(&root_item->drop_progress,
3694                                &disk_key, sizeof(disk_key));
3695                         root_item->drop_level = i;
3696                         return 0;
3697                 } else {
3698                         struct extent_buffer *parent;
3699
3700                         /*
3701                          * this whole node is done, free our reference
3702                          * on it and go up one level
3703                          */
3704                         if (path->nodes[*level] == root->node)
3705                                 parent = path->nodes[*level];
3706                         else
3707                                 parent = path->nodes[*level + 1];
3708
3709                         root_owner = btrfs_header_owner(parent);
3710                         root_gen = btrfs_header_generation(parent);
3711
3712                         clean_tree_block(trans, root, path->nodes[*level]);
3713                         ret = btrfs_free_extent(trans, root,
3714                                                 path->nodes[*level]->start,
3715                                                 path->nodes[*level]->len,
3716                                                 parent->start, root_owner,
3717                                                 root_gen, *level, 1);
3718                         BUG_ON(ret);
3719                         if (path->locks[*level]) {
3720                                 btrfs_tree_unlock(path->nodes[*level]);
3721                                 path->locks[*level] = 0;
3722                         }
3723                         free_extent_buffer(path->nodes[*level]);
3724                         path->nodes[*level] = NULL;
3725                         *level = i + 1;
3726                 }
3727         }
3728         return 1;
3729 }
3730
3731 /*
3732  * drop the reference count on the tree rooted at 'snap'.  This traverses
3733  * the tree freeing any blocks that have a ref count of zero after being
3734  * decremented.
3735  */
3736 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3737                         *root)
3738 {
3739         int ret = 0;
3740         int wret;
3741         int level;
3742         struct btrfs_path *path;
3743         int i;
3744         int orig_level;
3745         int update_count;
3746         struct btrfs_root_item *root_item = &root->root_item;
3747
3748         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3749         path = btrfs_alloc_path();
3750         BUG_ON(!path);
3751
3752         level = btrfs_header_level(root->node);
3753         orig_level = level;
3754         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3755                 path->nodes[level] = root->node;
3756                 extent_buffer_get(root->node);
3757                 path->slots[level] = 0;
3758         } else {
3759                 struct btrfs_key key;
3760                 struct btrfs_disk_key found_key;
3761                 struct extent_buffer *node;
3762
3763                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3764                 level = root_item->drop_level;
3765                 path->lowest_level = level;
3766                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3767                 if (wret < 0) {
3768                         ret = wret;
3769                         goto out;
3770                 }
3771                 node = path->nodes[level];
3772                 btrfs_node_key(node, &found_key, path->slots[level]);
3773                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3774                                sizeof(found_key)));
3775                 /*
3776                  * unlock our path, this is safe because only this
3777                  * function is allowed to delete this snapshot
3778                  */
3779                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3780                         if (path->nodes[i] && path->locks[i]) {
3781                                 path->locks[i] = 0;
3782                                 btrfs_tree_unlock(path->nodes[i]);
3783                         }
3784                 }
3785         }
3786         while (1) {
3787                 unsigned long update;
3788                 wret = walk_down_tree(trans, root, path, &level);
3789                 if (wret > 0)
3790                         break;
3791                 if (wret < 0)
3792                         ret = wret;
3793
3794                 wret = walk_up_tree(trans, root, path, &level,
3795                                     BTRFS_MAX_LEVEL);
3796                 if (wret > 0)
3797                         break;
3798                 if (wret < 0)
3799                         ret = wret;
3800                 if (trans->transaction->in_commit) {
3801                         ret = -EAGAIN;
3802                         break;
3803                 }
3804                 atomic_inc(&root->fs_info->throttle_gen);
3805                 wake_up(&root->fs_info->transaction_throttle);
3806                 for (update_count = 0; update_count < 16; update_count++) {
3807                         update = trans->delayed_ref_updates;
3808                         trans->delayed_ref_updates = 0;
3809                         if (update)
3810                                 btrfs_run_delayed_refs(trans, root, update);
3811                         else
3812                                 break;
3813                 }
3814         }
3815         for (i = 0; i <= orig_level; i++) {
3816                 if (path->nodes[i]) {
3817                         free_extent_buffer(path->nodes[i]);
3818                         path->nodes[i] = NULL;
3819                 }
3820         }
3821 out:
3822         btrfs_free_path(path);
3823         return ret;
3824 }
3825
3826 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3827                         struct btrfs_root *root,
3828                         struct extent_buffer *node,
3829                         struct extent_buffer *parent)
3830 {
3831         struct btrfs_path *path;
3832         int level;
3833         int parent_level;
3834         int ret = 0;
3835         int wret;
3836
3837         path = btrfs_alloc_path();
3838         BUG_ON(!path);
3839
3840         btrfs_assert_tree_locked(parent);
3841         parent_level = btrfs_header_level(parent);
3842         extent_buffer_get(parent);
3843         path->nodes[parent_level] = parent;
3844         path->slots[parent_level] = btrfs_header_nritems(parent);
3845
3846         btrfs_assert_tree_locked(node);
3847         level = btrfs_header_level(node);
3848         extent_buffer_get(node);
3849         path->nodes[level] = node;
3850         path->slots[level] = 0;
3851
3852         while (1) {
3853                 wret = walk_down_subtree(trans, root, path, &level);
3854                 if (wret < 0)
3855                         ret = wret;
3856                 if (wret != 0)
3857                         break;
3858
3859                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3860                 if (wret < 0)
3861                         ret = wret;
3862                 if (wret != 0)
3863                         break;
3864         }
3865
3866         btrfs_free_path(path);
3867         return ret;
3868 }
3869
3870 static unsigned long calc_ra(unsigned long start, unsigned long last,
3871                              unsigned long nr)
3872 {
3873         return min(last, start + nr - 1);
3874 }
3875
3876 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3877                                          u64 len)
3878 {
3879         u64 page_start;
3880         u64 page_end;
3881         unsigned long first_index;
3882         unsigned long last_index;
3883         unsigned long i;
3884         struct page *page;
3885         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3886         struct file_ra_state *ra;
3887         struct btrfs_ordered_extent *ordered;
3888         unsigned int total_read = 0;
3889         unsigned int total_dirty = 0;
3890         int ret = 0;
3891
3892         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3893
3894         mutex_lock(&inode->i_mutex);
3895         first_index = start >> PAGE_CACHE_SHIFT;
3896         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3897
3898         /* make sure the dirty trick played by the caller work */
3899         ret = invalidate_inode_pages2_range(inode->i_mapping,
3900                                             first_index, last_index);
3901         if (ret)
3902                 goto out_unlock;
3903
3904         file_ra_state_init(ra, inode->i_mapping);
3905
3906         for (i = first_index ; i <= last_index; i++) {
3907                 if (total_read % ra->ra_pages == 0) {
3908                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3909                                        calc_ra(i, last_index, ra->ra_pages));
3910                 }
3911                 total_read++;
3912 again:
3913                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3914                         BUG_ON(1);
3915                 page = grab_cache_page(inode->i_mapping, i);
3916                 if (!page) {
3917                         ret = -ENOMEM;
3918                         goto out_unlock;
3919                 }
3920                 if (!PageUptodate(page)) {
3921                         btrfs_readpage(NULL, page);
3922                         lock_page(page);
3923                         if (!PageUptodate(page)) {
3924                                 unlock_page(page);
3925                                 page_cache_release(page);
3926                                 ret = -EIO;
3927                                 goto out_unlock;
3928                         }
3929                 }
3930                 wait_on_page_writeback(page);
3931
3932                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3933                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3934                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3935
3936                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3937                 if (ordered) {
3938                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3939                         unlock_page(page);
3940                         page_cache_release(page);
3941                         btrfs_start_ordered_extent(inode, ordered, 1);
3942                         btrfs_put_ordered_extent(ordered);
3943                         goto again;
3944                 }
3945                 set_page_extent_mapped(page);
3946
3947                 if (i == first_index)
3948                         set_extent_bits(io_tree, page_start, page_end,
3949                                         EXTENT_BOUNDARY, GFP_NOFS);
3950                 btrfs_set_extent_delalloc(inode, page_start, page_end);
3951
3952                 set_page_dirty(page);
3953                 total_dirty++;
3954
3955                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3956                 unlock_page(page);
3957                 page_cache_release(page);
3958         }
3959
3960 out_unlock:
3961         kfree(ra);
3962         mutex_unlock(&inode->i_mutex);
3963         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3964         return ret;
3965 }
3966
3967 static noinline int relocate_data_extent(struct inode *reloc_inode,
3968                                          struct btrfs_key *extent_key,
3969                                          u64 offset)
3970 {
3971         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3972         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3973         struct extent_map *em;
3974         u64 start = extent_key->objectid - offset;
3975         u64 end = start + extent_key->offset - 1;
3976
3977         em = alloc_extent_map(GFP_NOFS);
3978         BUG_ON(!em || IS_ERR(em));
3979
3980         em->start = start;
3981         em->len = extent_key->offset;
3982         em->block_len = extent_key->offset;
3983         em->block_start = extent_key->objectid;
3984         em->bdev = root->fs_info->fs_devices->latest_bdev;
3985         set_bit(EXTENT_FLAG_PINNED, &em->flags);
3986
3987         /* setup extent map to cheat btrfs_readpage */
3988         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3989         while (1) {
3990                 int ret;
3991                 spin_lock(&em_tree->lock);
3992                 ret = add_extent_mapping(em_tree, em);
3993                 spin_unlock(&em_tree->lock);
3994                 if (ret != -EEXIST) {
3995                         free_extent_map(em);
3996                         break;
3997                 }
3998                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
3999         }
4000         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4001
4002         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4003 }
4004
4005 struct btrfs_ref_path {
4006         u64 extent_start;
4007         u64 nodes[BTRFS_MAX_LEVEL];
4008         u64 root_objectid;
4009         u64 root_generation;
4010         u64 owner_objectid;
4011         u32 num_refs;
4012         int lowest_level;
4013         int current_level;
4014         int shared_level;
4015
4016         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4017         u64 new_nodes[BTRFS_MAX_LEVEL];
4018 };
4019
4020 struct disk_extent {
4021         u64 ram_bytes;
4022         u64 disk_bytenr;
4023         u64 disk_num_bytes;
4024         u64 offset;
4025         u64 num_bytes;
4026         u8 compression;
4027         u8 encryption;
4028         u16 other_encoding;
4029 };
4030
4031 static int is_cowonly_root(u64 root_objectid)
4032 {
4033         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4034             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4035             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4036             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4037             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4038             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4039                 return 1;
4040         return 0;
4041 }
4042
4043 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4044                                     struct btrfs_root *extent_root,
4045                                     struct btrfs_ref_path *ref_path,
4046                                     int first_time)
4047 {
4048         struct extent_buffer *leaf;
4049         struct btrfs_path *path;
4050         struct btrfs_extent_ref *ref;
4051         struct btrfs_key key;
4052         struct btrfs_key found_key;
4053         u64 bytenr;
4054         u32 nritems;
4055         int level;
4056         int ret = 1;
4057
4058         path = btrfs_alloc_path();
4059         if (!path)
4060                 return -ENOMEM;
4061
4062         if (first_time) {
4063                 ref_path->lowest_level = -1;
4064                 ref_path->current_level = -1;
4065                 ref_path->shared_level = -1;
4066                 goto walk_up;
4067         }
4068 walk_down:
4069         level = ref_path->current_level - 1;
4070         while (level >= -1) {
4071                 u64 parent;
4072                 if (level < ref_path->lowest_level)
4073                         break;
4074
4075                 if (level >= 0)
4076                         bytenr = ref_path->nodes[level];
4077                 else
4078                         bytenr = ref_path->extent_start;
4079                 BUG_ON(bytenr == 0);
4080
4081                 parent = ref_path->nodes[level + 1];
4082                 ref_path->nodes[level + 1] = 0;
4083                 ref_path->current_level = level;
4084                 BUG_ON(parent == 0);
4085
4086                 key.objectid = bytenr;
4087                 key.offset = parent + 1;
4088                 key.type = BTRFS_EXTENT_REF_KEY;
4089
4090                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4091                 if (ret < 0)
4092                         goto out;
4093                 BUG_ON(ret == 0);
4094
4095                 leaf = path->nodes[0];
4096                 nritems = btrfs_header_nritems(leaf);
4097                 if (path->slots[0] >= nritems) {
4098                         ret = btrfs_next_leaf(extent_root, path);
4099                         if (ret < 0)
4100                                 goto out;
4101                         if (ret > 0)
4102                                 goto next;
4103                         leaf = path->nodes[0];
4104                 }
4105
4106                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4107                 if (found_key.objectid == bytenr &&
4108                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4109                         if (level < ref_path->shared_level)
4110                                 ref_path->shared_level = level;
4111                         goto found;
4112                 }
4113 next:
4114                 level--;
4115                 btrfs_release_path(extent_root, path);
4116                 cond_resched();
4117         }
4118         /* reached lowest level */
4119         ret = 1;
4120         goto out;
4121 walk_up:
4122         level = ref_path->current_level;
4123         while (level < BTRFS_MAX_LEVEL - 1) {
4124                 u64 ref_objectid;
4125
4126                 if (level >= 0)
4127                         bytenr = ref_path->nodes[level];
4128                 else
4129                         bytenr = ref_path->extent_start;
4130
4131                 BUG_ON(bytenr == 0);
4132
4133                 key.objectid = bytenr;
4134                 key.offset = 0;
4135                 key.type = BTRFS_EXTENT_REF_KEY;
4136
4137                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4138                 if (ret < 0)
4139                         goto out;
4140
4141                 leaf = path->nodes[0];
4142                 nritems = btrfs_header_nritems(leaf);
4143                 if (path->slots[0] >= nritems) {
4144                         ret = btrfs_next_leaf(extent_root, path);
4145                         if (ret < 0)
4146                                 goto out;
4147                         if (ret > 0) {
4148                                 /* the extent was freed by someone */
4149                                 if (ref_path->lowest_level == level)
4150                                         goto out;
4151                                 btrfs_release_path(extent_root, path);
4152                                 goto walk_down;
4153                         }
4154                         leaf = path->nodes[0];
4155                 }
4156
4157                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4158                 if (found_key.objectid != bytenr ||
4159                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4160                         /* the extent was freed by someone */
4161                         if (ref_path->lowest_level == level) {
4162                                 ret = 1;
4163                                 goto out;
4164                         }
4165                         btrfs_release_path(extent_root, path);
4166                         goto walk_down;
4167                 }
4168 found:
4169                 ref = btrfs_item_ptr(leaf, path->slots[0],
4170                                 struct btrfs_extent_ref);
4171                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4172                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4173                         if (first_time) {
4174                                 level = (int)ref_objectid;
4175                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4176                                 ref_path->lowest_level = level;
4177                                 ref_path->current_level = level;
4178                                 ref_path->nodes[level] = bytenr;
4179                         } else {
4180                                 WARN_ON(ref_objectid != level);
4181                         }
4182                 } else {
4183                         WARN_ON(level != -1);
4184                 }
4185                 first_time = 0;
4186
4187                 if (ref_path->lowest_level == level) {
4188                         ref_path->owner_objectid = ref_objectid;
4189                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4190                 }
4191
4192                 /*
4193                  * the block is tree root or the block isn't in reference
4194                  * counted tree.
4195                  */
4196                 if (found_key.objectid == found_key.offset ||
4197                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4198                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4199                         ref_path->root_generation =
4200                                 btrfs_ref_generation(leaf, ref);
4201                         if (level < 0) {
4202                                 /* special reference from the tree log */
4203                                 ref_path->nodes[0] = found_key.offset;
4204                                 ref_path->current_level = 0;
4205                         }
4206                         ret = 0;
4207                         goto out;
4208                 }
4209
4210                 level++;
4211                 BUG_ON(ref_path->nodes[level] != 0);
4212                 ref_path->nodes[level] = found_key.offset;
4213                 ref_path->current_level = level;
4214
4215                 /*
4216                  * the reference was created in the running transaction,
4217                  * no need to continue walking up.
4218                  */
4219                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4220                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4221                         ref_path->root_generation =
4222                                 btrfs_ref_generation(leaf, ref);
4223                         ret = 0;
4224                         goto out;
4225                 }
4226
4227                 btrfs_release_path(extent_root, path);
4228                 cond_resched();
4229         }
4230         /* reached max tree level, but no tree root found. */
4231         BUG();
4232 out:
4233         btrfs_free_path(path);
4234         return ret;
4235 }
4236
4237 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4238                                 struct btrfs_root *extent_root,
4239                                 struct btrfs_ref_path *ref_path,
4240                                 u64 extent_start)
4241 {
4242         memset(ref_path, 0, sizeof(*ref_path));
4243         ref_path->extent_start = extent_start;
4244
4245         return __next_ref_path(trans, extent_root, ref_path, 1);
4246 }
4247
4248 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4249                                struct btrfs_root *extent_root,
4250                                struct btrfs_ref_path *ref_path)
4251 {
4252         return __next_ref_path(trans, extent_root, ref_path, 0);
4253 }
4254
4255 static noinline int get_new_locations(struct inode *reloc_inode,
4256                                       struct btrfs_key *extent_key,
4257                                       u64 offset, int no_fragment,
4258                                       struct disk_extent **extents,
4259                                       int *nr_extents)
4260 {
4261         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4262         struct btrfs_path *path;
4263         struct btrfs_file_extent_item *fi;
4264         struct extent_buffer *leaf;
4265         struct disk_extent *exts = *extents;
4266         struct btrfs_key found_key;
4267         u64 cur_pos;
4268         u64 last_byte;
4269         u32 nritems;
4270         int nr = 0;
4271         int max = *nr_extents;
4272         int ret;
4273
4274         WARN_ON(!no_fragment && *extents);
4275         if (!exts) {
4276                 max = 1;
4277                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4278                 if (!exts)
4279                         return -ENOMEM;
4280         }
4281
4282         path = btrfs_alloc_path();
4283         BUG_ON(!path);
4284
4285         cur_pos = extent_key->objectid - offset;
4286         last_byte = extent_key->objectid + extent_key->offset;
4287         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4288                                        cur_pos, 0);
4289         if (ret < 0)
4290                 goto out;
4291         if (ret > 0) {
4292                 ret = -ENOENT;
4293                 goto out;
4294         }
4295
4296         while (1) {
4297                 leaf = path->nodes[0];
4298                 nritems = btrfs_header_nritems(leaf);
4299                 if (path->slots[0] >= nritems) {
4300                         ret = btrfs_next_leaf(root, path);
4301                         if (ret < 0)
4302                                 goto out;
4303                         if (ret > 0)
4304                                 break;
4305                         leaf = path->nodes[0];
4306                 }
4307
4308                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4309                 if (found_key.offset != cur_pos ||
4310                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4311                     found_key.objectid != reloc_inode->i_ino)
4312                         break;
4313
4314                 fi = btrfs_item_ptr(leaf, path->slots[0],
4315                                     struct btrfs_file_extent_item);
4316                 if (btrfs_file_extent_type(leaf, fi) !=
4317                     BTRFS_FILE_EXTENT_REG ||
4318                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4319                         break;
4320
4321                 if (nr == max) {
4322                         struct disk_extent *old = exts;
4323                         max *= 2;
4324                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4325                         memcpy(exts, old, sizeof(*exts) * nr);
4326                         if (old != *extents)
4327                                 kfree(old);
4328                 }
4329
4330                 exts[nr].disk_bytenr =
4331                         btrfs_file_extent_disk_bytenr(leaf, fi);
4332                 exts[nr].disk_num_bytes =
4333                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4334                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4335                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4336                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4337                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4338                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4339                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4340                                                                            fi);
4341                 BUG_ON(exts[nr].offset > 0);
4342                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4343                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4344
4345                 cur_pos += exts[nr].num_bytes;
4346                 nr++;
4347
4348                 if (cur_pos + offset >= last_byte)
4349                         break;
4350
4351                 if (no_fragment) {
4352                         ret = 1;
4353                         goto out;
4354                 }
4355                 path->slots[0]++;
4356         }
4357
4358         BUG_ON(cur_pos + offset > last_byte);
4359         if (cur_pos + offset < last_byte) {
4360                 ret = -ENOENT;
4361                 goto out;
4362         }
4363         ret = 0;
4364 out:
4365         btrfs_free_path(path);
4366         if (ret) {
4367                 if (exts != *extents)
4368                         kfree(exts);
4369         } else {
4370                 *extents = exts;
4371                 *nr_extents = nr;
4372         }
4373         return ret;
4374 }
4375
4376 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4377                                         struct btrfs_root *root,
4378                                         struct btrfs_path *path,
4379                                         struct btrfs_key *extent_key,
4380                                         struct btrfs_key *leaf_key,
4381                                         struct btrfs_ref_path *ref_path,
4382                                         struct disk_extent *new_extents,
4383                                         int nr_extents)
4384 {
4385         struct extent_buffer *leaf;
4386         struct btrfs_file_extent_item *fi;
4387         struct inode *inode = NULL;
4388         struct btrfs_key key;
4389         u64 lock_start = 0;
4390         u64 lock_end = 0;
4391         u64 num_bytes;
4392         u64 ext_offset;
4393         u64 search_end = (u64)-1;
4394         u32 nritems;
4395         int nr_scaned = 0;
4396         int extent_locked = 0;
4397         int extent_type;
4398         int ret;
4399
4400         memcpy(&key, leaf_key, sizeof(key));
4401         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4402                 if (key.objectid < ref_path->owner_objectid ||
4403                     (key.objectid == ref_path->owner_objectid &&
4404                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4405                         key.objectid = ref_path->owner_objectid;
4406                         key.type = BTRFS_EXTENT_DATA_KEY;
4407                         key.offset = 0;
4408                 }
4409         }
4410
4411         while (1) {
4412                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4413                 if (ret < 0)
4414                         goto out;
4415
4416                 leaf = path->nodes[0];
4417                 nritems = btrfs_header_nritems(leaf);
4418 next:
4419                 if (extent_locked && ret > 0) {
4420                         /*
4421                          * the file extent item was modified by someone
4422                          * before the extent got locked.
4423                          */
4424                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4425                                       lock_end, GFP_NOFS);
4426                         extent_locked = 0;
4427                 }
4428
4429                 if (path->slots[0] >= nritems) {
4430                         if (++nr_scaned > 2)
4431                                 break;
4432
4433                         BUG_ON(extent_locked);
4434                         ret = btrfs_next_leaf(root, path);
4435                         if (ret < 0)
4436                                 goto out;
4437                         if (ret > 0)
4438                                 break;
4439                         leaf = path->nodes[0];
4440                         nritems = btrfs_header_nritems(leaf);
4441                 }
4442
4443                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4444
4445                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4446                         if ((key.objectid > ref_path->owner_objectid) ||
4447                             (key.objectid == ref_path->owner_objectid &&
4448                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4449                             key.offset >= search_end)
4450                                 break;
4451                 }
4452
4453                 if (inode && key.objectid != inode->i_ino) {
4454                         BUG_ON(extent_locked);
4455                         btrfs_release_path(root, path);
4456                         mutex_unlock(&inode->i_mutex);
4457                         iput(inode);
4458                         inode = NULL;
4459                         continue;
4460                 }
4461
4462                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4463                         path->slots[0]++;
4464                         ret = 1;
4465                         goto next;
4466                 }
4467                 fi = btrfs_item_ptr(leaf, path->slots[0],
4468                                     struct btrfs_file_extent_item);
4469                 extent_type = btrfs_file_extent_type(leaf, fi);
4470                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4471                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4472                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4473                      extent_key->objectid)) {
4474                         path->slots[0]++;
4475                         ret = 1;
4476                         goto next;
4477                 }
4478
4479                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4480                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4481
4482                 if (search_end == (u64)-1) {
4483                         search_end = key.offset - ext_offset +
4484                                 btrfs_file_extent_ram_bytes(leaf, fi);
4485                 }
4486
4487                 if (!extent_locked) {
4488                         lock_start = key.offset;
4489                         lock_end = lock_start + num_bytes - 1;
4490                 } else {
4491                         if (lock_start > key.offset ||
4492                             lock_end + 1 < key.offset + num_bytes) {
4493                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4494                                               lock_start, lock_end, GFP_NOFS);
4495                                 extent_locked = 0;
4496                         }
4497                 }
4498
4499                 if (!inode) {
4500                         btrfs_release_path(root, path);
4501
4502                         inode = btrfs_iget_locked(root->fs_info->sb,
4503                                                   key.objectid, root);
4504                         if (inode->i_state & I_NEW) {
4505                                 BTRFS_I(inode)->root = root;
4506                                 BTRFS_I(inode)->location.objectid =
4507                                         key.objectid;
4508                                 BTRFS_I(inode)->location.type =
4509                                         BTRFS_INODE_ITEM_KEY;
4510                                 BTRFS_I(inode)->location.offset = 0;
4511                                 btrfs_read_locked_inode(inode);
4512                                 unlock_new_inode(inode);
4513                         }
4514                         /*
4515                          * some code call btrfs_commit_transaction while
4516                          * holding the i_mutex, so we can't use mutex_lock
4517                          * here.
4518                          */
4519                         if (is_bad_inode(inode) ||
4520                             !mutex_trylock(&inode->i_mutex)) {
4521                                 iput(inode);
4522                                 inode = NULL;
4523                                 key.offset = (u64)-1;
4524                                 goto skip;
4525                         }
4526                 }
4527
4528                 if (!extent_locked) {
4529                         struct btrfs_ordered_extent *ordered;
4530
4531                         btrfs_release_path(root, path);
4532
4533                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4534                                     lock_end, GFP_NOFS);
4535                         ordered = btrfs_lookup_first_ordered_extent(inode,
4536                                                                     lock_end);
4537                         if (ordered &&
4538                             ordered->file_offset <= lock_end &&
4539                             ordered->file_offset + ordered->len > lock_start) {
4540                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4541                                               lock_start, lock_end, GFP_NOFS);
4542                                 btrfs_start_ordered_extent(inode, ordered, 1);
4543                                 btrfs_put_ordered_extent(ordered);
4544                                 key.offset += num_bytes;
4545                                 goto skip;
4546                         }
4547                         if (ordered)
4548                                 btrfs_put_ordered_extent(ordered);
4549
4550                         extent_locked = 1;
4551                         continue;
4552                 }
4553
4554                 if (nr_extents == 1) {
4555                         /* update extent pointer in place */
4556                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4557                                                 new_extents[0].disk_bytenr);
4558                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4559                                                 new_extents[0].disk_num_bytes);
4560                         btrfs_mark_buffer_dirty(leaf);
4561
4562                         btrfs_drop_extent_cache(inode, key.offset,
4563                                                 key.offset + num_bytes - 1, 0);
4564
4565                         ret = btrfs_inc_extent_ref(trans, root,
4566                                                 new_extents[0].disk_bytenr,
4567                                                 new_extents[0].disk_num_bytes,
4568                                                 leaf->start,
4569                                                 root->root_key.objectid,
4570                                                 trans->transid,
4571                                                 key.objectid);
4572                         BUG_ON(ret);
4573
4574                         ret = btrfs_free_extent(trans, root,
4575                                                 extent_key->objectid,
4576                                                 extent_key->offset,
4577                                                 leaf->start,
4578                                                 btrfs_header_owner(leaf),
4579                                                 btrfs_header_generation(leaf),
4580                                                 key.objectid, 0);
4581                         BUG_ON(ret);
4582
4583                         btrfs_release_path(root, path);
4584                         key.offset += num_bytes;
4585                 } else {
4586                         BUG_ON(1);
4587 #if 0
4588                         u64 alloc_hint;
4589                         u64 extent_len;
4590                         int i;
4591                         /*
4592                          * drop old extent pointer at first, then insert the
4593                          * new pointers one bye one
4594                          */
4595                         btrfs_release_path(root, path);
4596                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4597                                                  key.offset + num_bytes,
4598                                                  key.offset, &alloc_hint);
4599                         BUG_ON(ret);
4600
4601                         for (i = 0; i < nr_extents; i++) {
4602                                 if (ext_offset >= new_extents[i].num_bytes) {
4603                                         ext_offset -= new_extents[i].num_bytes;
4604                                         continue;
4605                                 }
4606                                 extent_len = min(new_extents[i].num_bytes -
4607                                                  ext_offset, num_bytes);
4608
4609                                 ret = btrfs_insert_empty_item(trans, root,
4610                                                               path, &key,
4611                                                               sizeof(*fi));
4612                                 BUG_ON(ret);
4613
4614                                 leaf = path->nodes[0];
4615                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4616                                                 struct btrfs_file_extent_item);
4617                                 btrfs_set_file_extent_generation(leaf, fi,
4618                                                         trans->transid);
4619                                 btrfs_set_file_extent_type(leaf, fi,
4620                                                         BTRFS_FILE_EXTENT_REG);
4621                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4622                                                 new_extents[i].disk_bytenr);
4623                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4624                                                 new_extents[i].disk_num_bytes);
4625                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4626                                                 new_extents[i].ram_bytes);
4627
4628                                 btrfs_set_file_extent_compression(leaf, fi,
4629                                                 new_extents[i].compression);
4630                                 btrfs_set_file_extent_encryption(leaf, fi,
4631                                                 new_extents[i].encryption);
4632                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4633                                                 new_extents[i].other_encoding);
4634
4635                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4636                                                         extent_len);
4637                                 ext_offset += new_extents[i].offset;
4638                                 btrfs_set_file_extent_offset(leaf, fi,
4639                                                         ext_offset);
4640                                 btrfs_mark_buffer_dirty(leaf);
4641
4642                                 btrfs_drop_extent_cache(inode, key.offset,
4643                                                 key.offset + extent_len - 1, 0);
4644
4645                                 ret = btrfs_inc_extent_ref(trans, root,
4646                                                 new_extents[i].disk_bytenr,
4647                                                 new_extents[i].disk_num_bytes,
4648                                                 leaf->start,
4649                                                 root->root_key.objectid,
4650                                                 trans->transid, key.objectid);
4651                                 BUG_ON(ret);
4652                                 btrfs_release_path(root, path);
4653
4654                                 inode_add_bytes(inode, extent_len);
4655
4656                                 ext_offset = 0;
4657                                 num_bytes -= extent_len;
4658                                 key.offset += extent_len;
4659
4660                                 if (num_bytes == 0)
4661                                         break;
4662                         }
4663                         BUG_ON(i >= nr_extents);
4664 #endif
4665                 }
4666
4667                 if (extent_locked) {
4668                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4669                                       lock_end, GFP_NOFS);
4670                         extent_locked = 0;
4671                 }
4672 skip:
4673                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4674                     key.offset >= search_end)
4675                         break;
4676
4677                 cond_resched();
4678         }
4679         ret = 0;
4680 out:
4681         btrfs_release_path(root, path);
4682         if (inode) {
4683                 mutex_unlock(&inode->i_mutex);
4684                 if (extent_locked) {
4685                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4686                                       lock_end, GFP_NOFS);
4687                 }
4688                 iput(inode);
4689         }
4690         return ret;
4691 }
4692
4693 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4694                                struct btrfs_root *root,
4695                                struct extent_buffer *buf, u64 orig_start)
4696 {
4697         int level;
4698         int ret;
4699
4700         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4701         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4702
4703         level = btrfs_header_level(buf);
4704         if (level == 0) {
4705                 struct btrfs_leaf_ref *ref;
4706                 struct btrfs_leaf_ref *orig_ref;
4707
4708                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4709                 if (!orig_ref)
4710                         return -ENOENT;
4711
4712                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4713                 if (!ref) {
4714                         btrfs_free_leaf_ref(root, orig_ref);
4715                         return -ENOMEM;
4716                 }
4717
4718                 ref->nritems = orig_ref->nritems;
4719                 memcpy(ref->extents, orig_ref->extents,
4720                         sizeof(ref->extents[0]) * ref->nritems);
4721
4722                 btrfs_free_leaf_ref(root, orig_ref);
4723
4724                 ref->root_gen = trans->transid;
4725                 ref->bytenr = buf->start;
4726                 ref->owner = btrfs_header_owner(buf);
4727                 ref->generation = btrfs_header_generation(buf);
4728
4729                 ret = btrfs_add_leaf_ref(root, ref, 0);
4730                 WARN_ON(ret);
4731                 btrfs_free_leaf_ref(root, ref);
4732         }
4733         return 0;
4734 }
4735
4736 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4737                                         struct extent_buffer *leaf,
4738                                         struct btrfs_block_group_cache *group,
4739                                         struct btrfs_root *target_root)
4740 {
4741         struct btrfs_key key;
4742         struct inode *inode = NULL;
4743         struct btrfs_file_extent_item *fi;
4744         u64 num_bytes;
4745         u64 skip_objectid = 0;
4746         u32 nritems;
4747         u32 i;
4748
4749         nritems = btrfs_header_nritems(leaf);
4750         for (i = 0; i < nritems; i++) {
4751                 btrfs_item_key_to_cpu(leaf, &key, i);
4752                 if (key.objectid == skip_objectid ||
4753                     key.type != BTRFS_EXTENT_DATA_KEY)
4754                         continue;
4755                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4756                 if (btrfs_file_extent_type(leaf, fi) ==
4757                     BTRFS_FILE_EXTENT_INLINE)
4758                         continue;
4759                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4760                         continue;
4761                 if (!inode || inode->i_ino != key.objectid) {
4762                         iput(inode);
4763                         inode = btrfs_ilookup(target_root->fs_info->sb,
4764                                               key.objectid, target_root, 1);
4765                 }
4766                 if (!inode) {
4767                         skip_objectid = key.objectid;
4768                         continue;
4769                 }
4770                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4771
4772                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4773                             key.offset + num_bytes - 1, GFP_NOFS);
4774                 btrfs_drop_extent_cache(inode, key.offset,
4775                                         key.offset + num_bytes - 1, 1);
4776                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4777                               key.offset + num_bytes - 1, GFP_NOFS);
4778                 cond_resched();
4779         }
4780         iput(inode);
4781         return 0;
4782 }
4783
4784 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4785                                         struct btrfs_root *root,
4786                                         struct extent_buffer *leaf,
4787                                         struct btrfs_block_group_cache *group,
4788                                         struct inode *reloc_inode)
4789 {
4790         struct btrfs_key key;
4791         struct btrfs_key extent_key;
4792         struct btrfs_file_extent_item *fi;
4793         struct btrfs_leaf_ref *ref;
4794         struct disk_extent *new_extent;
4795         u64 bytenr;
4796         u64 num_bytes;
4797         u32 nritems;
4798         u32 i;
4799         int ext_index;
4800         int nr_extent;
4801         int ret;
4802
4803         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4804         BUG_ON(!new_extent);
4805
4806         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4807         BUG_ON(!ref);
4808
4809         ext_index = -1;
4810         nritems = btrfs_header_nritems(leaf);
4811         for (i = 0; i < nritems; i++) {
4812                 btrfs_item_key_to_cpu(leaf, &key, i);
4813                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4814                         continue;
4815                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4816                 if (btrfs_file_extent_type(leaf, fi) ==
4817                     BTRFS_FILE_EXTENT_INLINE)
4818                         continue;
4819                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4820                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4821                 if (bytenr == 0)
4822                         continue;
4823
4824                 ext_index++;
4825                 if (bytenr >= group->key.objectid + group->key.offset ||
4826                     bytenr + num_bytes <= group->key.objectid)
4827                         continue;
4828
4829                 extent_key.objectid = bytenr;
4830                 extent_key.offset = num_bytes;
4831                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4832                 nr_extent = 1;
4833                 ret = get_new_locations(reloc_inode, &extent_key,
4834                                         group->key.objectid, 1,
4835                                         &new_extent, &nr_extent);
4836                 if (ret > 0)
4837                         continue;
4838                 BUG_ON(ret < 0);
4839
4840                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4841                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4842                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4843                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4844
4845                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4846                                                 new_extent->disk_bytenr);
4847                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4848                                                 new_extent->disk_num_bytes);
4849                 btrfs_mark_buffer_dirty(leaf);
4850
4851                 ret = btrfs_inc_extent_ref(trans, root,
4852                                         new_extent->disk_bytenr,
4853                                         new_extent->disk_num_bytes,
4854                                         leaf->start,
4855                                         root->root_key.objectid,
4856                                         trans->transid, key.objectid);
4857                 BUG_ON(ret);
4858
4859                 ret = btrfs_free_extent(trans, root,
4860                                         bytenr, num_bytes, leaf->start,
4861                                         btrfs_header_owner(leaf),
4862                                         btrfs_header_generation(leaf),
4863                                         key.objectid, 0);
4864                 BUG_ON(ret);
4865                 cond_resched();
4866         }
4867         kfree(new_extent);
4868         BUG_ON(ext_index + 1 != ref->nritems);
4869         btrfs_free_leaf_ref(root, ref);
4870         return 0;
4871 }
4872
4873 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4874                           struct btrfs_root *root)
4875 {
4876         struct btrfs_root *reloc_root;
4877         int ret;
4878
4879         if (root->reloc_root) {
4880                 reloc_root = root->reloc_root;
4881                 root->reloc_root = NULL;
4882                 list_add(&reloc_root->dead_list,
4883                          &root->fs_info->dead_reloc_roots);
4884
4885                 btrfs_set_root_bytenr(&reloc_root->root_item,
4886                                       reloc_root->node->start);
4887                 btrfs_set_root_level(&root->root_item,
4888                                      btrfs_header_level(reloc_root->node));
4889                 memset(&reloc_root->root_item.drop_progress, 0,
4890                         sizeof(struct btrfs_disk_key));
4891                 reloc_root->root_item.drop_level = 0;
4892
4893                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4894                                         &reloc_root->root_key,
4895                                         &reloc_root->root_item);
4896                 BUG_ON(ret);
4897         }
4898         return 0;
4899 }
4900
4901 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4902 {
4903         struct btrfs_trans_handle *trans;
4904         struct btrfs_root *reloc_root;
4905         struct btrfs_root *prev_root = NULL;
4906         struct list_head dead_roots;
4907         int ret;
4908         unsigned long nr;
4909
4910         INIT_LIST_HEAD(&dead_roots);
4911         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4912
4913         while (!list_empty(&dead_roots)) {
4914                 reloc_root = list_entry(dead_roots.prev,
4915                                         struct btrfs_root, dead_list);
4916                 list_del_init(&reloc_root->dead_list);
4917
4918                 BUG_ON(reloc_root->commit_root != NULL);
4919                 while (1) {
4920                         trans = btrfs_join_transaction(root, 1);
4921                         BUG_ON(!trans);
4922
4923                         mutex_lock(&root->fs_info->drop_mutex);
4924                         ret = btrfs_drop_snapshot(trans, reloc_root);
4925                         if (ret != -EAGAIN)
4926                                 break;
4927                         mutex_unlock(&root->fs_info->drop_mutex);
4928
4929                         nr = trans->blocks_used;
4930                         ret = btrfs_end_transaction(trans, root);
4931                         BUG_ON(ret);
4932                         btrfs_btree_balance_dirty(root, nr);
4933                 }
4934
4935                 free_extent_buffer(reloc_root->node);
4936
4937                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4938                                      &reloc_root->root_key);
4939                 BUG_ON(ret);
4940                 mutex_unlock(&root->fs_info->drop_mutex);
4941
4942                 nr = trans->blocks_used;
4943                 ret = btrfs_end_transaction(trans, root);
4944                 BUG_ON(ret);
4945                 btrfs_btree_balance_dirty(root, nr);
4946
4947                 kfree(prev_root);
4948                 prev_root = reloc_root;
4949         }
4950         if (prev_root) {
4951                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4952                 kfree(prev_root);
4953         }
4954         return 0;
4955 }
4956
4957 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4958 {
4959         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4960         return 0;
4961 }
4962
4963 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4964 {
4965         struct btrfs_root *reloc_root;
4966         struct btrfs_trans_handle *trans;
4967         struct btrfs_key location;
4968         int found;
4969         int ret;
4970
4971         mutex_lock(&root->fs_info->tree_reloc_mutex);
4972         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4973         BUG_ON(ret);
4974         found = !list_empty(&root->fs_info->dead_reloc_roots);
4975         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4976
4977         if (found) {
4978                 trans = btrfs_start_transaction(root, 1);
4979                 BUG_ON(!trans);
4980                 ret = btrfs_commit_transaction(trans, root);
4981                 BUG_ON(ret);
4982         }
4983
4984         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4985         location.offset = (u64)-1;
4986         location.type = BTRFS_ROOT_ITEM_KEY;
4987
4988         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4989         BUG_ON(!reloc_root);
4990         btrfs_orphan_cleanup(reloc_root);
4991         return 0;
4992 }
4993
4994 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
4995                                     struct btrfs_root *root)
4996 {
4997         struct btrfs_root *reloc_root;
4998         struct extent_buffer *eb;
4999         struct btrfs_root_item *root_item;
5000         struct btrfs_key root_key;
5001         int ret;
5002
5003         BUG_ON(!root->ref_cows);
5004         if (root->reloc_root)
5005                 return 0;
5006
5007         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5008         BUG_ON(!root_item);
5009
5010         ret = btrfs_copy_root(trans, root, root->commit_root,
5011                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5012         BUG_ON(ret);
5013
5014         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5015         root_key.offset = root->root_key.objectid;
5016         root_key.type = BTRFS_ROOT_ITEM_KEY;
5017
5018         memcpy(root_item, &root->root_item, sizeof(root_item));
5019         btrfs_set_root_refs(root_item, 0);
5020         btrfs_set_root_bytenr(root_item, eb->start);
5021         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5022         btrfs_set_root_generation(root_item, trans->transid);
5023
5024         btrfs_tree_unlock(eb);
5025         free_extent_buffer(eb);
5026
5027         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5028                                 &root_key, root_item);
5029         BUG_ON(ret);
5030         kfree(root_item);
5031
5032         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5033                                                  &root_key);
5034         BUG_ON(!reloc_root);
5035         reloc_root->last_trans = trans->transid;
5036         reloc_root->commit_root = NULL;
5037         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5038
5039         root->reloc_root = reloc_root;
5040         return 0;
5041 }
5042
5043 /*
5044  * Core function of space balance.
5045  *
5046  * The idea is using reloc trees to relocate tree blocks in reference
5047  * counted roots. There is one reloc tree for each subvol, and all
5048  * reloc trees share same root key objectid. Reloc trees are snapshots
5049  * of the latest committed roots of subvols (root->commit_root).
5050  *
5051  * To relocate a tree block referenced by a subvol, there are two steps.
5052  * COW the block through subvol's reloc tree, then update block pointer
5053  * in the subvol to point to the new block. Since all reloc trees share
5054  * same root key objectid, doing special handing for tree blocks owned
5055  * by them is easy. Once a tree block has been COWed in one reloc tree,
5056  * we can use the resulting new block directly when the same block is
5057  * required to COW again through other reloc trees. By this way, relocated
5058  * tree blocks are shared between reloc trees, so they are also shared
5059  * between subvols.
5060  */
5061 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5062                                       struct btrfs_root *root,
5063                                       struct btrfs_path *path,
5064                                       struct btrfs_key *first_key,
5065                                       struct btrfs_ref_path *ref_path,
5066                                       struct btrfs_block_group_cache *group,
5067                                       struct inode *reloc_inode)
5068 {
5069         struct btrfs_root *reloc_root;
5070         struct extent_buffer *eb = NULL;
5071         struct btrfs_key *keys;
5072         u64 *nodes;
5073         int level;
5074         int shared_level;
5075         int lowest_level = 0;
5076         int ret;
5077
5078         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5079                 lowest_level = ref_path->owner_objectid;
5080
5081         if (!root->ref_cows) {
5082                 path->lowest_level = lowest_level;
5083                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5084                 BUG_ON(ret < 0);
5085                 path->lowest_level = 0;
5086                 btrfs_release_path(root, path);
5087                 return 0;
5088         }
5089
5090         mutex_lock(&root->fs_info->tree_reloc_mutex);
5091         ret = init_reloc_tree(trans, root);
5092         BUG_ON(ret);
5093         reloc_root = root->reloc_root;
5094
5095         shared_level = ref_path->shared_level;
5096         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5097
5098         keys = ref_path->node_keys;
5099         nodes = ref_path->new_nodes;
5100         memset(&keys[shared_level + 1], 0,
5101                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5102         memset(&nodes[shared_level + 1], 0,
5103                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5104
5105         if (nodes[lowest_level] == 0) {
5106                 path->lowest_level = lowest_level;
5107                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5108                                         0, 1);
5109                 BUG_ON(ret);
5110                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5111                         eb = path->nodes[level];
5112                         if (!eb || eb == reloc_root->node)
5113                                 break;
5114                         nodes[level] = eb->start;
5115                         if (level == 0)
5116                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5117                         else
5118                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5119                 }
5120                 if (nodes[0] &&
5121                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5122                         eb = path->nodes[0];
5123                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5124                                                       group, reloc_inode);
5125                         BUG_ON(ret);
5126                 }
5127                 btrfs_release_path(reloc_root, path);
5128         } else {
5129                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5130                                        lowest_level);
5131                 BUG_ON(ret);
5132         }
5133
5134         /*
5135          * replace tree blocks in the fs tree with tree blocks in
5136          * the reloc tree.
5137          */
5138         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5139         BUG_ON(ret < 0);
5140
5141         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5142                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5143                                         0, 0);
5144                 BUG_ON(ret);
5145                 extent_buffer_get(path->nodes[0]);
5146                 eb = path->nodes[0];
5147                 btrfs_release_path(reloc_root, path);
5148                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5149                 BUG_ON(ret);
5150                 free_extent_buffer(eb);
5151         }
5152
5153         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5154         path->lowest_level = 0;
5155         return 0;
5156 }
5157
5158 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5159                                         struct btrfs_root *root,
5160                                         struct btrfs_path *path,
5161                                         struct btrfs_key *first_key,
5162                                         struct btrfs_ref_path *ref_path)
5163 {
5164         int ret;
5165
5166         ret = relocate_one_path(trans, root, path, first_key,
5167                                 ref_path, NULL, NULL);
5168         BUG_ON(ret);
5169
5170         return 0;
5171 }
5172
5173 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5174                                     struct btrfs_root *extent_root,
5175                                     struct btrfs_path *path,
5176                                     struct btrfs_key *extent_key)
5177 {
5178         int ret;
5179
5180         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5181         if (ret)
5182                 goto out;
5183         ret = btrfs_del_item(trans, extent_root, path);
5184 out:
5185         btrfs_release_path(extent_root, path);
5186         return ret;
5187 }
5188
5189 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5190                                                 struct btrfs_ref_path *ref_path)
5191 {
5192         struct btrfs_key root_key;
5193
5194         root_key.objectid = ref_path->root_objectid;
5195         root_key.type = BTRFS_ROOT_ITEM_KEY;
5196         if (is_cowonly_root(ref_path->root_objectid))
5197                 root_key.offset = 0;
5198         else
5199                 root_key.offset = (u64)-1;
5200
5201         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5202 }
5203
5204 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5205                                         struct btrfs_path *path,
5206                                         struct btrfs_key *extent_key,
5207                                         struct btrfs_block_group_cache *group,
5208                                         struct inode *reloc_inode, int pass)
5209 {
5210         struct btrfs_trans_handle *trans;
5211         struct btrfs_root *found_root;
5212         struct btrfs_ref_path *ref_path = NULL;
5213         struct disk_extent *new_extents = NULL;
5214         int nr_extents = 0;
5215         int loops;
5216         int ret;
5217         int level;
5218         struct btrfs_key first_key;
5219         u64 prev_block = 0;
5220
5221
5222         trans = btrfs_start_transaction(extent_root, 1);
5223         BUG_ON(!trans);
5224
5225         if (extent_key->objectid == 0) {
5226                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5227                 goto out;
5228         }
5229
5230         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5231         if (!ref_path) {
5232                 ret = -ENOMEM;
5233                 goto out;
5234         }
5235
5236         for (loops = 0; ; loops++) {
5237                 if (loops == 0) {
5238                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5239                                                    extent_key->objectid);
5240                 } else {
5241                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5242                 }
5243                 if (ret < 0)
5244                         goto out;
5245                 if (ret > 0)
5246                         break;
5247
5248                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5249                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5250                         continue;
5251
5252                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5253                 BUG_ON(!found_root);
5254                 /*
5255                  * for reference counted tree, only process reference paths
5256                  * rooted at the latest committed root.
5257                  */
5258                 if (found_root->ref_cows &&
5259                     ref_path->root_generation != found_root->root_key.offset)
5260                         continue;
5261
5262                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5263                         if (pass == 0) {
5264                                 /*
5265                                  * copy data extents to new locations
5266                                  */
5267                                 u64 group_start = group->key.objectid;
5268                                 ret = relocate_data_extent(reloc_inode,
5269                                                            extent_key,
5270                                                            group_start);
5271                                 if (ret < 0)
5272                                         goto out;
5273                                 break;
5274                         }
5275                         level = 0;
5276                 } else {
5277                         level = ref_path->owner_objectid;
5278                 }
5279
5280                 if (prev_block != ref_path->nodes[level]) {
5281                         struct extent_buffer *eb;
5282                         u64 block_start = ref_path->nodes[level];
5283                         u64 block_size = btrfs_level_size(found_root, level);
5284
5285                         eb = read_tree_block(found_root, block_start,
5286                                              block_size, 0);
5287                         btrfs_tree_lock(eb);
5288                         BUG_ON(level != btrfs_header_level(eb));
5289
5290                         if (level == 0)
5291                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5292                         else
5293                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5294
5295                         btrfs_tree_unlock(eb);
5296                         free_extent_buffer(eb);
5297                         prev_block = block_start;
5298                 }
5299
5300                 mutex_lock(&extent_root->fs_info->trans_mutex);
5301                 btrfs_record_root_in_trans(found_root);
5302                 mutex_unlock(&extent_root->fs_info->trans_mutex);
5303                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5304                         /*
5305                          * try to update data extent references while
5306                          * keeping metadata shared between snapshots.
5307                          */
5308                         if (pass == 1) {
5309                                 ret = relocate_one_path(trans, found_root,
5310                                                 path, &first_key, ref_path,
5311                                                 group, reloc_inode);
5312                                 if (ret < 0)
5313                                         goto out;
5314                                 continue;
5315                         }
5316                         /*
5317                          * use fallback method to process the remaining
5318                          * references.
5319                          */
5320                         if (!new_extents) {
5321                                 u64 group_start = group->key.objectid;
5322                                 new_extents = kmalloc(sizeof(*new_extents),
5323                                                       GFP_NOFS);
5324                                 nr_extents = 1;
5325                                 ret = get_new_locations(reloc_inode,
5326                                                         extent_key,
5327                                                         group_start, 1,
5328                                                         &new_extents,
5329                                                         &nr_extents);
5330                                 if (ret)
5331                                         goto out;
5332                         }
5333                         ret = replace_one_extent(trans, found_root,
5334                                                 path, extent_key,
5335                                                 &first_key, ref_path,
5336                                                 new_extents, nr_extents);
5337                 } else {
5338                         ret = relocate_tree_block(trans, found_root, path,
5339                                                   &first_key, ref_path);
5340                 }
5341                 if (ret < 0)
5342                         goto out;
5343         }
5344         ret = 0;
5345 out:
5346         btrfs_end_transaction(trans, extent_root);
5347         kfree(new_extents);
5348         kfree(ref_path);
5349         return ret;
5350 }
5351
5352 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5353 {
5354         u64 num_devices;
5355         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5356                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5357
5358         num_devices = root->fs_info->fs_devices->rw_devices;
5359         if (num_devices == 1) {
5360                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5361                 stripped = flags & ~stripped;
5362
5363                 /* turn raid0 into single device chunks */
5364                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5365                         return stripped;
5366
5367                 /* turn mirroring into duplication */
5368                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5369                              BTRFS_BLOCK_GROUP_RAID10))
5370                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5371                 return flags;
5372         } else {
5373                 /* they already had raid on here, just return */
5374                 if (flags & stripped)
5375                         return flags;
5376
5377                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5378                 stripped = flags & ~stripped;
5379
5380                 /* switch duplicated blocks with raid1 */
5381                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5382                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5383
5384                 /* turn single device chunks into raid0 */
5385                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5386         }
5387         return flags;
5388 }
5389
5390 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5391                      struct btrfs_block_group_cache *shrink_block_group,
5392                      int force)
5393 {
5394         struct btrfs_trans_handle *trans;
5395         u64 new_alloc_flags;
5396         u64 calc;
5397
5398         spin_lock(&shrink_block_group->lock);
5399         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5400                 spin_unlock(&shrink_block_group->lock);
5401
5402                 trans = btrfs_start_transaction(root, 1);
5403                 spin_lock(&shrink_block_group->lock);
5404
5405                 new_alloc_flags = update_block_group_flags(root,
5406                                                    shrink_block_group->flags);
5407                 if (new_alloc_flags != shrink_block_group->flags) {
5408                         calc =
5409                              btrfs_block_group_used(&shrink_block_group->item);
5410                 } else {
5411                         calc = shrink_block_group->key.offset;
5412                 }
5413                 spin_unlock(&shrink_block_group->lock);
5414
5415                 do_chunk_alloc(trans, root->fs_info->extent_root,
5416                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5417
5418                 btrfs_end_transaction(trans, root);
5419         } else
5420                 spin_unlock(&shrink_block_group->lock);
5421         return 0;
5422 }
5423
5424 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5425                                  struct btrfs_root *root,
5426                                  u64 objectid, u64 size)
5427 {
5428         struct btrfs_path *path;
5429         struct btrfs_inode_item *item;
5430         struct extent_buffer *leaf;
5431         int ret;
5432
5433         path = btrfs_alloc_path();
5434         if (!path)
5435                 return -ENOMEM;
5436
5437         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5438         if (ret)
5439                 goto out;
5440
5441         leaf = path->nodes[0];
5442         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5443         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5444         btrfs_set_inode_generation(leaf, item, 1);
5445         btrfs_set_inode_size(leaf, item, size);
5446         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5447         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5448         btrfs_mark_buffer_dirty(leaf);
5449         btrfs_release_path(root, path);
5450 out:
5451         btrfs_free_path(path);
5452         return ret;
5453 }
5454
5455 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5456                                         struct btrfs_block_group_cache *group)
5457 {
5458         struct inode *inode = NULL;
5459         struct btrfs_trans_handle *trans;
5460         struct btrfs_root *root;
5461         struct btrfs_key root_key;
5462         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5463         int err = 0;
5464
5465         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5466         root_key.type = BTRFS_ROOT_ITEM_KEY;
5467         root_key.offset = (u64)-1;
5468         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5469         if (IS_ERR(root))
5470                 return ERR_CAST(root);
5471
5472         trans = btrfs_start_transaction(root, 1);
5473         BUG_ON(!trans);
5474
5475         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5476         if (err)
5477                 goto out;
5478
5479         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5480         BUG_ON(err);
5481
5482         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5483                                        group->key.offset, 0, group->key.offset,
5484                                        0, 0, 0);
5485         BUG_ON(err);
5486
5487         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5488         if (inode->i_state & I_NEW) {
5489                 BTRFS_I(inode)->root = root;
5490                 BTRFS_I(inode)->location.objectid = objectid;
5491                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5492                 BTRFS_I(inode)->location.offset = 0;
5493                 btrfs_read_locked_inode(inode);
5494                 unlock_new_inode(inode);
5495                 BUG_ON(is_bad_inode(inode));
5496         } else {
5497                 BUG_ON(1);
5498         }
5499         BTRFS_I(inode)->index_cnt = group->key.objectid;
5500
5501         err = btrfs_orphan_add(trans, inode);
5502 out:
5503         btrfs_end_transaction(trans, root);
5504         if (err) {
5505                 if (inode)
5506                         iput(inode);
5507                 inode = ERR_PTR(err);
5508         }
5509         return inode;
5510 }
5511
5512 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5513 {
5514
5515         struct btrfs_ordered_sum *sums;
5516         struct btrfs_sector_sum *sector_sum;
5517         struct btrfs_ordered_extent *ordered;
5518         struct btrfs_root *root = BTRFS_I(inode)->root;
5519         struct list_head list;
5520         size_t offset;
5521         int ret;
5522         u64 disk_bytenr;
5523
5524         INIT_LIST_HEAD(&list);
5525
5526         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5527         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5528
5529         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5530         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5531                                        disk_bytenr + len - 1, &list);
5532
5533         while (!list_empty(&list)) {
5534                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5535                 list_del_init(&sums->list);
5536
5537                 sector_sum = sums->sums;
5538                 sums->bytenr = ordered->start;
5539
5540                 offset = 0;
5541                 while (offset < sums->len) {
5542                         sector_sum->bytenr += ordered->start - disk_bytenr;
5543                         sector_sum++;
5544                         offset += root->sectorsize;
5545                 }
5546
5547                 btrfs_add_ordered_sum(inode, ordered, sums);
5548         }
5549         btrfs_put_ordered_extent(ordered);
5550         return 0;
5551 }
5552
5553 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5554 {
5555         struct btrfs_trans_handle *trans;
5556         struct btrfs_path *path;
5557         struct btrfs_fs_info *info = root->fs_info;
5558         struct extent_buffer *leaf;
5559         struct inode *reloc_inode;
5560         struct btrfs_block_group_cache *block_group;
5561         struct btrfs_key key;
5562         u64 skipped;
5563         u64 cur_byte;
5564         u64 total_found;
5565         u32 nritems;
5566         int ret;
5567         int progress;
5568         int pass = 0;
5569
5570         root = root->fs_info->extent_root;
5571
5572         block_group = btrfs_lookup_block_group(info, group_start);
5573         BUG_ON(!block_group);
5574
5575         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5576                (unsigned long long)block_group->key.objectid,
5577                (unsigned long long)block_group->flags);
5578
5579         path = btrfs_alloc_path();
5580         BUG_ON(!path);
5581
5582         reloc_inode = create_reloc_inode(info, block_group);
5583         BUG_ON(IS_ERR(reloc_inode));
5584
5585         __alloc_chunk_for_shrink(root, block_group, 1);
5586         set_block_group_readonly(block_group);
5587
5588         btrfs_start_delalloc_inodes(info->tree_root);
5589         btrfs_wait_ordered_extents(info->tree_root, 0);
5590 again:
5591         skipped = 0;
5592         total_found = 0;
5593         progress = 0;
5594         key.objectid = block_group->key.objectid;
5595         key.offset = 0;
5596         key.type = 0;
5597         cur_byte = key.objectid;
5598
5599         trans = btrfs_start_transaction(info->tree_root, 1);
5600         btrfs_commit_transaction(trans, info->tree_root);
5601
5602         mutex_lock(&root->fs_info->cleaner_mutex);
5603         btrfs_clean_old_snapshots(info->tree_root);
5604         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5605         mutex_unlock(&root->fs_info->cleaner_mutex);
5606
5607         trans = btrfs_start_transaction(info->tree_root, 1);
5608         btrfs_commit_transaction(trans, info->tree_root);
5609
5610         while (1) {
5611                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5612                 if (ret < 0)
5613                         goto out;
5614 next:
5615                 leaf = path->nodes[0];
5616                 nritems = btrfs_header_nritems(leaf);
5617                 if (path->slots[0] >= nritems) {
5618                         ret = btrfs_next_leaf(root, path);
5619                         if (ret < 0)
5620                                 goto out;
5621                         if (ret == 1) {
5622                                 ret = 0;
5623                                 break;
5624                         }
5625                         leaf = path->nodes[0];
5626                         nritems = btrfs_header_nritems(leaf);
5627                 }
5628
5629                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5630
5631                 if (key.objectid >= block_group->key.objectid +
5632                     block_group->key.offset)
5633                         break;
5634
5635                 if (progress && need_resched()) {
5636                         btrfs_release_path(root, path);
5637                         cond_resched();
5638                         progress = 0;
5639                         continue;
5640                 }
5641                 progress = 1;
5642
5643                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5644                     key.objectid + key.offset <= cur_byte) {
5645                         path->slots[0]++;
5646                         goto next;
5647                 }
5648
5649                 total_found++;
5650                 cur_byte = key.objectid + key.offset;
5651                 btrfs_release_path(root, path);
5652
5653                 __alloc_chunk_for_shrink(root, block_group, 0);
5654                 ret = relocate_one_extent(root, path, &key, block_group,
5655                                           reloc_inode, pass);
5656                 BUG_ON(ret < 0);
5657                 if (ret > 0)
5658                         skipped++;
5659
5660                 key.objectid = cur_byte;
5661                 key.type = 0;
5662                 key.offset = 0;
5663         }
5664
5665         btrfs_release_path(root, path);
5666
5667         if (pass == 0) {
5668                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5669                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5670         }
5671
5672         if (total_found > 0) {
5673                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5674                        (unsigned long long)total_found, pass);
5675                 pass++;
5676                 if (total_found == skipped && pass > 2) {
5677                         iput(reloc_inode);
5678                         reloc_inode = create_reloc_inode(info, block_group);
5679                         pass = 0;
5680                 }
5681                 goto again;
5682         }
5683
5684         /* delete reloc_inode */
5685         iput(reloc_inode);
5686
5687         /* unpin extents in this range */
5688         trans = btrfs_start_transaction(info->tree_root, 1);
5689         btrfs_commit_transaction(trans, info->tree_root);
5690
5691         spin_lock(&block_group->lock);
5692         WARN_ON(block_group->pinned > 0);
5693         WARN_ON(block_group->reserved > 0);
5694         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5695         spin_unlock(&block_group->lock);
5696         put_block_group(block_group);
5697         ret = 0;
5698 out:
5699         btrfs_free_path(path);
5700         return ret;
5701 }
5702
5703 static int find_first_block_group(struct btrfs_root *root,
5704                 struct btrfs_path *path, struct btrfs_key *key)
5705 {
5706         int ret = 0;
5707         struct btrfs_key found_key;
5708         struct extent_buffer *leaf;
5709         int slot;
5710
5711         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5712         if (ret < 0)
5713                 goto out;
5714
5715         while (1) {
5716                 slot = path->slots[0];
5717                 leaf = path->nodes[0];
5718                 if (slot >= btrfs_header_nritems(leaf)) {
5719                         ret = btrfs_next_leaf(root, path);
5720                         if (ret == 0)
5721                                 continue;
5722                         if (ret < 0)
5723                                 goto out;
5724                         break;
5725                 }
5726                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5727
5728                 if (found_key.objectid >= key->objectid &&
5729                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5730                         ret = 0;
5731                         goto out;
5732                 }
5733                 path->slots[0]++;
5734         }
5735         ret = -ENOENT;
5736 out:
5737         return ret;
5738 }
5739
5740 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5741 {
5742         struct btrfs_block_group_cache *block_group;
5743         struct btrfs_space_info *space_info;
5744         struct rb_node *n;
5745
5746         spin_lock(&info->block_group_cache_lock);
5747         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5748                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5749                                        cache_node);
5750                 rb_erase(&block_group->cache_node,
5751                          &info->block_group_cache_tree);
5752                 spin_unlock(&info->block_group_cache_lock);
5753
5754                 btrfs_remove_free_space_cache(block_group);
5755                 down_write(&block_group->space_info->groups_sem);
5756                 list_del(&block_group->list);
5757                 up_write(&block_group->space_info->groups_sem);
5758
5759                 WARN_ON(atomic_read(&block_group->count) != 1);
5760                 kfree(block_group);
5761
5762                 spin_lock(&info->block_group_cache_lock);
5763         }
5764         spin_unlock(&info->block_group_cache_lock);
5765
5766         /* now that all the block groups are freed, go through and
5767          * free all the space_info structs.  This is only called during
5768          * the final stages of unmount, and so we know nobody is
5769          * using them.  We call synchronize_rcu() once before we start,
5770          * just to be on the safe side.
5771          */
5772         synchronize_rcu();
5773
5774         while(!list_empty(&info->space_info)) {
5775                 space_info = list_entry(info->space_info.next,
5776                                         struct btrfs_space_info,
5777                                         list);
5778
5779                 list_del(&space_info->list);
5780                 kfree(space_info);
5781         }
5782         return 0;
5783 }
5784
5785 int btrfs_read_block_groups(struct btrfs_root *root)
5786 {
5787         struct btrfs_path *path;
5788         int ret;
5789         struct btrfs_block_group_cache *cache;
5790         struct btrfs_fs_info *info = root->fs_info;
5791         struct btrfs_space_info *space_info;
5792         struct btrfs_key key;
5793         struct btrfs_key found_key;
5794         struct extent_buffer *leaf;
5795
5796         root = info->extent_root;
5797         key.objectid = 0;
5798         key.offset = 0;
5799         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5800         path = btrfs_alloc_path();
5801         if (!path)
5802                 return -ENOMEM;
5803
5804         while (1) {
5805                 ret = find_first_block_group(root, path, &key);
5806                 if (ret > 0) {
5807                         ret = 0;
5808                         goto error;
5809                 }
5810                 if (ret != 0)
5811                         goto error;
5812
5813                 leaf = path->nodes[0];
5814                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5815                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5816                 if (!cache) {
5817                         ret = -ENOMEM;
5818                         break;
5819                 }
5820
5821                 atomic_set(&cache->count, 1);
5822                 spin_lock_init(&cache->lock);
5823                 mutex_init(&cache->alloc_mutex);
5824                 mutex_init(&cache->cache_mutex);
5825                 INIT_LIST_HEAD(&cache->list);
5826                 read_extent_buffer(leaf, &cache->item,
5827                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5828                                    sizeof(cache->item));
5829                 memcpy(&cache->key, &found_key, sizeof(found_key));
5830
5831                 key.objectid = found_key.objectid + found_key.offset;
5832                 btrfs_release_path(root, path);
5833                 cache->flags = btrfs_block_group_flags(&cache->item);
5834
5835                 ret = update_space_info(info, cache->flags, found_key.offset,
5836                                         btrfs_block_group_used(&cache->item),
5837                                         &space_info);
5838                 BUG_ON(ret);
5839                 cache->space_info = space_info;
5840                 down_write(&space_info->groups_sem);
5841                 list_add_tail(&cache->list, &space_info->block_groups);
5842                 up_write(&space_info->groups_sem);
5843
5844                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5845                 BUG_ON(ret);
5846
5847                 set_avail_alloc_bits(root->fs_info, cache->flags);
5848                 if (btrfs_chunk_readonly(root, cache->key.objectid))
5849                         set_block_group_readonly(cache);
5850         }
5851         ret = 0;
5852 error:
5853         btrfs_free_path(path);
5854         return ret;
5855 }
5856
5857 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5858                            struct btrfs_root *root, u64 bytes_used,
5859                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5860                            u64 size)
5861 {
5862         int ret;
5863         struct btrfs_root *extent_root;
5864         struct btrfs_block_group_cache *cache;
5865
5866         extent_root = root->fs_info->extent_root;
5867
5868         root->fs_info->last_trans_new_blockgroup = trans->transid;
5869
5870         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5871         if (!cache)
5872                 return -ENOMEM;
5873
5874         cache->key.objectid = chunk_offset;
5875         cache->key.offset = size;
5876         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5877         atomic_set(&cache->count, 1);
5878         spin_lock_init(&cache->lock);
5879         mutex_init(&cache->alloc_mutex);
5880         mutex_init(&cache->cache_mutex);
5881         INIT_LIST_HEAD(&cache->list);
5882
5883         btrfs_set_block_group_used(&cache->item, bytes_used);
5884         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5885         cache->flags = type;
5886         btrfs_set_block_group_flags(&cache->item, type);
5887
5888         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5889                                 &cache->space_info);
5890         BUG_ON(ret);
5891         down_write(&cache->space_info->groups_sem);
5892         list_add_tail(&cache->list, &cache->space_info->block_groups);
5893         up_write(&cache->space_info->groups_sem);
5894
5895         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5896         BUG_ON(ret);
5897
5898         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5899                                 sizeof(cache->item));
5900         BUG_ON(ret);
5901
5902         set_avail_alloc_bits(extent_root->fs_info, type);
5903
5904         return 0;
5905 }
5906
5907 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5908                              struct btrfs_root *root, u64 group_start)
5909 {
5910         struct btrfs_path *path;
5911         struct btrfs_block_group_cache *block_group;
5912         struct btrfs_key key;
5913         int ret;
5914
5915         root = root->fs_info->extent_root;
5916
5917         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5918         BUG_ON(!block_group);
5919         BUG_ON(!block_group->ro);
5920
5921         memcpy(&key, &block_group->key, sizeof(key));
5922
5923         path = btrfs_alloc_path();
5924         BUG_ON(!path);
5925
5926         spin_lock(&root->fs_info->block_group_cache_lock);
5927         rb_erase(&block_group->cache_node,
5928                  &root->fs_info->block_group_cache_tree);
5929         spin_unlock(&root->fs_info->block_group_cache_lock);
5930         btrfs_remove_free_space_cache(block_group);
5931         down_write(&block_group->space_info->groups_sem);
5932         list_del(&block_group->list);
5933         up_write(&block_group->space_info->groups_sem);
5934
5935         spin_lock(&block_group->space_info->lock);
5936         block_group->space_info->total_bytes -= block_group->key.offset;
5937         block_group->space_info->bytes_readonly -= block_group->key.offset;
5938         spin_unlock(&block_group->space_info->lock);
5939         block_group->space_info->full = 0;
5940
5941         put_block_group(block_group);
5942         put_block_group(block_group);
5943
5944         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5945         if (ret > 0)
5946                 ret = -EIO;
5947         if (ret < 0)
5948                 goto out;
5949
5950         ret = btrfs_del_item(trans, root, path);
5951 out:
5952         btrfs_free_path(path);
5953         return ret;
5954 }