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 "compat.h"
  24 #include "hash.h"
  25 #include "crc32c.h"
  26 #include "ctree.h"
  27 #include "disk-io.h"
  28 #include "print-tree.h"
  29 #include "transaction.h"
  30 #include "volumes.h"
  31 #include "locking.h"
  32 #include "ref-cache.h"
  33
  34 #define PENDING_EXTENT_INSERT 0
  35 #define PENDING_EXTENT_DELETE 1
  36 #define PENDING_BACKREF_UPDATE 2
  37
  38 struct pending_extent_op {
  39         int type;
  40         u64 bytenr;
  41         u64 num_bytes;
  42         u64 parent;
  43         u64 orig_parent;
  44         u64 generation;
  45         u64 orig_generation;
  46         int level;
  47         struct list_head list;
  48         int del;
  49 };
  50
  51 static int finish_current_insert(struct btrfs_trans_handle *trans,
  52                                  struct btrfs_root *extent_root, int all);
  53 static int del_pending_extents(struct btrfs_trans_handle *trans,
  54                                struct btrfs_root *extent_root, int all);
  55 static int pin_down_bytes(struct btrfs_trans_handle *trans,
  56                           struct btrfs_root *root,
  57                           u64 bytenr, u64 num_bytes, int is_data);
  58 static int update_block_group(struct btrfs_trans_handle *trans,
  59                               struct btrfs_root *root,
  60                               u64 bytenr, u64 num_bytes, int alloc,
  61                               int mark_free);
  62
  63 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  64 {
  65         return (cache->flags & bits) == bits;
  66 }
  67
  68 /*
  69  * this adds the block group to the fs_info rb tree for the block group
  70  * cache
  71  */
  72 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  73                                 struct btrfs_block_group_cache *block_group)
  74 {
  75         struct rb_node **p;
  76         struct rb_node *parent = NULL;
  77         struct btrfs_block_group_cache *cache;
  78
  79         spin_lock(&info->block_group_cache_lock);
  80         p = &info->block_group_cache_tree.rb_node;
  81
  82         while (*p) {
  83                 parent = *p;
  84                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  85                                  cache_node);
  86                 if (block_group->key.objectid < cache->key.objectid) {
  87                         p = &(*p)->rb_left;
  88                 } else if (block_group->key.objectid > cache->key.objectid) {
  89                         p = &(*p)->rb_right;
  90                 } else {
  91                         spin_unlock(&info->block_group_cache_lock);
  92                         return -EEXIST;
  93                 }
  94         }
  95
  96         rb_link_node(&block_group->cache_node, parent, p);
  97         rb_insert_color(&block_group->cache_node,
  98                         &info->block_group_cache_tree);
  99         spin_unlock(&info->block_group_cache_lock);
 100
 101         return 0;
 102 }
 103
 104 /*
 105  * This will return the block group at or after bytenr if contains is 0, else
 106  * it will return the block group that contains the bytenr
 107  */
 108 static struct btrfs_block_group_cache *
 109 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 110                               int contains)
 111 {
 112         struct btrfs_block_group_cache *cache, *ret = NULL;
 113         struct rb_node *n;
 114         u64 end, start;
 115
 116         spin_lock(&info->block_group_cache_lock);
 117         n = info->block_group_cache_tree.rb_node;
 118
 119         while (n) {
 120                 cache = rb_entry(n, struct btrfs_block_group_cache,
 121                                  cache_node);
 122                 end = cache->key.objectid + cache->key.offset - 1;
 123                 start = cache->key.objectid;
 124
 125                 if (bytenr < start) {
 126                         if (!contains && (!ret || start < ret->key.objectid))
 127                                 ret = cache;
 128                         n = n->rb_left;
 129                 } else if (bytenr > start) {
 130                         if (contains && bytenr <= end) {
 131                                 ret = cache;
 132                                 break;
 133                         }
 134                         n = n->rb_right;
 135                 } else {
 136                         ret = cache;
 137                         break;
 138                 }
 139         }
 140         if (ret)
 141                 atomic_inc(&ret->count);
 142         spin_unlock(&info->block_group_cache_lock);
 143
 144         return ret;
 145 }
 146
 147 /*
 148  * this is only called by cache_block_group, since we could have freed extents
 149  * we need to check the pinned_extents for any extents that can't be used yet
 150  * since their free space will be released as soon as the transaction commits.
 151  */
 152 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 153                               struct btrfs_fs_info *info, u64 start, u64 end)
 154 {
 155         u64 extent_start, extent_end, size;
 156         int ret;
 157
 158         mutex_lock(&info->pinned_mutex);
 159         while (start < end) {
 160                 ret = find_first_extent_bit(&info->pinned_extents, start,
 161                                             &extent_start, &extent_end,
 162                                             EXTENT_DIRTY);
 163                 if (ret)
 164                         break;
 165
 166                 if (extent_start == start) {
 167                         start = extent_end + 1;
 168                 } else if (extent_start > start && extent_start < end) {
 169                         size = extent_start - start;
 170                         ret = btrfs_add_free_space(block_group, start,
 171                                                    size);
 172                         BUG_ON(ret);
 173                         start = extent_end + 1;
 174                 } else {
 175                         break;
 176                 }
 177         }
 178
 179         if (start < end) {
 180                 size = end - start;
 181                 ret = btrfs_add_free_space(block_group, start, size);
 182                 BUG_ON(ret);
 183         }
 184         mutex_unlock(&info->pinned_mutex);
 185
 186         return 0;
 187 }
 188
 189 static int remove_sb_from_cache(struct btrfs_root *root,
 190                                 struct btrfs_block_group_cache *cache)
 191 {
 192         u64 bytenr;
 193         u64 *logical;
 194         int stripe_len;
 195         int i, nr, ret;
 196
 197         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 198                 bytenr = btrfs_sb_offset(i);
 199                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 200                                        cache->key.objectid, bytenr, 0,
 201                                        &logical, &nr, &stripe_len);
 202                 BUG_ON(ret);
 203                 while (nr--) {
 204                         btrfs_remove_free_space(cache, logical[nr],
 205                                                 stripe_len);
 206                 }
 207                 kfree(logical);
 208         }
 209         return 0;
 210 }
 211
 212 static int cache_block_group(struct btrfs_root *root,
 213                              struct btrfs_block_group_cache *block_group)
 214 {
 215         struct btrfs_path *path;
 216         int ret = 0;
 217         struct btrfs_key key;
 218         struct extent_buffer *leaf;
 219         int slot;
 220         u64 last;
 221
 222         if (!block_group)
 223                 return 0;
 224
 225         root = root->fs_info->extent_root;
 226
 227         if (block_group->cached)
 228                 return 0;
 229
 230         path = btrfs_alloc_path();
 231         if (!path)
 232                 return -ENOMEM;
 233
 234         path->reada = 2;
 235         /*
 236          * we get into deadlocks with paths held by callers of this function.
 237          * since the alloc_mutex is protecting things right now, just
 238          * skip the locking here
 239          */
 240         path->skip_locking = 1;
 241         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 242         key.objectid = last;
 243         key.offset = 0;
 244         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 245         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 246         if (ret < 0)
 247                 goto err;
 248
 249         while (1) {
 250                 leaf = path->nodes[0];
 251                 slot = path->slots[0];
 252                 if (slot >= btrfs_header_nritems(leaf)) {
 253                         ret = btrfs_next_leaf(root, path);
 254                         if (ret < 0)
 255                                 goto err;
 256                         if (ret == 0)
 257                                 continue;
 258                         else
 259                                 break;
 260                 }
 261                 btrfs_item_key_to_cpu(leaf, &key, slot);
 262                 if (key.objectid < block_group->key.objectid)
 263                         goto next;
 264
 265                 if (key.objectid >= block_group->key.objectid +
 266                     block_group->key.offset)
 267                         break;
 268
 269                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 270                         add_new_free_space(block_group, root->fs_info, last,
 271                                            key.objectid);
 272
 273                         last = key.objectid + key.offset;
 274                 }
 275 next:
 276                 path->slots[0]++;
 277         }
 278
 279         add_new_free_space(block_group, root->fs_info, last,
 280                            block_group->key.objectid +
 281                            block_group->key.offset);
 282
 283         remove_sb_from_cache(root, block_group);
 284         block_group->cached = 1;
 285         ret = 0;
 286 err:
 287         btrfs_free_path(path);
 288         return ret;
 289 }
 290
 291 /*
 292  * return the block group that starts at or after bytenr
 293  */
 294 static struct btrfs_block_group_cache *
 295 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 296 {
 297         struct btrfs_block_group_cache *cache;
 298
 299         cache = block_group_cache_tree_search(info, bytenr, 0);
 300
 301         return cache;
 302 }
 303
 304 /*
 305  * return the block group that contains teh given bytenr
 306  */
 307 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 308                                                  struct btrfs_fs_info *info,
 309                                                  u64 bytenr)
 310 {
 311         struct btrfs_block_group_cache *cache;
 312
 313         cache = block_group_cache_tree_search(info, bytenr, 1);
 314
 315         return cache;
 316 }
 317
 318 static inline void put_block_group(struct btrfs_block_group_cache *cache)
 319 {
 320         if (atomic_dec_and_test(&cache->count))
 321                 kfree(cache);
 322 }
 323
 324 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 325                                                   u64 flags)
 326 {
 327         struct list_head *head = &info->space_info;
 328         struct btrfs_space_info *found;
 329         list_for_each_entry(found, head, list) {
 330                 if (found->flags == flags)
 331                         return found;
 332         }
 333         return NULL;
 334 }
 335
 336 static u64 div_factor(u64 num, int factor)
 337 {
 338         if (factor == 10)
 339                 return num;
 340         num *= factor;
 341         do_div(num, 10);
 342         return num;
 343 }
 344
 345 u64 btrfs_find_block_group(struct btrfs_root *root,
 346                            u64 search_start, u64 search_hint, int owner)
 347 {
 348         struct btrfs_block_group_cache *cache;
 349         u64 used;
 350         u64 last = max(search_hint, search_start);
 351         u64 group_start = 0;
 352         int full_search = 0;
 353         int factor = 9;
 354         int wrapped = 0;
 355 again:
 356         while (1) {
 357                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 358                 if (!cache)
 359                         break;
 360
 361                 spin_lock(&cache->lock);
 362                 last = cache->key.objectid + cache->key.offset;
 363                 used = btrfs_block_group_used(&cache->item);
 364
 365                 if ((full_search || !cache->ro) &&
 366                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 367                         if (used + cache->pinned + cache->reserved <
 368                             div_factor(cache->key.offset, factor)) {
 369                                 group_start = cache->key.objectid;
 370                                 spin_unlock(&cache->lock);
 371                                 put_block_group(cache);
 372                                 goto found;
 373                         }
 374                 }
 375                 spin_unlock(&cache->lock);
 376                 put_block_group(cache);
 377                 cond_resched();
 378         }
 379         if (!wrapped) {
 380                 last = search_start;
 381                 wrapped = 1;
 382                 goto again;
 383         }
 384         if (!full_search && factor < 10) {
 385                 last = search_start;
 386                 full_search = 1;
 387                 factor = 10;
 388                 goto again;
 389         }
 390 found:
 391         return group_start;
 392 }
 393
 394 /* simple helper to search for an existing extent at a given offset */
 395 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 396 {
 397         int ret;
 398         struct btrfs_key key;
 399         struct btrfs_path *path;
 400
 401         path = btrfs_alloc_path();
 402         BUG_ON(!path);
 403         key.objectid = start;
 404         key.offset = len;
 405         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 406         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 407                                 0, 0);
 408         btrfs_free_path(path);
 409         return ret;
 410 }
 411
 412 /*
 413  * Back reference rules.  Back refs have three main goals:
 414  *
 415  * 1) differentiate between all holders of references to an extent so that
 416  *    when a reference is dropped we can make sure it was a valid reference
 417  *    before freeing the extent.
 418  *
 419  * 2) Provide enough information to quickly find the holders of an extent
 420  *    if we notice a given block is corrupted or bad.
 421  *
 422  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 423  *    maintenance.  This is actually the same as #2, but with a slightly
 424  *    different use case.
 425  *
 426  * File extents can be referenced by:
 427  *
 428  * - multiple snapshots, subvolumes, or different generations in one subvol
 429  * - different files inside a single subvolume
 430  * - different offsets inside a file (bookend extents in file.c)
 431  *
 432  * The extent ref structure has fields for:
 433  *
 434  * - Objectid of the subvolume root
 435  * - Generation number of the tree holding the reference
 436  * - objectid of the file holding the reference
 437  * - number of references holding by parent node (alway 1 for tree blocks)
 438  *
 439  * Btree leaf may hold multiple references to a file extent. In most cases,
 440  * these references are from same file and the corresponding offsets inside
 441  * the file are close together.
 442  *
 443  * When a file extent is allocated the fields are filled in:
 444  *     (root_key.objectid, trans->transid, inode objectid, 1)
 445  *
 446  * When a leaf is cow'd new references are added for every file extent found
 447  * in the leaf.  It looks similar to the create case, but trans->transid will
 448  * be different when the block is cow'd.
 449  *
 450  *     (root_key.objectid, trans->transid, inode objectid,
 451  *      number of references in the leaf)
 452  *
 453  * When a file extent is removed either during snapshot deletion or
 454  * file truncation, we find the corresponding back reference and check
 455  * the following fields:
 456  *
 457  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 458  *      inode objectid)
 459  *
 460  * Btree extents can be referenced by:
 461  *
 462  * - Different subvolumes
 463  * - Different generations of the same subvolume
 464  *
 465  * When a tree block is created, back references are inserted:
 466  *
 467  * (root->root_key.objectid, trans->transid, level, 1)
 468  *
 469  * When a tree block is cow'd, new back references are added for all the
 470  * blocks it points to. If the tree block isn't in reference counted root,
 471  * the old back references are removed. These new back references are of
 472  * the form (trans->transid will have increased since creation):
 473  *
 474  * (root->root_key.objectid, trans->transid, level, 1)
 475  *
 476  * When a backref is in deleting, the following fields are checked:
 477  *
 478  * if backref was for a tree root:
 479  *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
 480  * else
 481  *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
 482  *
 483  * Back Reference Key composing:
 484  *
 485  * The key objectid corresponds to the first byte in the extent, the key
 486  * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
 487  * byte of parent extent. If a extent is tree root, the key offset is set
 488  * to the key objectid.
 489  */
 490
 491 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
 492                                           struct btrfs_root *root,
 493                                           struct btrfs_path *path,
 494                                           u64 bytenr, u64 parent,
 495                                           u64 ref_root, u64 ref_generation,
 496                                           u64 owner_objectid, int del)
 497 {
 498         struct btrfs_key key;
 499         struct btrfs_extent_ref *ref;
 500         struct extent_buffer *leaf;
 501         u64 ref_objectid;
 502         int ret;
 503
 504         key.objectid = bytenr;
 505         key.type = BTRFS_EXTENT_REF_KEY;
 506         key.offset = parent;
 507
 508         ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
 509         if (ret < 0)
 510                 goto out;
 511         if (ret > 0) {
 512                 ret = -ENOENT;
 513                 goto out;
 514         }
 515
 516         leaf = path->nodes[0];
 517         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 518         ref_objectid = btrfs_ref_objectid(leaf, ref);
 519         if (btrfs_ref_root(leaf, ref) != ref_root ||
 520             btrfs_ref_generation(leaf, ref) != ref_generation ||
 521             (ref_objectid != owner_objectid &&
 522              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 523                 ret = -EIO;
 524                 WARN_ON(1);
 525                 goto out;
 526         }
 527         ret = 0;
 528 out:
 529         return ret;
 530 }
 531
 532 /*
 533  * updates all the backrefs that are pending on update_list for the
 534  * extent_root
 535  */
 536 static noinline int update_backrefs(struct btrfs_trans_handle *trans,
 537                                     struct btrfs_root *extent_root,
 538                                     struct btrfs_path *path,
 539                                     struct list_head *update_list)
 540 {
 541         struct btrfs_key key;
 542         struct btrfs_extent_ref *ref;
 543         struct btrfs_fs_info *info = extent_root->fs_info;
 544         struct pending_extent_op *op;
 545         struct extent_buffer *leaf;
 546         int ret = 0;
 547         struct list_head *cur = update_list->next;
 548         u64 ref_objectid;
 549         u64 ref_root = extent_root->root_key.objectid;
 550
 551         op = list_entry(cur, struct pending_extent_op, list);
 552
 553 search:
 554         key.objectid = op->bytenr;
 555         key.type = BTRFS_EXTENT_REF_KEY;
 556         key.offset = op->orig_parent;
 557
 558         ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
 559         BUG_ON(ret);
 560
 561         leaf = path->nodes[0];
 562
 563 loop:
 564         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 565
 566         ref_objectid = btrfs_ref_objectid(leaf, ref);
 567
 568         if (btrfs_ref_root(leaf, ref) != ref_root ||
 569             btrfs_ref_generation(leaf, ref) != op->orig_generation ||
 570             (ref_objectid != op->level &&
 571              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 572                 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
 573                        "root %llu, owner %u\n",
 574                        (unsigned long long)op->bytenr,
 575                        (unsigned long long)op->orig_parent,
 576                        (unsigned long long)ref_root, op->level);
 577                 btrfs_print_leaf(extent_root, leaf);
 578                 BUG();
 579         }
 580
 581         key.objectid = op->bytenr;
 582         key.offset = op->parent;
 583         key.type = BTRFS_EXTENT_REF_KEY;
 584         ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
 585         BUG_ON(ret);
 586         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 587         btrfs_set_ref_generation(leaf, ref, op->generation);
 588
 589         cur = cur->next;
 590
 591         list_del_init(&op->list);
 592         unlock_extent(&info->extent_ins, op->bytenr,
 593                       op->bytenr + op->num_bytes - 1, GFP_NOFS);
 594         kfree(op);
 595
 596         if (cur == update_list) {
 597                 btrfs_mark_buffer_dirty(path->nodes[0]);
 598                 btrfs_release_path(extent_root, path);
 599                 goto out;
 600         }
 601
 602         op = list_entry(cur, struct pending_extent_op, list);
 603
 604         path->slots[0]++;
 605         while (path->slots[0] < btrfs_header_nritems(leaf)) {
 606                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 607                 if (key.objectid == op->bytenr &&
 608                     key.type == BTRFS_EXTENT_REF_KEY)
 609                         goto loop;
 610                 path->slots[0]++;
 611         }
 612
 613         btrfs_mark_buffer_dirty(path->nodes[0]);
 614         btrfs_release_path(extent_root, path);
 615         goto search;
 616
 617 out:
 618         return 0;
 619 }
 620
 621 static noinline int insert_extents(struct btrfs_trans_handle *trans,
 622                                    struct btrfs_root *extent_root,
 623                                    struct btrfs_path *path,
 624                                    struct list_head *insert_list, int nr)
 625 {
 626         struct btrfs_key *keys;
 627         u32 *data_size;
 628         struct pending_extent_op *op;
 629         struct extent_buffer *leaf;
 630         struct list_head *cur = insert_list->next;
 631         struct btrfs_fs_info *info = extent_root->fs_info;
 632         u64 ref_root = extent_root->root_key.objectid;
 633         int i = 0, last = 0, ret;
 634         int total = nr * 2;
 635
 636         if (!nr)
 637                 return 0;
 638
 639         keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
 640         if (!keys)
 641                 return -ENOMEM;
 642
 643         data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
 644         if (!data_size) {
 645                 kfree(keys);
 646                 return -ENOMEM;
 647         }
 648
 649         list_for_each_entry(op, insert_list, list) {
 650                 keys[i].objectid = op->bytenr;
 651                 keys[i].offset = op->num_bytes;
 652                 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
 653                 data_size[i] = sizeof(struct btrfs_extent_item);
 654                 i++;
 655
 656                 keys[i].objectid = op->bytenr;
 657                 keys[i].offset = op->parent;
 658                 keys[i].type = BTRFS_EXTENT_REF_KEY;
 659                 data_size[i] = sizeof(struct btrfs_extent_ref);
 660                 i++;
 661         }
 662
 663         op = list_entry(cur, struct pending_extent_op, list);
 664         i = 0;
 665         while (i < total) {
 666                 int c;
 667                 ret = btrfs_insert_some_items(trans, extent_root, path,
 668                                               keys+i, data_size+i, total-i);
 669                 BUG_ON(ret < 0);
 670
 671                 if (last && ret > 1)
 672                         BUG();
 673
 674                 leaf = path->nodes[0];
 675                 for (c = 0; c < ret; c++) {
 676                         int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
 677
 678                         /*
 679                          * if the first item we inserted was a backref, then
 680                          * the EXTENT_ITEM will be the odd c's, else it will
 681                          * be the even c's
 682                          */
 683                         if ((ref_first && (c % 2)) ||
 684                             (!ref_first && !(c % 2))) {
 685                                 struct btrfs_extent_item *itm;
 686
 687                                 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
 688                                                      struct btrfs_extent_item);
 689                                 btrfs_set_extent_refs(path->nodes[0], itm, 1);
 690                                 op->del++;
 691                         } else {
 692                                 struct btrfs_extent_ref *ref;
 693
 694                                 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
 695                                                      struct btrfs_extent_ref);
 696                                 btrfs_set_ref_root(leaf, ref, ref_root);
 697                                 btrfs_set_ref_generation(leaf, ref,
 698                                                          op->generation);
 699                                 btrfs_set_ref_objectid(leaf, ref, op->level);
 700                                 btrfs_set_ref_num_refs(leaf, ref, 1);
 701                                 op->del++;
 702                         }
 703
 704                         /*
 705                          * using del to see when its ok to free up the
 706                          * pending_extent_op.  In the case where we insert the
 707                          * last item on the list in order to help do batching
 708                          * we need to not free the extent op until we actually
 709                          * insert the extent_item
 710                          */
 711                         if (op->del == 2) {
 712                                 unlock_extent(&info->extent_ins, op->bytenr,
 713                                               op->bytenr + op->num_bytes - 1,
 714                                               GFP_NOFS);
 715                                 cur = cur->next;
 716                                 list_del_init(&op->list);
 717                                 kfree(op);
 718                                 if (cur != insert_list)
 719                                         op = list_entry(cur,
 720                                                 struct pending_extent_op,
 721                                                 list);
 722                         }
 723                 }
 724                 btrfs_mark_buffer_dirty(leaf);
 725                 btrfs_release_path(extent_root, path);
 726
 727                 /*
 728                  * Ok backref's and items usually go right next to eachother,
 729                  * but if we could only insert 1 item that means that we
 730                  * inserted on the end of a leaf, and we have no idea what may
 731                  * be on the next leaf so we just play it safe.  In order to
 732                  * try and help this case we insert the last thing on our
 733                  * insert list so hopefully it will end up being the last
 734                  * thing on the leaf and everything else will be before it,
 735                  * which will let us insert a whole bunch of items at the same
 736                  * time.
 737                  */
 738                 if (ret == 1 && !last && (i + ret < total)) {
 739                         /*
 740                          * last: where we will pick up the next time around
 741                          * i: our current key to insert, will be total - 1
 742                          * cur: the current op we are screwing with
 743                          * op: duh
 744                          */
 745                         last = i + ret;
 746                         i = total - 1;
 747                         cur = insert_list->prev;
 748                         op = list_entry(cur, struct pending_extent_op, list);
 749                 } else if (last) {
 750                         /*
 751                          * ok we successfully inserted the last item on the
 752                          * list, lets reset everything
 753                          *
 754                          * i: our current key to insert, so where we left off
 755                          *    last time
 756                          * last: done with this
 757                          * cur: the op we are messing with
 758                          * op: duh
 759                          * total: since we inserted the last key, we need to
 760                          *        decrement total so we dont overflow
 761                          */
 762                         i = last;
 763                         last = 0;
 764                         total--;
 765                         if (i < total) {
 766                                 cur = insert_list->next;
 767                                 op = list_entry(cur, struct pending_extent_op,
 768                                                 list);
 769                         }
 770                 } else {
 771                         i += ret;
 772                 }
 773
 774                 cond_resched();
 775         }
 776         ret = 0;
 777         kfree(keys);
 778         kfree(data_size);
 779         return ret;
 780 }
 781
 782 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
 783                                           struct btrfs_root *root,
 784                                           struct btrfs_path *path,
 785                                           u64 bytenr, u64 parent,
 786                                           u64 ref_root, u64 ref_generation,
 787                                           u64 owner_objectid)
 788 {
 789         struct btrfs_key key;
 790         struct extent_buffer *leaf;
 791         struct btrfs_extent_ref *ref;
 792         u32 num_refs;
 793         int ret;
 794
 795         key.objectid = bytenr;
 796         key.type = BTRFS_EXTENT_REF_KEY;
 797         key.offset = parent;
 798
 799         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
 800         if (ret == 0) {
 801                 leaf = path->nodes[0];
 802                 ref = btrfs_item_ptr(leaf, path->slots[0],
 803                                      struct btrfs_extent_ref);
 804                 btrfs_set_ref_root(leaf, ref, ref_root);
 805                 btrfs_set_ref_generation(leaf, ref, ref_generation);
 806                 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
 807                 btrfs_set_ref_num_refs(leaf, ref, 1);
 808         } else if (ret == -EEXIST) {
 809                 u64 existing_owner;
 810                 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
 811                 leaf = path->nodes[0];
 812                 ref = btrfs_item_ptr(leaf, path->slots[0],
 813                                      struct btrfs_extent_ref);
 814                 if (btrfs_ref_root(leaf, ref) != ref_root ||
 815                     btrfs_ref_generation(leaf, ref) != ref_generation) {
 816                         ret = -EIO;
 817                         WARN_ON(1);
 818                         goto out;
 819                 }
 820
 821                 num_refs = btrfs_ref_num_refs(leaf, ref);
 822                 BUG_ON(num_refs == 0);
 823                 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
 824
 825                 existing_owner = btrfs_ref_objectid(leaf, ref);
 826                 if (existing_owner != owner_objectid &&
 827                     existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
 828                         btrfs_set_ref_objectid(leaf, ref,
 829                                         BTRFS_MULTIPLE_OBJECTIDS);
 830                 }
 831                 ret = 0;
 832         } else {
 833                 goto out;
 834         }
 835         btrfs_mark_buffer_dirty(path->nodes[0]);
 836 out:
 837         btrfs_release_path(root, path);
 838         return ret;
 839 }
 840
 841 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
 842                                           struct btrfs_root *root,
 843                                           struct btrfs_path *path)
 844 {
 845         struct extent_buffer *leaf;
 846         struct btrfs_extent_ref *ref;
 847         u32 num_refs;
 848         int ret = 0;
 849
 850         leaf = path->nodes[0];
 851         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 852         num_refs = btrfs_ref_num_refs(leaf, ref);
 853         BUG_ON(num_refs == 0);
 854         num_refs -= 1;
 855         if (num_refs == 0) {
 856                 ret = btrfs_del_item(trans, root, path);
 857         } else {
 858                 btrfs_set_ref_num_refs(leaf, ref, num_refs);
 859                 btrfs_mark_buffer_dirty(leaf);
 860         }
 861         btrfs_release_path(root, path);
 862         return ret;
 863 }
 864
 865 #ifdef BIO_RW_DISCARD
 866 static void btrfs_issue_discard(struct block_device *bdev,
 867                                 u64 start, u64 len)
 868 {
 869         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
 870 }
 871 #endif
 872
 873 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
 874                                 u64 num_bytes)
 875 {
 876 #ifdef BIO_RW_DISCARD
 877         int ret;
 878         u64 map_length = num_bytes;
 879         struct btrfs_multi_bio *multi = NULL;
 880
 881         /* Tell the block device(s) that the sectors can be discarded */
 882         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
 883                               bytenr, &map_length, &multi, 0);
 884         if (!ret) {
 885                 struct btrfs_bio_stripe *stripe = multi->stripes;
 886                 int i;
 887
 888                 if (map_length > num_bytes)
 889                         map_length = num_bytes;
 890
 891                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
 892                         btrfs_issue_discard(stripe->dev->bdev,
 893                                             stripe->physical,
 894                                             map_length);
 895                 }
 896                 kfree(multi);
 897         }
 898
 899         return ret;
 900 #else
 901         return 0;
 902 #endif
 903 }
 904
 905 static noinline int free_extents(struct btrfs_trans_handle *trans,
 906                                  struct btrfs_root *extent_root,
 907                                  struct list_head *del_list)
 908 {
 909         struct btrfs_fs_info *info = extent_root->fs_info;
 910         struct btrfs_path *path;
 911         struct btrfs_key key, found_key;
 912         struct extent_buffer *leaf;
 913         struct list_head *cur;
 914         struct pending_extent_op *op;
 915         struct btrfs_extent_item *ei;
 916         int ret, num_to_del, extent_slot = 0, found_extent = 0;
 917         u32 refs;
 918         u64 bytes_freed = 0;
 919
 920         path = btrfs_alloc_path();
 921         if (!path)
 922                 return -ENOMEM;
 923         path->reada = 1;
 924
 925 search:
 926         /* search for the backref for the current ref we want to delete */
 927         cur = del_list->next;
 928         op = list_entry(cur, struct pending_extent_op, list);
 929         ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
 930                                     op->orig_parent,
 931                                     extent_root->root_key.objectid,
 932                                     op->orig_generation, op->level, 1);
 933         if (ret) {
 934                 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
 935                        "root %llu gen %llu owner %u\n",
 936                        (unsigned long long)op->bytenr,
 937                        (unsigned long long)extent_root->root_key.objectid,
 938                        (unsigned long long)op->orig_generation, op->level);
 939                 btrfs_print_leaf(extent_root, path->nodes[0]);
 940                 WARN_ON(1);
 941                 goto out;
 942         }
 943
 944         extent_slot = path->slots[0];
 945         num_to_del = 1;
 946         found_extent = 0;
 947
 948         /*
 949          * if we aren't the first item on the leaf we can move back one and see
 950          * if our ref is right next to our extent item
 951          */
 952         if (likely(extent_slot)) {
 953                 extent_slot--;
 954                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 955                                       extent_slot);
 956                 if (found_key.objectid == op->bytenr &&
 957                     found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 958                     found_key.offset == op->num_bytes) {
 959                         num_to_del++;
 960                         found_extent = 1;
 961                 }
 962         }
 963
 964         /*
 965          * if we didn't find the extent we need to delete the backref and then
 966          * search for the extent item key so we can update its ref count
 967          */
 968         if (!found_extent) {
 969                 key.objectid = op->bytenr;
 970                 key.type = BTRFS_EXTENT_ITEM_KEY;
 971                 key.offset = op->num_bytes;
 972
 973                 ret = remove_extent_backref(trans, extent_root, path);
 974                 BUG_ON(ret);
 975                 btrfs_release_path(extent_root, path);
 976                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
 977                 BUG_ON(ret);
 978                 extent_slot = path->slots[0];
 979         }
 980
 981         /* this is where we update the ref count for the extent */
 982         leaf = path->nodes[0];
 983         ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
 984         refs = btrfs_extent_refs(leaf, ei);
 985         BUG_ON(refs == 0);
 986         refs--;
 987         btrfs_set_extent_refs(leaf, ei, refs);
 988
 989         btrfs_mark_buffer_dirty(leaf);
 990
 991         /*
 992          * This extent needs deleting.  The reason cur_slot is extent_slot +
 993          * num_to_del is because extent_slot points to the slot where the extent
 994          * is, and if the backref was not right next to the extent we will be
 995          * deleting at least 1 item, and will want to start searching at the
 996          * slot directly next to extent_slot.  However if we did find the
 997          * backref next to the extent item them we will be deleting at least 2
 998          * items and will want to start searching directly after the ref slot
 999          */
1000         if (!refs) {
1001                 struct list_head *pos, *n, *end;
1002                 int cur_slot = extent_slot+num_to_del;
1003                 u64 super_used;
1004                 u64 root_used;
1005
1006                 path->slots[0] = extent_slot;
1007                 bytes_freed = op->num_bytes;
1008
1009                 mutex_lock(&info->pinned_mutex);
1010                 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1011                                      op->num_bytes, op->level >=
1012                                      BTRFS_FIRST_FREE_OBJECTID);
1013                 mutex_unlock(&info->pinned_mutex);
1014                 BUG_ON(ret < 0);
1015                 op->del = ret;
1016
1017                 /*
1018                  * we need to see if we can delete multiple things at once, so
1019                  * start looping through the list of extents we are wanting to
1020                  * delete and see if their extent/backref's are right next to
1021                  * eachother and the extents only have 1 ref
1022                  */
1023                 for (pos = cur->next; pos != del_list; pos = pos->next) {
1024                         struct pending_extent_op *tmp;
1025
1026                         tmp = list_entry(pos, struct pending_extent_op, list);
1027
1028                         /* we only want to delete extent+ref at this stage */
1029                         if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1030                                 break;
1031
1032                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1033                         if (found_key.objectid != tmp->bytenr ||
1034                             found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1035                             found_key.offset != tmp->num_bytes)
1036                                 break;
1037
1038                         /* check to make sure this extent only has one ref */
1039                         ei = btrfs_item_ptr(leaf, cur_slot,
1040                                             struct btrfs_extent_item);
1041                         if (btrfs_extent_refs(leaf, ei) != 1)
1042                                 break;
1043
1044                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1045                         if (found_key.objectid != tmp->bytenr ||
1046                             found_key.type != BTRFS_EXTENT_REF_KEY ||
1047                             found_key.offset != tmp->orig_parent)
1048                                 break;
1049
1050                         /*
1051                          * the ref is right next to the extent, we can set the
1052                          * ref count to 0 since we will delete them both now
1053                          */
1054                         btrfs_set_extent_refs(leaf, ei, 0);
1055
1056                         /* pin down the bytes for this extent */
1057                         mutex_lock(&info->pinned_mutex);
1058                         ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1059                                              tmp->num_bytes, tmp->level >=
1060                                              BTRFS_FIRST_FREE_OBJECTID);
1061                         mutex_unlock(&info->pinned_mutex);
1062                         BUG_ON(ret < 0);
1063
1064                         /*
1065                          * use the del field to tell if we need to go ahead and
1066                          * free up the extent when we delete the item or not.
1067                          */
1068                         tmp->del = ret;
1069                         bytes_freed += tmp->num_bytes;
1070
1071                         num_to_del += 2;
1072                         cur_slot += 2;
1073                 }
1074                 end = pos;
1075
1076                 /* update the free space counters */
1077                 spin_lock(&info->delalloc_lock);
1078                 super_used = btrfs_super_bytes_used(&info->super_copy);
1079                 btrfs_set_super_bytes_used(&info->super_copy,
1080                                            super_used - bytes_freed);
1081
1082                 root_used = btrfs_root_used(&extent_root->root_item);
1083                 btrfs_set_root_used(&extent_root->root_item,
1084                                     root_used - bytes_freed);
1085                 spin_unlock(&info->delalloc_lock);
1086
1087                 /* delete the items */
1088                 ret = btrfs_del_items(trans, extent_root, path,
1089                                       path->slots[0], num_to_del);
1090                 BUG_ON(ret);
1091
1092                 /*
1093                  * loop through the extents we deleted and do the cleanup work
1094                  * on them
1095                  */
1096                 for (pos = cur, n = pos->next; pos != end;
1097                      pos = n, n = pos->next) {
1098                         struct pending_extent_op *tmp;
1099                         tmp = list_entry(pos, struct pending_extent_op, list);
1100
1101                         /*
1102                          * remember tmp->del tells us wether or not we pinned
1103                          * down the extent
1104                          */
1105                         ret = update_block_group(trans, extent_root,
1106                                                  tmp->bytenr, tmp->num_bytes, 0,
1107                                                  tmp->del);
1108                         BUG_ON(ret);
1109
1110                         list_del_init(&tmp->list);
1111                         unlock_extent(&info->extent_ins, tmp->bytenr,
1112                                       tmp->bytenr + tmp->num_bytes - 1,
1113                                       GFP_NOFS);
1114                         kfree(tmp);
1115                 }
1116         } else if (refs && found_extent) {
1117                 /*
1118                  * the ref and extent were right next to eachother, but the
1119                  * extent still has a ref, so just free the backref and keep
1120                  * going
1121                  */
1122                 ret = remove_extent_backref(trans, extent_root, path);
1123                 BUG_ON(ret);
1124
1125                 list_del_init(&op->list);
1126                 unlock_extent(&info->extent_ins, op->bytenr,
1127                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1128                 kfree(op);
1129         } else {
1130                 /*
1131                  * the extent has multiple refs and the backref we were looking
1132                  * for was not right next to it, so just unlock and go next,
1133                  * we're good to go
1134                  */
1135                 list_del_init(&op->list);
1136                 unlock_extent(&info->extent_ins, op->bytenr,
1137                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1138                 kfree(op);
1139         }
1140
1141         btrfs_release_path(extent_root, path);
1142         if (!list_empty(del_list))
1143                 goto search;
1144
1145 out:
1146         btrfs_free_path(path);
1147         return ret;
1148 }
1149
1150 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1151                                      struct btrfs_root *root, u64 bytenr,
1152                                      u64 orig_parent, u64 parent,
1153                                      u64 orig_root, u64 ref_root,
1154                                      u64 orig_generation, u64 ref_generation,
1155                                      u64 owner_objectid)
1156 {
1157         int ret;
1158         struct btrfs_root *extent_root = root->fs_info->extent_root;
1159         struct btrfs_path *path;
1160
1161         if (root == root->fs_info->extent_root) {
1162                 struct pending_extent_op *extent_op;
1163                 u64 num_bytes;
1164
1165                 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1166                 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1167                 mutex_lock(&root->fs_info->extent_ins_mutex);
1168                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1169                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1170                         u64 priv;
1171                         ret = get_state_private(&root->fs_info->extent_ins,
1172                                                 bytenr, &priv);
1173                         BUG_ON(ret);
1174                         extent_op = (struct pending_extent_op *)
1175                                                         (unsigned long)priv;
1176                         BUG_ON(extent_op->parent != orig_parent);
1177                         BUG_ON(extent_op->generation != orig_generation);
1178
1179                         extent_op->parent = parent;
1180                         extent_op->generation = ref_generation;
1181                 } else {
1182                         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1183                         BUG_ON(!extent_op);
1184
1185                         extent_op->type = PENDING_BACKREF_UPDATE;
1186                         extent_op->bytenr = bytenr;
1187                         extent_op->num_bytes = num_bytes;
1188                         extent_op->parent = parent;
1189                         extent_op->orig_parent = orig_parent;
1190                         extent_op->generation = ref_generation;
1191                         extent_op->orig_generation = orig_generation;
1192                         extent_op->level = (int)owner_objectid;
1193                         INIT_LIST_HEAD(&extent_op->list);
1194                         extent_op->del = 0;
1195
1196                         set_extent_bits(&root->fs_info->extent_ins,
1197                                         bytenr, bytenr + num_bytes - 1,
1198                                         EXTENT_WRITEBACK, GFP_NOFS);
1199                         set_state_private(&root->fs_info->extent_ins,
1200                                           bytenr, (unsigned long)extent_op);
1201                 }
1202                 mutex_unlock(&root->fs_info->extent_ins_mutex);
1203                 return 0;
1204         }
1205
1206         path = btrfs_alloc_path();
1207         if (!path)
1208                 return -ENOMEM;
1209         ret = lookup_extent_backref(trans, extent_root, path,
1210                                     bytenr, orig_parent, orig_root,
1211                                     orig_generation, owner_objectid, 1);
1212         if (ret)
1213                 goto out;
1214         ret = remove_extent_backref(trans, extent_root, path);
1215         if (ret)
1216                 goto out;
1217         ret = insert_extent_backref(trans, extent_root, path, bytenr,
1218                                     parent, ref_root, ref_generation,
1219                                     owner_objectid);
1220         BUG_ON(ret);
1221         finish_current_insert(trans, extent_root, 0);
1222         del_pending_extents(trans, extent_root, 0);
1223 out:
1224         btrfs_free_path(path);
1225         return ret;
1226 }
1227
1228 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1229                             struct btrfs_root *root, u64 bytenr,
1230                             u64 orig_parent, u64 parent,
1231                             u64 ref_root, u64 ref_generation,
1232                             u64 owner_objectid)
1233 {
1234         int ret;
1235         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1236             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1237                 return 0;
1238         ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1239                                         parent, ref_root, ref_root,
1240                                         ref_generation, ref_generation,
1241                                         owner_objectid);
1242         return ret;
1243 }
1244
1245 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1246                                   struct btrfs_root *root, u64 bytenr,
1247                                   u64 orig_parent, u64 parent,
1248                                   u64 orig_root, u64 ref_root,
1249                                   u64 orig_generation, u64 ref_generation,
1250                                   u64 owner_objectid)
1251 {
1252         struct btrfs_path *path;
1253         int ret;
1254         struct btrfs_key key;
1255         struct extent_buffer *l;
1256         struct btrfs_extent_item *item;
1257         u32 refs;
1258
1259         path = btrfs_alloc_path();
1260         if (!path)
1261                 return -ENOMEM;
1262
1263         path->reada = 1;
1264         key.objectid = bytenr;
1265         key.type = BTRFS_EXTENT_ITEM_KEY;
1266         key.offset = (u64)-1;
1267
1268         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1269                                 0, 1);
1270         if (ret < 0)
1271                 return ret;
1272         BUG_ON(ret == 0 || path->slots[0] == 0);
1273
1274         path->slots[0]--;
1275         l = path->nodes[0];
1276
1277         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1278         if (key.objectid != bytenr) {
1279                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1280                 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
1281                        (unsigned long long)bytenr,
1282                        (unsigned long long)key.objectid);
1283                 BUG();
1284         }
1285         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1286
1287         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1288         refs = btrfs_extent_refs(l, item);
1289         btrfs_set_extent_refs(l, item, refs + 1);
1290         btrfs_mark_buffer_dirty(path->nodes[0]);
1291
1292         btrfs_release_path(root->fs_info->extent_root, path);
1293
1294         path->reada = 1;
1295         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1296                                     path, bytenr, parent,
1297                                     ref_root, ref_generation,
1298                                     owner_objectid);
1299         BUG_ON(ret);
1300         finish_current_insert(trans, root->fs_info->extent_root, 0);
1301         del_pending_extents(trans, root->fs_info->extent_root, 0);
1302
1303         btrfs_free_path(path);
1304         return 0;
1305 }
1306
1307 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1308                          struct btrfs_root *root,
1309                          u64 bytenr, u64 num_bytes, u64 parent,
1310                          u64 ref_root, u64 ref_generation,
1311                          u64 owner_objectid)
1312 {
1313         int ret;
1314         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1315             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1316                 return 0;
1317         ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1318                                      0, ref_root, 0, ref_generation,
1319                                      owner_objectid);
1320         return ret;
1321 }
1322
1323 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1324                          struct btrfs_root *root)
1325 {
1326         u64 start;
1327         u64 end;
1328         int ret;
1329
1330         while(1) {
1331                 finish_current_insert(trans, root->fs_info->extent_root, 1);
1332                 del_pending_extents(trans, root->fs_info->extent_root, 1);
1333
1334                 /* is there more work to do? */
1335                 ret = find_first_extent_bit(&root->fs_info->pending_del,
1336                                             0, &start, &end, EXTENT_WRITEBACK);
1337                 if (!ret)
1338                         continue;
1339                 ret = find_first_extent_bit(&root->fs_info->extent_ins,
1340                                             0, &start, &end, EXTENT_WRITEBACK);
1341                 if (!ret)
1342                         continue;
1343                 break;
1344         }
1345         return 0;
1346 }
1347
1348 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1349                             struct btrfs_root *root, u64 bytenr,
1350                             u64 num_bytes, u32 *refs)
1351 {
1352         struct btrfs_path *path;
1353         int ret;
1354         struct btrfs_key key;
1355         struct extent_buffer *l;
1356         struct btrfs_extent_item *item;
1357
1358         WARN_ON(num_bytes < root->sectorsize);
1359         path = btrfs_alloc_path();
1360         path->reada = 1;
1361         key.objectid = bytenr;
1362         key.offset = num_bytes;
1363         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1364         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1365                                 0, 0);
1366         if (ret < 0)
1367                 goto out;
1368         if (ret != 0) {
1369                 btrfs_print_leaf(root, path->nodes[0]);
1370                 printk(KERN_INFO "btrfs failed to find block number %llu\n",
1371                        (unsigned long long)bytenr);
1372                 BUG();
1373         }
1374         l = path->nodes[0];
1375         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1376         *refs = btrfs_extent_refs(l, item);
1377 out:
1378         btrfs_free_path(path);
1379         return 0;
1380 }
1381
1382 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1383                           struct btrfs_root *root, u64 objectid, u64 bytenr)
1384 {
1385         struct btrfs_root *extent_root = root->fs_info->extent_root;
1386         struct btrfs_path *path;
1387         struct extent_buffer *leaf;
1388         struct btrfs_extent_ref *ref_item;
1389         struct btrfs_key key;
1390         struct btrfs_key found_key;
1391         u64 ref_root;
1392         u64 last_snapshot;
1393         u32 nritems;
1394         int ret;
1395
1396         key.objectid = bytenr;
1397         key.offset = (u64)-1;
1398         key.type = BTRFS_EXTENT_ITEM_KEY;
1399
1400         path = btrfs_alloc_path();
1401         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1402         if (ret < 0)
1403                 goto out;
1404         BUG_ON(ret == 0);
1405
1406         ret = -ENOENT;
1407         if (path->slots[0] == 0)
1408                 goto out;
1409
1410         path->slots[0]--;
1411         leaf = path->nodes[0];
1412         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1413
1414         if (found_key.objectid != bytenr ||
1415             found_key.type != BTRFS_EXTENT_ITEM_KEY)
1416                 goto out;
1417
1418         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1419         while (1) {
1420                 leaf = path->nodes[0];
1421                 nritems = btrfs_header_nritems(leaf);
1422                 if (path->slots[0] >= nritems) {
1423                         ret = btrfs_next_leaf(extent_root, path);
1424                         if (ret < 0)
1425                                 goto out;
1426                         if (ret == 0)
1427                                 continue;
1428                         break;
1429                 }
1430                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1431                 if (found_key.objectid != bytenr)
1432                         break;
1433
1434                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1435                         path->slots[0]++;
1436                         continue;
1437                 }
1438
1439                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1440                                           struct btrfs_extent_ref);
1441                 ref_root = btrfs_ref_root(leaf, ref_item);
1442                 if ((ref_root != root->root_key.objectid &&
1443                      ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1444                      objectid != btrfs_ref_objectid(leaf, ref_item)) {
1445                         ret = 1;
1446                         goto out;
1447                 }
1448                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1449                         ret = 1;
1450                         goto out;
1451                 }
1452
1453                 path->slots[0]++;
1454         }
1455         ret = 0;
1456 out:
1457         btrfs_free_path(path);
1458         return ret;
1459 }
1460
1461 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1462                     struct extent_buffer *buf, u32 nr_extents)
1463 {
1464         struct btrfs_key key;
1465         struct btrfs_file_extent_item *fi;
1466         u64 root_gen;
1467         u32 nritems;
1468         int i;
1469         int level;
1470         int ret = 0;
1471         int shared = 0;
1472
1473         if (!root->ref_cows)
1474                 return 0;
1475
1476         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1477                 shared = 0;
1478                 root_gen = root->root_key.offset;
1479         } else {
1480                 shared = 1;
1481                 root_gen = trans->transid - 1;
1482         }
1483
1484         level = btrfs_header_level(buf);
1485         nritems = btrfs_header_nritems(buf);
1486
1487         if (level == 0) {
1488                 struct btrfs_leaf_ref *ref;
1489                 struct btrfs_extent_info *info;
1490
1491                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1492                 if (!ref) {
1493                         ret = -ENOMEM;
1494                         goto out;
1495                 }
1496
1497                 ref->root_gen = root_gen;
1498                 ref->bytenr = buf->start;
1499                 ref->owner = btrfs_header_owner(buf);
1500                 ref->generation = btrfs_header_generation(buf);
1501                 ref->nritems = nr_extents;
1502                 info = ref->extents;
1503
1504                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1505                         u64 disk_bytenr;
1506                         btrfs_item_key_to_cpu(buf, &key, i);
1507                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1508                                 continue;
1509                         fi = btrfs_item_ptr(buf, i,
1510                                             struct btrfs_file_extent_item);
1511                         if (btrfs_file_extent_type(buf, fi) ==
1512                             BTRFS_FILE_EXTENT_INLINE)
1513                                 continue;
1514                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1515                         if (disk_bytenr == 0)
1516                                 continue;
1517
1518                         info->bytenr = disk_bytenr;
1519                         info->num_bytes =
1520                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1521                         info->objectid = key.objectid;
1522                         info->offset = key.offset;
1523                         info++;
1524                 }
1525
1526                 ret = btrfs_add_leaf_ref(root, ref, shared);
1527                 if (ret == -EEXIST && shared) {
1528                         struct btrfs_leaf_ref *old;
1529                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1530                         BUG_ON(!old);
1531                         btrfs_remove_leaf_ref(root, old);
1532                         btrfs_free_leaf_ref(root, old);
1533                         ret = btrfs_add_leaf_ref(root, ref, shared);
1534                 }
1535                 WARN_ON(ret);
1536                 btrfs_free_leaf_ref(root, ref);
1537         }
1538 out:
1539         return ret;
1540 }
1541
1542 /* when a block goes through cow, we update the reference counts of
1543  * everything that block points to.  The internal pointers of the block
1544  * can be in just about any order, and it is likely to have clusters of
1545  * things that are close together and clusters of things that are not.
1546  *
1547  * To help reduce the seeks that come with updating all of these reference
1548  * counts, sort them by byte number before actual updates are done.
1549  *
1550  * struct refsort is used to match byte number to slot in the btree block.
1551  * we sort based on the byte number and then use the slot to actually
1552  * find the item.
1553  *
1554  * struct refsort is smaller than strcut btrfs_item and smaller than
1555  * struct btrfs_key_ptr.  Since we're currently limited to the page size
1556  * for a btree block, there's no way for a kmalloc of refsorts for a
1557  * single node to be bigger than a page.
1558  */
1559 struct refsort {
1560         u64 bytenr;
1561         u32 slot;
1562 };
1563
1564 /*
1565  * for passing into sort()
1566  */
1567 static int refsort_cmp(const void *a_void, const void *b_void)
1568 {
1569         const struct refsort *a = a_void;
1570         const struct refsort *b = b_void;
1571
1572         if (a->bytenr < b->bytenr)
1573                 return -1;
1574         if (a->bytenr > b->bytenr)
1575                 return 1;
1576         return 0;
1577 }
1578
1579
1580 noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1581                            struct btrfs_root *root,
1582                            struct extent_buffer *orig_buf,
1583                            struct extent_buffer *buf, u32 *nr_extents)
1584 {
1585         u64 bytenr;
1586         u64 ref_root;
1587         u64 orig_root;
1588         u64 ref_generation;
1589         u64 orig_generation;
1590         struct refsort *sorted;
1591         u32 nritems;
1592         u32 nr_file_extents = 0;
1593         struct btrfs_key key;
1594         struct btrfs_file_extent_item *fi;
1595         int i;
1596         int level;
1597         int ret = 0;
1598         int faili = 0;
1599         int refi = 0;
1600         int slot;
1601         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1602                             u64, u64, u64, u64, u64, u64, u64, u64);
1603
1604         ref_root = btrfs_header_owner(buf);
1605         ref_generation = btrfs_header_generation(buf);
1606         orig_root = btrfs_header_owner(orig_buf);
1607         orig_generation = btrfs_header_generation(orig_buf);
1608
1609         nritems = btrfs_header_nritems(buf);
1610         level = btrfs_header_level(buf);
1611
1612         sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS);
1613         BUG_ON(!sorted);
1614
1615         if (root->ref_cows) {
1616                 process_func = __btrfs_inc_extent_ref;
1617         } else {
1618                 if (level == 0 &&
1619                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1620                         goto out;
1621                 if (level != 0 &&
1622                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1623                         goto out;
1624                 process_func = __btrfs_update_extent_ref;
1625         }
1626
1627         /*
1628          * we make two passes through the items.  In the first pass we
1629          * only record the byte number and slot.  Then we sort based on
1630          * byte number and do the actual work based on the sorted results
1631          */
1632         for (i = 0; i < nritems; i++) {
1633                 cond_resched();
1634                 if (level == 0) {
1635                         btrfs_item_key_to_cpu(buf, &key, i);
1636                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1637                                 continue;
1638                         fi = btrfs_item_ptr(buf, i,
1639                                             struct btrfs_file_extent_item);
1640                         if (btrfs_file_extent_type(buf, fi) ==
1641                             BTRFS_FILE_EXTENT_INLINE)
1642                                 continue;
1643                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1644                         if (bytenr == 0)
1645                                 continue;
1646
1647                         nr_file_extents++;
1648                         sorted[refi].bytenr = bytenr;
1649                         sorted[refi].slot = i;
1650                         refi++;
1651                 } else {
1652                         bytenr = btrfs_node_blockptr(buf, i);
1653                         sorted[refi].bytenr = bytenr;
1654                         sorted[refi].slot = i;
1655                         refi++;
1656                 }
1657         }
1658         /*
1659          * if refi == 0, we didn't actually put anything into the sorted
1660          * array and we're done
1661          */
1662         if (refi == 0)
1663                 goto out;
1664
1665         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
1666
1667         for (i = 0; i < refi; i++) {
1668                 cond_resched();
1669                 slot = sorted[i].slot;
1670                 bytenr = sorted[i].bytenr;
1671
1672                 if (level == 0) {
1673                         btrfs_item_key_to_cpu(buf, &key, slot);
1674
1675                         ret = process_func(trans, root, bytenr,
1676                                            orig_buf->start, buf->start,
1677                                            orig_root, ref_root,
1678                                            orig_generation, ref_generation,
1679                                            key.objectid);
1680
1681                         if (ret) {
1682                                 faili = slot;
1683                                 WARN_ON(1);
1684                                 goto fail;
1685                         }
1686                 } else {
1687                         ret = process_func(trans, root, bytenr,
1688                                            orig_buf->start, buf->start,
1689                                            orig_root, ref_root,
1690                                            orig_generation, ref_generation,
1691                                            level - 1);
1692                         if (ret) {
1693                                 faili = slot;
1694                                 WARN_ON(1);
1695                                 goto fail;
1696                         }
1697                 }
1698         }
1699 out:
1700         kfree(sorted);
1701         if (nr_extents) {
1702                 if (level == 0)
1703                         *nr_extents = nr_file_extents;
1704                 else
1705                         *nr_extents = nritems;
1706         }
1707         return 0;
1708 fail:
1709         kfree(sorted);
1710         WARN_ON(1);
1711         return ret;
1712 }
1713
1714 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1715                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1716                      struct extent_buffer *buf, int start_slot, int nr)
1717
1718 {
1719         u64 bytenr;
1720         u64 ref_root;
1721         u64 orig_root;
1722         u64 ref_generation;
1723         u64 orig_generation;
1724         struct btrfs_key key;
1725         struct btrfs_file_extent_item *fi;
1726         int i;
1727         int ret;
1728         int slot;
1729         int level;
1730
1731         BUG_ON(start_slot < 0);
1732         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1733
1734         ref_root = btrfs_header_owner(buf);
1735         ref_generation = btrfs_header_generation(buf);
1736         orig_root = btrfs_header_owner(orig_buf);
1737         orig_generation = btrfs_header_generation(orig_buf);
1738         level = btrfs_header_level(buf);
1739
1740         if (!root->ref_cows) {
1741                 if (level == 0 &&
1742                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1743                         return 0;
1744                 if (level != 0 &&
1745                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1746                         return 0;
1747         }
1748
1749         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1750                 cond_resched();
1751                 if (level == 0) {
1752                         btrfs_item_key_to_cpu(buf, &key, slot);
1753                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1754                                 continue;
1755                         fi = btrfs_item_ptr(buf, slot,
1756                                             struct btrfs_file_extent_item);
1757                         if (btrfs_file_extent_type(buf, fi) ==
1758                             BTRFS_FILE_EXTENT_INLINE)
1759                                 continue;
1760                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1761                         if (bytenr == 0)
1762                                 continue;
1763                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1764                                             orig_buf->start, buf->start,
1765                                             orig_root, ref_root,
1766                                             orig_generation, ref_generation,
1767                                             key.objectid);
1768                         if (ret)
1769                                 goto fail;
1770                 } else {
1771                         bytenr = btrfs_node_blockptr(buf, slot);
1772                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1773                                             orig_buf->start, buf->start,
1774                                             orig_root, ref_root,
1775                                             orig_generation, ref_generation,
1776                                             level - 1);
1777                         if (ret)
1778                                 goto fail;
1779                 }
1780         }
1781         return 0;
1782 fail:
1783         WARN_ON(1);
1784         return -1;
1785 }
1786
1787 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1788                                  struct btrfs_root *root,
1789                                  struct btrfs_path *path,
1790                                  struct btrfs_block_group_cache *cache)
1791 {
1792         int ret;
1793         int pending_ret;
1794         struct btrfs_root *extent_root = root->fs_info->extent_root;
1795         unsigned long bi;
1796         struct extent_buffer *leaf;
1797
1798         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1799         if (ret < 0)
1800                 goto fail;
1801         BUG_ON(ret);
1802
1803         leaf = path->nodes[0];
1804         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1805         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1806         btrfs_mark_buffer_dirty(leaf);
1807         btrfs_release_path(extent_root, path);
1808 fail:
1809         finish_current_insert(trans, extent_root, 0);
1810         pending_ret = del_pending_extents(trans, extent_root, 0);
1811         if (ret)
1812                 return ret;
1813         if (pending_ret)
1814                 return pending_ret;
1815         return 0;
1816
1817 }
1818
1819 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1820                                    struct btrfs_root *root)
1821 {
1822         struct btrfs_block_group_cache *cache, *entry;
1823         struct rb_node *n;
1824         int err = 0;
1825         int werr = 0;
1826         struct btrfs_path *path;
1827         u64 last = 0;
1828
1829         path = btrfs_alloc_path();
1830         if (!path)
1831                 return -ENOMEM;
1832
1833         while (1) {
1834                 cache = NULL;
1835                 spin_lock(&root->fs_info->block_group_cache_lock);
1836                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1837                      n; n = rb_next(n)) {
1838                         entry = rb_entry(n, struct btrfs_block_group_cache,
1839                                          cache_node);
1840                         if (entry->dirty) {
1841                                 cache = entry;
1842                                 break;
1843                         }
1844                 }
1845                 spin_unlock(&root->fs_info->block_group_cache_lock);
1846
1847                 if (!cache)
1848                         break;
1849
1850                 cache->dirty = 0;
1851                 last += cache->key.offset;
1852
1853                 err = write_one_cache_group(trans, root,
1854                                             path, cache);
1855                 /*
1856                  * if we fail to write the cache group, we want
1857                  * to keep it marked dirty in hopes that a later
1858                  * write will work
1859                  */
1860                 if (err) {
1861                         werr = err;
1862                         continue;
1863                 }
1864         }
1865         btrfs_free_path(path);
1866         return werr;
1867 }
1868
1869 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1870 {
1871         struct btrfs_block_group_cache *block_group;
1872         int readonly = 0;
1873
1874         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1875         if (!block_group || block_group->ro)
1876                 readonly = 1;
1877         if (block_group)
1878                 put_block_group(block_group);
1879         return readonly;
1880 }
1881
1882 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1883                              u64 total_bytes, u64 bytes_used,
1884                              struct btrfs_space_info **space_info)
1885 {
1886         struct btrfs_space_info *found;
1887
1888         found = __find_space_info(info, flags);
1889         if (found) {
1890                 spin_lock(&found->lock);
1891                 found->total_bytes += total_bytes;
1892                 found->bytes_used += bytes_used;
1893                 found->full = 0;
1894                 spin_unlock(&found->lock);
1895                 *space_info = found;
1896                 return 0;
1897         }
1898         found = kzalloc(sizeof(*found), GFP_NOFS);
1899         if (!found)
1900                 return -ENOMEM;
1901
1902         list_add(&found->list, &info->space_info);
1903         INIT_LIST_HEAD(&found->block_groups);
1904         init_rwsem(&found->groups_sem);
1905         spin_lock_init(&found->lock);
1906         found->flags = flags;
1907         found->total_bytes = total_bytes;
1908         found->bytes_used = bytes_used;
1909         found->bytes_pinned = 0;
1910         found->bytes_reserved = 0;
1911         found->bytes_readonly = 0;
1912         found->full = 0;
1913         found->force_alloc = 0;
1914         *space_info = found;
1915         return 0;
1916 }
1917
1918 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1919 {
1920         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1921                                    BTRFS_BLOCK_GROUP_RAID1 |
1922                                    BTRFS_BLOCK_GROUP_RAID10 |
1923                                    BTRFS_BLOCK_GROUP_DUP);
1924         if (extra_flags) {
1925                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1926                         fs_info->avail_data_alloc_bits |= extra_flags;
1927                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1928                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1929                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1930                         fs_info->avail_system_alloc_bits |= extra_flags;
1931         }
1932 }
1933
1934 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1935 {
1936         spin_lock(&cache->space_info->lock);
1937         spin_lock(&cache->lock);
1938         if (!cache->ro) {
1939                 cache->space_info->bytes_readonly += cache->key.offset -
1940                                         btrfs_block_group_used(&cache->item);
1941                 cache->ro = 1;
1942         }
1943         spin_unlock(&cache->lock);
1944         spin_unlock(&cache->space_info->lock);
1945 }
1946
1947 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1948 {
1949         u64 num_devices = root->fs_info->fs_devices->rw_devices;
1950
1951         if (num_devices == 1)
1952                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1953         if (num_devices < 4)
1954                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1955
1956         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1957             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1958                       BTRFS_BLOCK_GROUP_RAID10))) {
1959                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1960         }
1961
1962         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1963             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1964                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1965         }
1966
1967         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1968             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1969              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1970              (flags & BTRFS_BLOCK_GROUP_DUP)))
1971                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1972         return flags;
1973 }
1974
1975 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1976                           struct btrfs_root *extent_root, u64 alloc_bytes,
1977                           u64 flags, int force)
1978 {
1979         struct btrfs_space_info *space_info;
1980         u64 thresh;
1981         int ret = 0;
1982
1983         mutex_lock(&extent_root->fs_info->chunk_mutex);
1984
1985         flags = btrfs_reduce_alloc_profile(extent_root, flags);
1986
1987         space_info = __find_space_info(extent_root->fs_info, flags);
1988         if (!space_info) {
1989                 ret = update_space_info(extent_root->fs_info, flags,
1990                                         0, 0, &space_info);
1991                 BUG_ON(ret);
1992         }
1993         BUG_ON(!space_info);
1994
1995         spin_lock(&space_info->lock);
1996         if (space_info->force_alloc) {
1997                 force = 1;
1998                 space_info->force_alloc = 0;
1999         }
2000         if (space_info->full) {
2001                 spin_unlock(&space_info->lock);
2002                 goto out;
2003         }
2004
2005         thresh = space_info->total_bytes - space_info->bytes_readonly;
2006         thresh = div_factor(thresh, 6);
2007         if (!force &&
2008            (space_info->bytes_used + space_info->bytes_pinned +
2009             space_info->bytes_reserved + alloc_bytes) < thresh) {
2010                 spin_unlock(&space_info->lock);
2011                 goto out;
2012         }
2013         spin_unlock(&space_info->lock);
2014
2015         ret = btrfs_alloc_chunk(trans, extent_root, flags);
2016         if (ret)
2017                 space_info->full = 1;
2018 out:
2019         mutex_unlock(&extent_root->fs_info->chunk_mutex);
2020         return ret;
2021 }
2022
2023 static int update_block_group(struct btrfs_trans_handle *trans,
2024                               struct btrfs_root *root,
2025                               u64 bytenr, u64 num_bytes, int alloc,
2026                               int mark_free)
2027 {
2028         struct btrfs_block_group_cache *cache;
2029         struct btrfs_fs_info *info = root->fs_info;
2030         u64 total = num_bytes;
2031         u64 old_val;
2032         u64 byte_in_group;
2033
2034         while (total) {
2035                 cache = btrfs_lookup_block_group(info, bytenr);
2036                 if (!cache)
2037                         return -1;
2038                 byte_in_group = bytenr - cache->key.objectid;
2039                 WARN_ON(byte_in_group > cache->key.offset);
2040
2041                 spin_lock(&cache->space_info->lock);
2042                 spin_lock(&cache->lock);
2043                 cache->dirty = 1;
2044                 old_val = btrfs_block_group_used(&cache->item);
2045                 num_bytes = min(total, cache->key.offset - byte_in_group);
2046                 if (alloc) {
2047                         old_val += num_bytes;
2048                         cache->space_info->bytes_used += num_bytes;
2049                         if (cache->ro)
2050                                 cache->space_info->bytes_readonly -= num_bytes;
2051                         btrfs_set_block_group_used(&cache->item, old_val);
2052                         spin_unlock(&cache->lock);
2053                         spin_unlock(&cache->space_info->lock);
2054                 } else {
2055                         old_val -= num_bytes;
2056                         cache->space_info->bytes_used -= num_bytes;
2057                         if (cache->ro)
2058                                 cache->space_info->bytes_readonly += num_bytes;
2059                         btrfs_set_block_group_used(&cache->item, old_val);
2060                         spin_unlock(&cache->lock);
2061                         spin_unlock(&cache->space_info->lock);
2062                         if (mark_free) {
2063                                 int ret;
2064
2065                                 ret = btrfs_discard_extent(root, bytenr,
2066                                                            num_bytes);
2067                                 WARN_ON(ret);
2068
2069                                 ret = btrfs_add_free_space(cache, bytenr,
2070                                                            num_bytes);
2071                                 WARN_ON(ret);
2072                         }
2073                 }
2074                 put_block_group(cache);
2075                 total -= num_bytes;
2076                 bytenr += num_bytes;
2077         }
2078         return 0;
2079 }
2080
2081 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2082 {
2083         struct btrfs_block_group_cache *cache;
2084         u64 bytenr;
2085
2086         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2087         if (!cache)
2088                 return 0;
2089
2090         bytenr = cache->key.objectid;
2091         put_block_group(cache);
2092
2093         return bytenr;
2094 }
2095
2096 int btrfs_update_pinned_extents(struct btrfs_root *root,
2097                                 u64 bytenr, u64 num, int pin)
2098 {
2099         u64 len;
2100         struct btrfs_block_group_cache *cache;
2101         struct btrfs_fs_info *fs_info = root->fs_info;
2102
2103         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2104         if (pin) {
2105                 set_extent_dirty(&fs_info->pinned_extents,
2106                                 bytenr, bytenr + num - 1, GFP_NOFS);
2107         } else {
2108                 clear_extent_dirty(&fs_info->pinned_extents,
2109                                 bytenr, bytenr + num - 1, GFP_NOFS);
2110         }
2111         while (num > 0) {
2112                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2113                 BUG_ON(!cache);
2114                 len = min(num, cache->key.offset -
2115                           (bytenr - cache->key.objectid));
2116                 if (pin) {
2117                         spin_lock(&cache->space_info->lock);
2118                         spin_lock(&cache->lock);
2119                         cache->pinned += len;
2120                         cache->space_info->bytes_pinned += len;
2121                         spin_unlock(&cache->lock);
2122                         spin_unlock(&cache->space_info->lock);
2123                         fs_info->total_pinned += len;
2124                 } else {
2125                         spin_lock(&cache->space_info->lock);
2126                         spin_lock(&cache->lock);
2127                         cache->pinned -= len;
2128                         cache->space_info->bytes_pinned -= len;
2129                         spin_unlock(&cache->lock);
2130                         spin_unlock(&cache->space_info->lock);
2131                         fs_info->total_pinned -= len;
2132                         if (cache->cached)
2133                                 btrfs_add_free_space(cache, bytenr, len);
2134                 }
2135                 put_block_group(cache);
2136                 bytenr += len;
2137                 num -= len;
2138         }
2139         return 0;
2140 }
2141
2142 static int update_reserved_extents(struct btrfs_root *root,
2143                                    u64 bytenr, u64 num, int reserve)
2144 {
2145         u64 len;
2146         struct btrfs_block_group_cache *cache;
2147         struct btrfs_fs_info *fs_info = root->fs_info;
2148
2149         while (num > 0) {
2150                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2151                 BUG_ON(!cache);
2152                 len = min(num, cache->key.offset -
2153                           (bytenr - cache->key.objectid));
2154
2155                 spin_lock(&cache->space_info->lock);
2156                 spin_lock(&cache->lock);
2157                 if (reserve) {
2158                         cache->reserved += len;
2159                         cache->space_info->bytes_reserved += len;
2160                 } else {
2161                         cache->reserved -= len;
2162                         cache->space_info->bytes_reserved -= len;
2163                 }
2164                 spin_unlock(&cache->lock);
2165                 spin_unlock(&cache->space_info->lock);
2166                 put_block_group(cache);
2167                 bytenr += len;
2168                 num -= len;
2169         }
2170         return 0;
2171 }
2172
2173 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2174 {
2175         u64 last = 0;
2176         u64 start;
2177         u64 end;
2178         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2179         int ret;
2180
2181         mutex_lock(&root->fs_info->pinned_mutex);
2182         while (1) {
2183                 ret = find_first_extent_bit(pinned_extents, last,
2184                                             &start, &end, EXTENT_DIRTY);
2185                 if (ret)
2186                         break;
2187                 set_extent_dirty(copy, start, end, GFP_NOFS);
2188                 last = end + 1;
2189         }
2190         mutex_unlock(&root->fs_info->pinned_mutex);
2191         return 0;
2192 }
2193
2194 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2195                                struct btrfs_root *root,
2196                                struct extent_io_tree *unpin)
2197 {
2198         u64 start;
2199         u64 end;
2200         int ret;
2201
2202         mutex_lock(&root->fs_info->pinned_mutex);
2203         while (1) {
2204                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2205                                             EXTENT_DIRTY);
2206                 if (ret)
2207                         break;
2208
2209                 ret = btrfs_discard_extent(root, start, end + 1 - start);
2210
2211                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2212                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2213
2214                 if (need_resched()) {
2215                         mutex_unlock(&root->fs_info->pinned_mutex);
2216                         cond_resched();
2217                         mutex_lock(&root->fs_info->pinned_mutex);
2218                 }
2219         }
2220         mutex_unlock(&root->fs_info->pinned_mutex);
2221         return ret;
2222 }
2223
2224 static int finish_current_insert(struct btrfs_trans_handle *trans,
2225                                  struct btrfs_root *extent_root, int all)
2226 {
2227         u64 start;
2228         u64 end;
2229         u64 priv;
2230         u64 search = 0;
2231         struct btrfs_fs_info *info = extent_root->fs_info;
2232         struct btrfs_path *path;
2233         struct pending_extent_op *extent_op, *tmp;
2234         struct list_head insert_list, update_list;
2235         int ret;
2236         int num_inserts = 0, max_inserts, restart = 0;
2237
2238         path = btrfs_alloc_path();
2239         INIT_LIST_HEAD(&insert_list);
2240         INIT_LIST_HEAD(&update_list);
2241
2242         max_inserts = extent_root->leafsize /
2243                 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2244                  sizeof(struct btrfs_extent_ref) +
2245                  sizeof(struct btrfs_extent_item));
2246 again:
2247         mutex_lock(&info->extent_ins_mutex);
2248         while (1) {
2249                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2250                                             &end, EXTENT_WRITEBACK);
2251                 if (ret) {
2252                         if (restart && !num_inserts &&
2253                             list_empty(&update_list)) {
2254                                 restart = 0;
2255                                 search = 0;
2256                                 continue;
2257                         }
2258                         break;
2259                 }
2260
2261                 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2262                 if (!ret) {
2263                         if (all)
2264                                 restart = 1;
2265                         search = end + 1;
2266                         if (need_resched()) {
2267                                 mutex_unlock(&info->extent_ins_mutex);
2268                                 cond_resched();
2269                                 mutex_lock(&info->extent_ins_mutex);
2270                         }
2271                         continue;
2272                 }
2273
2274                 ret = get_state_private(&info->extent_ins, start, &priv);
2275                 BUG_ON(ret);
2276                 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2277
2278                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2279                         num_inserts++;
2280                         list_add_tail(&extent_op->list, &insert_list);
2281                         search = end + 1;
2282                         if (num_inserts == max_inserts) {
2283                                 restart = 1;
2284                                 break;
2285                         }
2286                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2287                         list_add_tail(&extent_op->list, &update_list);
2288                         search = end + 1;
2289                 } else {
2290                         BUG();
2291                 }
2292         }
2293
2294         /*
2295          * process the update list, clear the writeback bit for it, and if
2296          * somebody marked this thing for deletion then just unlock it and be
2297          * done, the free_extents will handle it
2298          */
2299         list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2300                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2301                                   extent_op->bytenr + extent_op->num_bytes - 1,
2302                                   EXTENT_WRITEBACK, GFP_NOFS);
2303                 if (extent_op->del) {
2304                         list_del_init(&extent_op->list);
2305                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2306                                       extent_op->bytenr + extent_op->num_bytes
2307                                       - 1, GFP_NOFS);
2308                         kfree(extent_op);
2309                 }
2310         }
2311         mutex_unlock(&info->extent_ins_mutex);
2312
2313         /*
2314          * still have things left on the update list, go ahead an update
2315          * everything
2316          */
2317         if (!list_empty(&update_list)) {
2318                 ret = update_backrefs(trans, extent_root, path, &update_list);
2319                 BUG_ON(ret);
2320
2321                 /* we may have COW'ed new blocks, so lets start over */
2322                 if (all)
2323                         restart = 1;
2324         }
2325
2326         /*
2327          * if no inserts need to be done, but we skipped some extents and we
2328          * need to make sure everything is cleaned then reset everything and
2329          * go back to the beginning
2330          */
2331         if (!num_inserts && restart) {
2332                 search = 0;
2333                 restart = 0;
2334                 INIT_LIST_HEAD(&update_list);
2335                 INIT_LIST_HEAD(&insert_list);
2336                 goto again;
2337         } else if (!num_inserts) {
2338                 goto out;
2339         }
2340
2341         /*
2342          * process the insert extents list.  Again if we are deleting this
2343          * extent, then just unlock it, pin down the bytes if need be, and be
2344          * done with it.  Saves us from having to actually insert the extent
2345          * into the tree and then subsequently come along and delete it
2346          */
2347         mutex_lock(&info->extent_ins_mutex);
2348         list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2349                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2350                                   extent_op->bytenr + extent_op->num_bytes - 1,
2351                                   EXTENT_WRITEBACK, GFP_NOFS);
2352                 if (extent_op->del) {
2353                         u64 used;
2354                         list_del_init(&extent_op->list);
2355                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2356                                       extent_op->bytenr + extent_op->num_bytes
2357                                       - 1, GFP_NOFS);
2358
2359                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2360                         ret = pin_down_bytes(trans, extent_root,
2361                                              extent_op->bytenr,
2362                                              extent_op->num_bytes, 0);
2363                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2364
2365                         spin_lock(&info->delalloc_lock);
2366                         used = btrfs_super_bytes_used(&info->super_copy);
2367                         btrfs_set_super_bytes_used(&info->super_copy,
2368                                         used - extent_op->num_bytes);
2369                         used = btrfs_root_used(&extent_root->root_item);
2370                         btrfs_set_root_used(&extent_root->root_item,
2371                                         used - extent_op->num_bytes);
2372                         spin_unlock(&info->delalloc_lock);
2373
2374                         ret = update_block_group(trans, extent_root,
2375                                                  extent_op->bytenr,
2376                                                  extent_op->num_bytes,
2377                                                  0, ret > 0);
2378                         BUG_ON(ret);
2379                         kfree(extent_op);
2380                         num_inserts--;
2381                 }
2382         }
2383         mutex_unlock(&info->extent_ins_mutex);
2384
2385         ret = insert_extents(trans, extent_root, path, &insert_list,
2386                              num_inserts);
2387         BUG_ON(ret);
2388
2389         /*
2390          * if restart is set for whatever reason we need to go back and start
2391          * searching through the pending list again.
2392          *
2393          * We just inserted some extents, which could have resulted in new
2394          * blocks being allocated, which would result in new blocks needing
2395          * updates, so if all is set we _must_ restart to get the updated
2396          * blocks.
2397          */
2398         if (restart || all) {
2399                 INIT_LIST_HEAD(&insert_list);
2400                 INIT_LIST_HEAD(&update_list);
2401                 search = 0;
2402                 restart = 0;
2403                 num_inserts = 0;
2404                 goto again;
2405         }
2406 out:
2407         btrfs_free_path(path);
2408         return 0;
2409 }
2410
2411 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2412                           struct btrfs_root *root,
2413                           u64 bytenr, u64 num_bytes, int is_data)
2414 {
2415         int err = 0;
2416         struct extent_buffer *buf;
2417
2418         if (is_data)
2419                 goto pinit;
2420
2421         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2422         if (!buf)
2423                 goto pinit;
2424
2425         /* we can reuse a block if it hasn't been written
2426          * and it is from this transaction.  We can't
2427          * reuse anything from the tree log root because
2428          * it has tiny sub-transactions.
2429          */
2430         if (btrfs_buffer_uptodate(buf, 0) &&
2431             btrfs_try_tree_lock(buf)) {
2432                 u64 header_owner = btrfs_header_owner(buf);
2433                 u64 header_transid = btrfs_header_generation(buf);
2434                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2435                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2436                     header_transid == trans->transid &&
2437                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2438                         clean_tree_block(NULL, root, buf);
2439                         btrfs_tree_unlock(buf);
2440                         free_extent_buffer(buf);
2441                         return 1;
2442                 }
2443                 btrfs_tree_unlock(buf);
2444         }
2445         free_extent_buffer(buf);
2446 pinit:
2447         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2448
2449         BUG_ON(err < 0);
2450         return 0;
2451 }
2452
2453 /*
2454  * remove an extent from the root, returns 0 on success
2455  */
2456 static int __free_extent(struct btrfs_trans_handle *trans,
2457                          struct btrfs_root *root,
2458                          u64 bytenr, u64 num_bytes, u64 parent,
2459                          u64 root_objectid, u64 ref_generation,
2460                          u64 owner_objectid, int pin, int mark_free)
2461 {
2462         struct btrfs_path *path;
2463         struct btrfs_key key;
2464         struct btrfs_fs_info *info = root->fs_info;
2465         struct btrfs_root *extent_root = info->extent_root;
2466         struct extent_buffer *leaf;
2467         int ret;
2468         int extent_slot = 0;
2469         int found_extent = 0;
2470         int num_to_del = 1;
2471         struct btrfs_extent_item *ei;
2472         u32 refs;
2473
2474         key.objectid = bytenr;
2475         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2476         key.offset = num_bytes;
2477         path = btrfs_alloc_path();
2478         if (!path)
2479                 return -ENOMEM;
2480
2481         path->reada = 1;
2482         ret = lookup_extent_backref(trans, extent_root, path,
2483                                     bytenr, parent, root_objectid,
2484                                     ref_generation, owner_objectid, 1);
2485         if (ret == 0) {
2486                 struct btrfs_key found_key;
2487                 extent_slot = path->slots[0];
2488                 while (extent_slot > 0) {
2489                         extent_slot--;
2490                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2491                                               extent_slot);
2492                         if (found_key.objectid != bytenr)
2493                                 break;
2494                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2495                             found_key.offset == num_bytes) {
2496                                 found_extent = 1;
2497                                 break;
2498                         }
2499                         if (path->slots[0] - extent_slot > 5)
2500                                 break;
2501                 }
2502                 if (!found_extent) {
2503                         ret = remove_extent_backref(trans, extent_root, path);
2504                         BUG_ON(ret);
2505                         btrfs_release_path(extent_root, path);
2506                         ret = btrfs_search_slot(trans, extent_root,
2507                                                 &key, path, -1, 1);
2508                         if (ret) {
2509                                 printk(KERN_ERR "umm, got %d back from search"
2510                                        ", was looking for %llu\n", ret,
2511                                        (unsigned long long)bytenr);
2512                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2513                         }
2514                         BUG_ON(ret);
2515                         extent_slot = path->slots[0];
2516                 }
2517         } else {
2518                 btrfs_print_leaf(extent_root, path->nodes[0]);
2519                 WARN_ON(1);
2520                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2521                        "root %llu gen %llu owner %llu\n",
2522                        (unsigned long long)bytenr,
2523                        (unsigned long long)root_objectid,
2524                        (unsigned long long)ref_generation,
2525                        (unsigned long long)owner_objectid);
2526         }
2527
2528         leaf = path->nodes[0];
2529         ei = btrfs_item_ptr(leaf, extent_slot,
2530                             struct btrfs_extent_item);
2531         refs = btrfs_extent_refs(leaf, ei);
2532         BUG_ON(refs == 0);
2533         refs -= 1;
2534         btrfs_set_extent_refs(leaf, ei, refs);
2535
2536         btrfs_mark_buffer_dirty(leaf);
2537
2538         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2539                 struct btrfs_extent_ref *ref;
2540                 ref = btrfs_item_ptr(leaf, path->slots[0],
2541                                      struct btrfs_extent_ref);
2542                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2543                 /* if the back ref and the extent are next to each other
2544                  * they get deleted below in one shot
2545                  */
2546                 path->slots[0] = extent_slot;
2547                 num_to_del = 2;
2548         } else if (found_extent) {
2549                 /* otherwise delete the extent back ref */
2550                 ret = remove_extent_backref(trans, extent_root, path);
2551                 BUG_ON(ret);
2552                 /* if refs are 0, we need to setup the path for deletion */
2553                 if (refs == 0) {
2554                         btrfs_release_path(extent_root, path);
2555                         ret = btrfs_search_slot(trans, extent_root, &key, path,
2556                                                 -1, 1);
2557                         BUG_ON(ret);
2558                 }
2559         }
2560
2561         if (refs == 0) {
2562                 u64 super_used;
2563                 u64 root_used;
2564
2565                 if (pin) {
2566                         mutex_lock(&root->fs_info->pinned_mutex);
2567                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2568                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2569                         mutex_unlock(&root->fs_info->pinned_mutex);
2570                         if (ret > 0)
2571                                 mark_free = 1;
2572                         BUG_ON(ret < 0);
2573                 }
2574                 /* block accounting for super block */
2575                 spin_lock(&info->delalloc_lock);
2576                 super_used = btrfs_super_bytes_used(&info->super_copy);
2577                 btrfs_set_super_bytes_used(&info->super_copy,
2578                                            super_used - num_bytes);
2579
2580                 /* block accounting for root item */
2581                 root_used = btrfs_root_used(&root->root_item);
2582                 btrfs_set_root_used(&root->root_item,
2583                                            root_used - num_bytes);
2584                 spin_unlock(&info->delalloc_lock);
2585                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2586                                       num_to_del);
2587                 BUG_ON(ret);
2588                 btrfs_release_path(extent_root, path);
2589
2590                 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2591                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2592                         BUG_ON(ret);
2593                 }
2594
2595                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2596                                          mark_free);
2597                 BUG_ON(ret);
2598         }
2599         btrfs_free_path(path);
2600         finish_current_insert(trans, extent_root, 0);
2601         return ret;
2602 }
2603
2604 /*
2605  * find all the blocks marked as pending in the radix tree and remove
2606  * them from the extent map
2607  */
2608 static int del_pending_extents(struct btrfs_trans_handle *trans,
2609                                struct btrfs_root *extent_root, int all)
2610 {
2611         int ret;
2612         int err = 0;
2613         u64 start;
2614         u64 end;
2615         u64 priv;
2616         u64 search = 0;
2617         int nr = 0, skipped = 0;
2618         struct extent_io_tree *pending_del;
2619         struct extent_io_tree *extent_ins;
2620         struct pending_extent_op *extent_op;
2621         struct btrfs_fs_info *info = extent_root->fs_info;
2622         struct list_head delete_list;
2623
2624         INIT_LIST_HEAD(&delete_list);
2625         extent_ins = &extent_root->fs_info->extent_ins;
2626         pending_del = &extent_root->fs_info->pending_del;
2627
2628 again:
2629         mutex_lock(&info->extent_ins_mutex);
2630         while (1) {
2631                 ret = find_first_extent_bit(pending_del, search, &start, &end,
2632                                             EXTENT_WRITEBACK);
2633                 if (ret) {
2634                         if (all && skipped && !nr) {
2635                                 search = 0;
2636                                 skipped = 0;
2637                                 continue;
2638                         }
2639                         mutex_unlock(&info->extent_ins_mutex);
2640                         break;
2641                 }
2642
2643                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2644                 if (!ret) {
2645                         search = end+1;
2646                         skipped = 1;
2647
2648                         if (need_resched()) {
2649                                 mutex_unlock(&info->extent_ins_mutex);
2650                                 cond_resched();
2651                                 mutex_lock(&info->extent_ins_mutex);
2652                         }
2653
2654                         continue;
2655                 }
2656                 BUG_ON(ret < 0);
2657
2658                 ret = get_state_private(pending_del, start, &priv);
2659                 BUG_ON(ret);
2660                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2661
2662                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2663                                   GFP_NOFS);
2664                 if (!test_range_bit(extent_ins, start, end,
2665                                     EXTENT_WRITEBACK, 0)) {
2666                         list_add_tail(&extent_op->list, &delete_list);
2667                         nr++;
2668                 } else {
2669                         kfree(extent_op);
2670
2671                         ret = get_state_private(&info->extent_ins, start,
2672                                                 &priv);
2673                         BUG_ON(ret);
2674                         extent_op = (struct pending_extent_op *)
2675                                                 (unsigned long)priv;
2676
2677                         clear_extent_bits(&info->extent_ins, start, end,
2678                                           EXTENT_WRITEBACK, GFP_NOFS);
2679
2680                         if (extent_op->type == PENDING_BACKREF_UPDATE) {
2681                                 list_add_tail(&extent_op->list, &delete_list);
2682                                 search = end + 1;
2683                                 nr++;
2684                                 continue;
2685                         }
2686
2687                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2688                         ret = pin_down_bytes(trans, extent_root, start,
2689                                              end + 1 - start, 0);
2690                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2691
2692                         ret = update_block_group(trans, extent_root, start,
2693                                                 end + 1 - start, 0, ret > 0);
2694
2695                         unlock_extent(extent_ins, start, end, GFP_NOFS);
2696                         BUG_ON(ret);
2697                         kfree(extent_op);
2698                 }
2699                 if (ret)
2700                         err = ret;
2701
2702                 search = end + 1;
2703
2704                 if (need_resched()) {
2705                         mutex_unlock(&info->extent_ins_mutex);
2706                         cond_resched();
2707                         mutex_lock(&info->extent_ins_mutex);
2708                 }
2709         }
2710
2711         if (nr) {
2712                 ret = free_extents(trans, extent_root, &delete_list);
2713                 BUG_ON(ret);
2714         }
2715
2716         if (all && skipped) {
2717                 INIT_LIST_HEAD(&delete_list);
2718                 search = 0;
2719                 nr = 0;
2720                 goto again;
2721         }
2722
2723         if (!err)
2724                 finish_current_insert(trans, extent_root, 0);
2725         return err;
2726 }
2727
2728 /*
2729  * remove an extent from the root, returns 0 on success
2730  */
2731 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2732                                struct btrfs_root *root,
2733                                u64 bytenr, u64 num_bytes, u64 parent,
2734                                u64 root_objectid, u64 ref_generation,
2735                                u64 owner_objectid, int pin)
2736 {
2737         struct btrfs_root *extent_root = root->fs_info->extent_root;
2738         int pending_ret;
2739         int ret;
2740
2741         WARN_ON(num_bytes < root->sectorsize);
2742         if (root == extent_root) {
2743                 struct pending_extent_op *extent_op = NULL;
2744
2745                 mutex_lock(&root->fs_info->extent_ins_mutex);
2746                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2747                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2748                         u64 priv;
2749                         ret = get_state_private(&root->fs_info->extent_ins,
2750                                                 bytenr, &priv);
2751                         BUG_ON(ret);
2752                         extent_op = (struct pending_extent_op *)
2753                                                 (unsigned long)priv;
2754
2755                         extent_op->del = 1;
2756                         if (extent_op->type == PENDING_EXTENT_INSERT) {
2757                                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2758                                 return 0;
2759                         }
2760                 }
2761
2762                 if (extent_op) {
2763                         ref_generation = extent_op->orig_generation;
2764                         parent = extent_op->orig_parent;
2765                 }
2766
2767                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2768                 BUG_ON(!extent_op);
2769
2770                 extent_op->type = PENDING_EXTENT_DELETE;
2771                 extent_op->bytenr = bytenr;
2772                 extent_op->num_bytes = num_bytes;
2773                 extent_op->parent = parent;
2774                 extent_op->orig_parent = parent;
2775                 extent_op->generation = ref_generation;
2776                 extent_op->orig_generation = ref_generation;
2777                 extent_op->level = (int)owner_objectid;
2778                 INIT_LIST_HEAD(&extent_op->list);
2779                 extent_op->del = 0;
2780
2781                 set_extent_bits(&root->fs_info->pending_del,
2782                                 bytenr, bytenr + num_bytes - 1,
2783                                 EXTENT_WRITEBACK, GFP_NOFS);
2784                 set_state_private(&root->fs_info->pending_del,
2785                                   bytenr, (unsigned long)extent_op);
2786                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2787                 return 0;
2788         }
2789         /* if metadata always pin */
2790         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2791                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2792                         mutex_lock(&root->fs_info->pinned_mutex);
2793                         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2794                         mutex_unlock(&root->fs_info->pinned_mutex);
2795                         update_reserved_extents(root, bytenr, num_bytes, 0);
2796                         return 0;
2797                 }
2798                 pin = 1;
2799         }
2800
2801         /* if data pin when any transaction has committed this */
2802         if (ref_generation != trans->transid)
2803                 pin = 1;
2804
2805         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2806                             root_objectid, ref_generation,
2807                             owner_objectid, pin, pin == 0);
2808
2809         finish_current_insert(trans, root->fs_info->extent_root, 0);
2810         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2811         return ret ? ret : pending_ret;
2812 }
2813
2814 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2815                       struct btrfs_root *root,
2816                       u64 bytenr, u64 num_bytes, u64 parent,
2817                       u64 root_objectid, u64 ref_generation,
2818                       u64 owner_objectid, int pin)
2819 {
2820         int ret;
2821
2822         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2823                                   root_objectid, ref_generation,
2824                                   owner_objectid, pin);
2825         return ret;
2826 }
2827
2828 static u64 stripe_align(struct btrfs_root *root, u64 val)
2829 {
2830         u64 mask = ((u64)root->stripesize - 1);
2831         u64 ret = (val + mask) & ~mask;
2832         return ret;
2833 }
2834
2835 /*
2836  * walks the btree of allocated extents and find a hole of a given size.
2837  * The key ins is changed to record the hole:
2838  * ins->objectid == block start
2839  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2840  * ins->offset == number of blocks
2841  * Any available blocks before search_start are skipped.
2842  */
2843 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2844                                      struct btrfs_root *orig_root,
2845                                      u64 num_bytes, u64 empty_size,
2846                                      u64 search_start, u64 search_end,
2847                                      u64 hint_byte, struct btrfs_key *ins,
2848                                      u64 exclude_start, u64 exclude_nr,
2849                                      int data)
2850 {
2851         int ret = 0;
2852         struct btrfs_root *root = orig_root->fs_info->extent_root;
2853         u64 total_needed = num_bytes;
2854         u64 *last_ptr = NULL;
2855         u64 last_wanted = 0;
2856         struct btrfs_block_group_cache *block_group = NULL;
2857         int chunk_alloc_done = 0;
2858         int empty_cluster = 2 * 1024 * 1024;
2859         int allowed_chunk_alloc = 0;
2860         struct list_head *head = NULL, *cur = NULL;
2861         int loop = 0;
2862         int extra_loop = 0;
2863         struct btrfs_space_info *space_info;
2864
2865         WARN_ON(num_bytes < root->sectorsize);
2866         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2867         ins->objectid = 0;
2868         ins->offset = 0;
2869
2870         if (orig_root->ref_cows || empty_size)
2871                 allowed_chunk_alloc = 1;
2872
2873         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2874                 last_ptr = &root->fs_info->last_alloc;
2875                 if (!btrfs_test_opt(root, SSD))
2876                         empty_cluster = 64 * 1024;
2877         }
2878
2879         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2880                 last_ptr = &root->fs_info->last_data_alloc;
2881
2882         if (last_ptr) {
2883                 if (*last_ptr) {
2884                         hint_byte = *last_ptr;
2885                         last_wanted = *last_ptr;
2886                 } else
2887                         empty_size += empty_cluster;
2888         } else {
2889                 empty_cluster = 0;
2890         }
2891         search_start = max(search_start, first_logical_byte(root, 0));
2892         search_start = max(search_start, hint_byte);
2893
2894         if (last_wanted && search_start != last_wanted) {
2895                 last_wanted = 0;
2896                 empty_size += empty_cluster;
2897         }
2898
2899         total_needed += empty_size;
2900         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2901         if (!block_group)
2902                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2903                                                              search_start);
2904         space_info = __find_space_info(root->fs_info, data);
2905
2906         down_read(&space_info->groups_sem);
2907         while (1) {
2908                 struct btrfs_free_space *free_space;
2909                 /*
2910                  * the only way this happens if our hint points to a block
2911                  * group thats not of the proper type, while looping this
2912                  * should never happen
2913                  */
2914                 if (empty_size)
2915                         extra_loop = 1;
2916
2917                 if (!block_group)
2918                         goto new_group_no_lock;
2919
2920                 if (unlikely(!block_group->cached)) {
2921                         mutex_lock(&block_group->cache_mutex);
2922                         ret = cache_block_group(root, block_group);
2923                         mutex_unlock(&block_group->cache_mutex);
2924                         if (ret)
2925                                 break;
2926                 }
2927
2928                 mutex_lock(&block_group->alloc_mutex);
2929                 if (unlikely(!block_group_bits(block_group, data)))
2930                         goto new_group;
2931
2932                 if (unlikely(block_group->ro))
2933                         goto new_group;
2934
2935                 free_space = btrfs_find_free_space(block_group, search_start,
2936                                                    total_needed);
2937                 if (free_space) {
2938                         u64 start = block_group->key.objectid;
2939                         u64 end = block_group->key.objectid +
2940                                 block_group->key.offset;
2941
2942                         search_start = stripe_align(root, free_space->offset);
2943
2944                         /* move on to the next group */
2945                         if (search_start + num_bytes >= search_end)
2946                                 goto new_group;
2947
2948                         /* move on to the next group */
2949                         if (search_start + num_bytes > end)
2950                                 goto new_group;
2951
2952                         if (last_wanted && search_start != last_wanted) {
2953                                 total_needed += empty_cluster;
2954                                 empty_size += empty_cluster;
2955                                 last_wanted = 0;
2956                                 /*
2957                                  * if search_start is still in this block group
2958                                  * then we just re-search this block group
2959                                  */
2960                                 if (search_start >= start &&
2961                                     search_start < end) {
2962                                         mutex_unlock(&block_group->alloc_mutex);
2963                                         continue;
2964                                 }
2965
2966                                 /* else we go to the next block group */
2967                                 goto new_group;
2968                         }
2969
2970                         if (exclude_nr > 0 &&
2971                             (search_start + num_bytes > exclude_start &&
2972                              search_start < exclude_start + exclude_nr)) {
2973                                 search_start = exclude_start + exclude_nr;
2974                                 /*
2975                                  * if search_start is still in this block group
2976                                  * then we just re-search this block group
2977                                  */
2978                                 if (search_start >= start &&
2979                                     search_start < end) {
2980                                         mutex_unlock(&block_group->alloc_mutex);
2981                                         last_wanted = 0;
2982                                         continue;
2983                                 }
2984
2985                                 /* else we go to the next block group */
2986                                 goto new_group;
2987                         }
2988
2989                         ins->objectid = search_start;
2990                         ins->offset = num_bytes;
2991
2992                         btrfs_remove_free_space_lock(block_group, search_start,
2993                                                      num_bytes);
2994                         /* we are all good, lets return */
2995                         mutex_unlock(&block_group->alloc_mutex);
2996                         break;
2997                 }
2998 new_group:
2999                 mutex_unlock(&block_group->alloc_mutex);
3000                 put_block_group(block_group);
3001                 block_group = NULL;
3002 new_group_no_lock:
3003                 /* don't try to compare new allocations against the
3004                  * last allocation any more
3005                  */
3006                 last_wanted = 0;
3007
3008                 /*
3009                  * Here's how this works.
3010                  * loop == 0: we were searching a block group via a hint
3011                  *              and didn't find anything, so we start at
3012                  *              the head of the block groups and keep searching
3013                  * loop == 1: we're searching through all of the block groups
3014                  *              if we hit the head again we have searched
3015                  *              all of the block groups for this space and we
3016                  *              need to try and allocate, if we cant error out.
3017                  * loop == 2: we allocated more space and are looping through
3018                  *              all of the block groups again.
3019                  */
3020                 if (loop == 0) {
3021                         head = &space_info->block_groups;
3022                         cur = head->next;
3023                         loop++;
3024                 } else if (loop == 1 && cur == head) {
3025                         int keep_going;
3026
3027                         /* at this point we give up on the empty_size
3028                          * allocations and just try to allocate the min
3029                          * space.
3030                          *
3031                          * The extra_loop field was set if an empty_size
3032                          * allocation was attempted above, and if this
3033                          * is try we need to try the loop again without
3034                          * the additional empty_size.
3035                          */
3036                         total_needed -= empty_size;
3037                         empty_size = 0;
3038                         keep_going = extra_loop;
3039                         loop++;
3040
3041                         if (allowed_chunk_alloc && !chunk_alloc_done) {
3042                                 up_read(&space_info->groups_sem);
3043                                 ret = do_chunk_alloc(trans, root, num_bytes +
3044                                                      2 * 1024 * 1024, data, 1);
3045                                 down_read(&space_info->groups_sem);
3046                                 if (ret < 0)
3047                                         goto loop_check;
3048                                 head = &space_info->block_groups;
3049                                 /*
3050                                  * we've allocated a new chunk, keep
3051                                  * trying
3052                                  */
3053                                 keep_going = 1;
3054                                 chunk_alloc_done = 1;
3055                         } else if (!allowed_chunk_alloc) {
3056                                 space_info->force_alloc = 1;
3057                         }
3058 loop_check:
3059                         if (keep_going) {
3060                                 cur = head->next;
3061                                 extra_loop = 0;
3062                         } else {
3063                                 break;
3064                         }
3065                 } else if (cur == head) {
3066                         break;
3067                 }
3068
3069                 block_group = list_entry(cur, struct btrfs_block_group_cache,
3070                                          list);
3071                 atomic_inc(&block_group->count);
3072
3073                 search_start = block_group->key.objectid;
3074                 cur = cur->next;
3075         }
3076
3077         /* we found what we needed */
3078         if (ins->objectid) {
3079                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3080                         trans->block_group = block_group->key.objectid;
3081
3082                 if (last_ptr)
3083                         *last_ptr = ins->objectid + ins->offset;
3084                 ret = 0;
3085         } else if (!ret) {
3086                 printk(KERN_ERR "btrfs searching for %llu bytes, "
3087                        "num_bytes %llu, loop %d, allowed_alloc %d\n",
3088                        (unsigned long long)total_needed,
3089                        (unsigned long long)num_bytes,
3090                        loop, allowed_chunk_alloc);
3091                 ret = -ENOSPC;
3092         }
3093         if (block_group)
3094                 put_block_group(block_group);
3095
3096         up_read(&space_info->groups_sem);
3097         return ret;
3098 }
3099
3100 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3101 {
3102         struct btrfs_block_group_cache *cache;
3103
3104         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3105                (unsigned long long)(info->total_bytes - info->bytes_used -
3106                                     info->bytes_pinned - info->bytes_reserved),
3107                (info->full) ? "" : "not ");
3108
3109         down_read(&info->groups_sem);
3110         list_for_each_entry(cache, &info->block_groups, list) {
3111                 spin_lock(&cache->lock);
3112                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3113                        "%llu pinned %llu reserved\n",
3114                        (unsigned long long)cache->key.objectid,
3115                        (unsigned long long)cache->key.offset,
3116                        (unsigned long long)btrfs_block_group_used(&cache->item),
3117                        (unsigned long long)cache->pinned,
3118                        (unsigned long long)cache->reserved);
3119                 btrfs_dump_free_space(cache, bytes);
3120                 spin_unlock(&cache->lock);
3121         }
3122         up_read(&info->groups_sem);
3123 }
3124
3125 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3126                                   struct btrfs_root *root,
3127                                   u64 num_bytes, u64 min_alloc_size,
3128                                   u64 empty_size, u64 hint_byte,
3129                                   u64 search_end, struct btrfs_key *ins,
3130                                   u64 data)
3131 {
3132         int ret;
3133         u64 search_start = 0;
3134         u64 alloc_profile;
3135         struct btrfs_fs_info *info = root->fs_info;
3136
3137         if (data) {
3138                 alloc_profile = info->avail_data_alloc_bits &
3139                         info->data_alloc_profile;
3140                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3141         } else if (root == root->fs_info->chunk_root) {
3142                 alloc_profile = info->avail_system_alloc_bits &
3143                         info->system_alloc_profile;
3144                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3145         } else {
3146                 alloc_profile = info->avail_metadata_alloc_bits &
3147                         info->metadata_alloc_profile;
3148                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3149         }
3150 again:
3151         data = btrfs_reduce_alloc_profile(root, data);
3152         /*
3153          * the only place that sets empty_size is btrfs_realloc_node, which
3154          * is not called recursively on allocations
3155          */
3156         if (empty_size || root->ref_cows) {
3157                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3158                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3159                                      2 * 1024 * 1024,
3160                                      BTRFS_BLOCK_GROUP_METADATA |
3161                                      (info->metadata_alloc_profile &
3162                                       info->avail_metadata_alloc_bits), 0);
3163                 }
3164                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3165                                      num_bytes + 2 * 1024 * 1024, data, 0);
3166         }
3167
3168         WARN_ON(num_bytes < root->sectorsize);
3169         ret = find_free_extent(trans, root, num_bytes, empty_size,
3170                                search_start, search_end, hint_byte, ins,
3171                                trans->alloc_exclude_start,
3172                                trans->alloc_exclude_nr, data);
3173
3174         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3175                 num_bytes = num_bytes >> 1;
3176                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3177                 num_bytes = max(num_bytes, min_alloc_size);
3178                 do_chunk_alloc(trans, root->fs_info->extent_root,
3179                                num_bytes, data, 1);
3180                 goto again;
3181         }
3182         if (ret) {
3183                 struct btrfs_space_info *sinfo;
3184
3185                 sinfo = __find_space_info(root->fs_info, data);
3186                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3187                        "wanted %llu\n", (unsigned long long)data,
3188                        (unsigned long long)num_bytes);
3189                 dump_space_info(sinfo, num_bytes);
3190                 BUG();
3191         }
3192
3193         return ret;
3194 }
3195
3196 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3197 {
3198         struct btrfs_block_group_cache *cache;
3199         int ret = 0;
3200
3201         cache = btrfs_lookup_block_group(root->fs_info, start);
3202         if (!cache) {
3203                 printk(KERN_ERR "Unable to find block group for %llu\n",
3204                        (unsigned long long)start);
3205                 return -ENOSPC;
3206         }
3207
3208         ret = btrfs_discard_extent(root, start, len);
3209
3210         btrfs_add_free_space(cache, start, len);
3211         put_block_group(cache);
3212         update_reserved_extents(root, start, len, 0);
3213
3214         return ret;
3215 }
3216
3217 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3218                                   struct btrfs_root *root,
3219                                   u64 num_bytes, u64 min_alloc_size,
3220                                   u64 empty_size, u64 hint_byte,
3221                                   u64 search_end, struct btrfs_key *ins,
3222                                   u64 data)
3223 {
3224         int ret;
3225         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3226                                      empty_size, hint_byte, search_end, ins,
3227                                      data);
3228         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3229         return ret;
3230 }
3231
3232 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3233                                          struct btrfs_root *root, u64 parent,
3234                                          u64 root_objectid, u64 ref_generation,
3235                                          u64 owner, struct btrfs_key *ins)
3236 {
3237         int ret;
3238         int pending_ret;
3239         u64 super_used;
3240         u64 root_used;
3241         u64 num_bytes = ins->offset;
3242         u32 sizes[2];
3243         struct btrfs_fs_info *info = root->fs_info;
3244         struct btrfs_root *extent_root = info->extent_root;
3245         struct btrfs_extent_item *extent_item;
3246         struct btrfs_extent_ref *ref;
3247         struct btrfs_path *path;
3248         struct btrfs_key keys[2];
3249
3250         if (parent == 0)
3251                 parent = ins->objectid;
3252
3253         /* block accounting for super block */
3254         spin_lock(&info->delalloc_lock);
3255         super_used = btrfs_super_bytes_used(&info->super_copy);
3256         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3257
3258         /* block accounting for root item */
3259         root_used = btrfs_root_used(&root->root_item);
3260         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3261         spin_unlock(&info->delalloc_lock);
3262
3263         if (root == extent_root) {
3264                 struct pending_extent_op *extent_op;
3265
3266                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3267                 BUG_ON(!extent_op);
3268
3269                 extent_op->type = PENDING_EXTENT_INSERT;
3270                 extent_op->bytenr = ins->objectid;
3271                 extent_op->num_bytes = ins->offset;
3272                 extent_op->parent = parent;
3273                 extent_op->orig_parent = 0;
3274                 extent_op->generation = ref_generation;
3275                 extent_op->orig_generation = 0;
3276                 extent_op->level = (int)owner;
3277                 INIT_LIST_HEAD(&extent_op->list);
3278                 extent_op->del = 0;
3279
3280                 mutex_lock(&root->fs_info->extent_ins_mutex);
3281                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3282                                 ins->objectid + ins->offset - 1,
3283                                 EXTENT_WRITEBACK, GFP_NOFS);
3284                 set_state_private(&root->fs_info->extent_ins,
3285                                   ins->objectid, (unsigned long)extent_op);
3286                 mutex_unlock(&root->fs_info->extent_ins_mutex);
3287                 goto update_block;
3288         }
3289
3290         memcpy(&keys[0], ins, sizeof(*ins));
3291         keys[1].objectid = ins->objectid;
3292         keys[1].type = BTRFS_EXTENT_REF_KEY;
3293         keys[1].offset = parent;
3294         sizes[0] = sizeof(*extent_item);
3295         sizes[1] = sizeof(*ref);
3296
3297         path = btrfs_alloc_path();
3298         BUG_ON(!path);
3299
3300         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3301                                        sizes, 2);
3302         BUG_ON(ret);
3303
3304         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3305                                      struct btrfs_extent_item);
3306         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3307         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3308                              struct btrfs_extent_ref);
3309
3310         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3311         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3312         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3313         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3314
3315         btrfs_mark_buffer_dirty(path->nodes[0]);
3316
3317         trans->alloc_exclude_start = 0;
3318         trans->alloc_exclude_nr = 0;
3319         btrfs_free_path(path);
3320         finish_current_insert(trans, extent_root, 0);
3321         pending_ret = del_pending_extents(trans, extent_root, 0);
3322
3323         if (ret)
3324                 goto out;
3325         if (pending_ret) {
3326                 ret = pending_ret;
3327                 goto out;
3328         }
3329
3330 update_block:
3331         ret = update_block_group(trans, root, ins->objectid,
3332                                  ins->offset, 1, 0);
3333         if (ret) {
3334                 printk(KERN_ERR "btrfs update block group failed for %llu "
3335                        "%llu\n", (unsigned long long)ins->objectid,
3336                        (unsigned long long)ins->offset);
3337                 BUG();
3338         }
3339 out:
3340         return ret;
3341 }
3342
3343 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3344                                 struct btrfs_root *root, u64 parent,
3345                                 u64 root_objectid, u64 ref_generation,
3346                                 u64 owner, struct btrfs_key *ins)
3347 {
3348         int ret;
3349
3350         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3351                 return 0;
3352         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3353                                             ref_generation, owner, ins);
3354         update_reserved_extents(root, ins->objectid, ins->offset, 0);
3355         return ret;
3356 }
3357
3358 /*
3359  * this is used by the tree logging recovery code.  It records that
3360  * an extent has been allocated and makes sure to clear the free
3361  * space cache bits as well
3362  */
3363 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3364                                 struct btrfs_root *root, u64 parent,
3365                                 u64 root_objectid, u64 ref_generation,
3366                                 u64 owner, struct btrfs_key *ins)
3367 {
3368         int ret;
3369         struct btrfs_block_group_cache *block_group;
3370
3371         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3372         mutex_lock(&block_group->cache_mutex);
3373         cache_block_group(root, block_group);
3374         mutex_unlock(&block_group->cache_mutex);
3375
3376         ret = btrfs_remove_free_space(block_group, ins->objectid,
3377                                       ins->offset);
3378         BUG_ON(ret);
3379         put_block_group(block_group);
3380         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3381                                             ref_generation, owner, ins);
3382         return ret;
3383 }
3384
3385 /*
3386  * finds a free extent and does all the dirty work required for allocation
3387  * returns the key for the extent through ins, and a tree buffer for
3388  * the first block of the extent through buf.
3389  *
3390  * returns 0 if everything worked, non-zero otherwise.
3391  */
3392 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3393                        struct btrfs_root *root,
3394                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3395                        u64 root_objectid, u64 ref_generation,
3396                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3397                        u64 search_end, struct btrfs_key *ins, u64 data)
3398 {
3399         int ret;
3400
3401         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3402                                      min_alloc_size, empty_size, hint_byte,
3403                                      search_end, ins, data);
3404         BUG_ON(ret);
3405         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3406                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3407                                         root_objectid, ref_generation,
3408                                         owner_objectid, ins);
3409                 BUG_ON(ret);
3410
3411         } else {
3412                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3413         }
3414         return ret;
3415 }
3416
3417 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3418                                             struct btrfs_root *root,
3419                                             u64 bytenr, u32 blocksize)
3420 {
3421         struct extent_buffer *buf;
3422
3423         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3424         if (!buf)
3425                 return ERR_PTR(-ENOMEM);
3426         btrfs_set_header_generation(buf, trans->transid);
3427         btrfs_tree_lock(buf);
3428         clean_tree_block(trans, root, buf);
3429
3430         btrfs_set_lock_blocking(buf);
3431         btrfs_set_buffer_uptodate(buf);
3432
3433         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3434                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3435                          buf->start + buf->len - 1, GFP_NOFS);
3436         } else {
3437                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3438                          buf->start + buf->len - 1, GFP_NOFS);
3439         }
3440         trans->blocks_used++;
3441         /* this returns a buffer locked for blocking */
3442         return buf;
3443 }
3444
3445 /*
3446  * helper function to allocate a block for a given tree
3447  * returns the tree buffer or NULL.
3448  */
3449 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3450                                              struct btrfs_root *root,
3451                                              u32 blocksize, u64 parent,
3452                                              u64 root_objectid,
3453                                              u64 ref_generation,
3454                                              int level,
3455                                              u64 hint,
3456                                              u64 empty_size)
3457 {
3458         struct btrfs_key ins;
3459         int ret;
3460         struct extent_buffer *buf;
3461
3462         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3463                                  root_objectid, ref_generation, level,
3464                                  empty_size, hint, (u64)-1, &ins, 0);
3465         if (ret) {
3466                 BUG_ON(ret > 0);
3467                 return ERR_PTR(ret);
3468         }
3469
3470         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3471         return buf;
3472 }
3473
3474 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3475                         struct btrfs_root *root, struct extent_buffer *leaf)
3476 {
3477         u64 leaf_owner;
3478         u64 leaf_generation;
3479         struct refsort *sorted;
3480         struct btrfs_key key;
3481         struct btrfs_file_extent_item *fi;
3482         int i;
3483         int nritems;
3484         int ret;
3485         int refi = 0;
3486         int slot;
3487
3488         BUG_ON(!btrfs_is_leaf(leaf));
3489         nritems = btrfs_header_nritems(leaf);
3490         leaf_owner = btrfs_header_owner(leaf);
3491         leaf_generation = btrfs_header_generation(leaf);
3492
3493         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3494         /* we do this loop twice.  The first time we build a list
3495          * of the extents we have a reference on, then we sort the list
3496          * by bytenr.  The second time around we actually do the
3497          * extent freeing.
3498          */
3499         for (i = 0; i < nritems; i++) {
3500                 u64 disk_bytenr;
3501                 cond_resched();
3502
3503                 btrfs_item_key_to_cpu(leaf, &key, i);
3504
3505                 /* only extents have references, skip everything else */
3506                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3507                         continue;
3508
3509                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3510
3511                 /* inline extents live in the btree, they don't have refs */
3512                 if (btrfs_file_extent_type(leaf, fi) ==
3513                     BTRFS_FILE_EXTENT_INLINE)
3514                         continue;
3515
3516                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3517
3518                 /* holes don't have refs */
3519                 if (disk_bytenr == 0)
3520                         continue;
3521
3522                 sorted[refi].bytenr = disk_bytenr;
3523                 sorted[refi].slot = i;
3524                 refi++;
3525         }
3526
3527         if (refi == 0)
3528                 goto out;
3529
3530         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3531
3532         for (i = 0; i < refi; i++) {
3533                 u64 disk_bytenr;
3534
3535                 disk_bytenr = sorted[i].bytenr;
3536                 slot = sorted[i].slot;
3537
3538                 cond_resched();
3539
3540                 btrfs_item_key_to_cpu(leaf, &key, slot);
3541                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3542                         continue;
3543
3544                 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3545
3546                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3547                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3548                                 leaf->start, leaf_owner, leaf_generation,
3549                                 key.objectid, 0);
3550                 BUG_ON(ret);
3551
3552                 atomic_inc(&root->fs_info->throttle_gen);
3553                 wake_up(&root->fs_info->transaction_throttle);
3554                 cond_resched();
3555         }
3556 out:
3557         kfree(sorted);
3558         return 0;
3559 }
3560
3561 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3562                                         struct btrfs_root *root,
3563                                         struct btrfs_leaf_ref *ref)
3564 {
3565         int i;
3566         int ret;
3567         struct btrfs_extent_info *info;
3568         struct refsort *sorted;
3569
3570         if (ref->nritems == 0)
3571                 return 0;
3572
3573         sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
3574         for (i = 0; i < ref->nritems; i++) {
3575                 sorted[i].bytenr = ref->extents[i].bytenr;
3576                 sorted[i].slot = i;
3577         }
3578         sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
3579
3580         /*
3581          * the items in the ref were sorted when the ref was inserted
3582          * into the ref cache, so this is already in order
3583          */
3584         for (i = 0; i < ref->nritems; i++) {
3585                 info = ref->extents + sorted[i].slot;
3586                 ret = __btrfs_free_extent(trans, root, info->bytenr,
3587                                           info->num_bytes, ref->bytenr,
3588                                           ref->owner, ref->generation,
3589                                           info->objectid, 0);
3590
3591                 atomic_inc(&root->fs_info->throttle_gen);
3592                 wake_up(&root->fs_info->transaction_throttle);
3593                 cond_resched();
3594
3595                 BUG_ON(ret);
3596                 info++;
3597         }
3598
3599         kfree(sorted);
3600         return 0;
3601 }
3602
3603 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3604                                      u64 len, u32 *refs)
3605 {
3606         int ret;
3607
3608         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3609         BUG_ON(ret);
3610
3611 #if 0 /* some debugging code in case we see problems here */
3612         /* if the refs count is one, it won't get increased again.  But
3613          * if the ref count is > 1, someone may be decreasing it at
3614          * the same time we are.
3615          */
3616         if (*refs != 1) {
3617                 struct extent_buffer *eb = NULL;
3618                 eb = btrfs_find_create_tree_block(root, start, len);
3619                 if (eb)
3620                         btrfs_tree_lock(eb);
3621
3622                 mutex_lock(&root->fs_info->alloc_mutex);
3623                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3624                 BUG_ON(ret);
3625                 mutex_unlock(&root->fs_info->alloc_mutex);
3626
3627                 if (eb) {
3628                         btrfs_tree_unlock(eb);
3629                         free_extent_buffer(eb);
3630                 }
3631                 if (*refs == 1) {
3632                         printk(KERN_ERR "btrfs block %llu went down to one "
3633                                "during drop_snap\n", (unsigned long long)start);
3634                 }
3635
3636         }
3637 #endif
3638
3639         cond_resched();
3640         return ret;
3641 }
3642
3643 /*
3644  * this is used while deleting old snapshots, and it drops the refs
3645  * on a whole subtree starting from a level 1 node.
3646  *
3647  * The idea is to sort all the leaf pointers, and then drop the
3648  * ref on all the leaves in order.  Most of the time the leaves
3649  * will have ref cache entries, so no leaf IOs will be required to
3650  * find the extents they have references on.
3651  *
3652  * For each leaf, any references it has are also dropped in order
3653  *
3654  * This ends up dropping the references in something close to optimal
3655  * order for reading and modifying the extent allocation tree.
3656  */
3657 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3658                                         struct btrfs_root *root,
3659                                         struct btrfs_path *path)
3660 {
3661         u64 bytenr;
3662         u64 root_owner;
3663         u64 root_gen;
3664         struct extent_buffer *eb = path->nodes[1];
3665         struct extent_buffer *leaf;
3666         struct btrfs_leaf_ref *ref;
3667         struct refsort *sorted = NULL;
3668         int nritems = btrfs_header_nritems(eb);
3669         int ret;
3670         int i;
3671         int refi = 0;
3672         int slot = path->slots[1];
3673         u32 blocksize = btrfs_level_size(root, 0);
3674         u32 refs;
3675
3676         if (nritems == 0)
3677                 goto out;
3678
3679         root_owner = btrfs_header_owner(eb);
3680         root_gen = btrfs_header_generation(eb);
3681         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3682
3683         /*
3684          * step one, sort all the leaf pointers so we don't scribble
3685          * randomly into the extent allocation tree
3686          */
3687         for (i = slot; i < nritems; i++) {
3688                 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
3689                 sorted[refi].slot = i;
3690                 refi++;
3691         }
3692
3693         /*
3694          * nritems won't be zero, but if we're picking up drop_snapshot
3695          * after a crash, slot might be > 0, so double check things
3696          * just in case.
3697          */
3698         if (refi == 0)
3699                 goto out;
3700
3701         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3702
3703         /*
3704          * the first loop frees everything the leaves point to
3705          */
3706         for (i = 0; i < refi; i++) {
3707                 u64 ptr_gen;
3708
3709                 bytenr = sorted[i].bytenr;
3710
3711                 /*
3712                  * check the reference count on this leaf.  If it is > 1
3713                  * we just decrement it below and don't update any
3714                  * of the refs the leaf points to.
3715                  */
3716                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3717                 BUG_ON(ret);
3718                 if (refs != 1)
3719                         continue;
3720
3721                 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
3722
3723                 /*
3724                  * the leaf only had one reference, which means the
3725                  * only thing pointing to this leaf is the snapshot
3726                  * we're deleting.  It isn't possible for the reference
3727                  * count to increase again later
3728                  *
3729                  * The reference cache is checked for the leaf,
3730                  * and if found we'll be able to drop any refs held by
3731                  * the leaf without needing to read it in.
3732                  */
3733                 ref = btrfs_lookup_leaf_ref(root, bytenr);
3734                 if (ref && ref->generation != ptr_gen) {
3735                         btrfs_free_leaf_ref(root, ref);
3736                         ref = NULL;
3737                 }
3738                 if (ref) {
3739                         ret = cache_drop_leaf_ref(trans, root, ref);
3740                         BUG_ON(ret);
3741                         btrfs_remove_leaf_ref(root, ref);
3742                         btrfs_free_leaf_ref(root, ref);
3743                 } else {
3744                         /*
3745                          * the leaf wasn't in the reference cache, so
3746                          * we have to read it.
3747                          */
3748                         leaf = read_tree_block(root, bytenr, blocksize,
3749                                                ptr_gen);
3750                         ret = btrfs_drop_leaf_ref(trans, root, leaf);
3751                         BUG_ON(ret);
3752                         free_extent_buffer(leaf);
3753                 }
3754                 atomic_inc(&root->fs_info->throttle_gen);
3755                 wake_up(&root->fs_info->transaction_throttle);
3756                 cond_resched();
3757         }
3758
3759         /*
3760          * run through the loop again to free the refs on the leaves.
3761          * This is faster than doing it in the loop above because
3762          * the leaves are likely to be clustered together.  We end up
3763          * working in nice chunks on the extent allocation tree.
3764          */
3765         for (i = 0; i < refi; i++) {
3766                 bytenr = sorted[i].bytenr;
3767                 ret = __btrfs_free_extent(trans, root, bytenr,
3768                                         blocksize, eb->start,
3769                                         root_owner, root_gen, 0, 1);
3770                 BUG_ON(ret);
3771
3772                 atomic_inc(&root->fs_info->throttle_gen);
3773                 wake_up(&root->fs_info->transaction_throttle);
3774                 cond_resched();
3775         }
3776 out:
3777         kfree(sorted);
3778
3779         /*
3780          * update the path to show we've processed the entire level 1
3781          * node.  This will get saved into the root's drop_snapshot_progress
3782          * field so these drops are not repeated again if this transaction
3783          * commits.
3784          */
3785         path->slots[1] = nritems;
3786         return 0;
3787 }
3788
3789 /*
3790  * helper function for drop_snapshot, this walks down the tree dropping ref
3791  * counts as it goes.
3792  */
3793 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3794                                    struct btrfs_root *root,
3795                                    struct btrfs_path *path, int *level)
3796 {
3797         u64 root_owner;
3798         u64 root_gen;
3799         u64 bytenr;
3800         u64 ptr_gen;
3801         struct extent_buffer *next;
3802         struct extent_buffer *cur;
3803         struct extent_buffer *parent;
3804         u32 blocksize;
3805         int ret;
3806         u32 refs;
3807
3808         WARN_ON(*level < 0);
3809         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3810         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3811                                 path->nodes[*level]->len, &refs);
3812         BUG_ON(ret);
3813         if (refs > 1)
3814                 goto out;
3815
3816         /*
3817          * walk down to the last node level and free all the leaves
3818          */
3819         while (*level >= 0) {
3820                 WARN_ON(*level < 0);
3821                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3822                 cur = path->nodes[*level];
3823
3824                 if (btrfs_header_level(cur) != *level)
3825                         WARN_ON(1);
3826
3827                 if (path->slots[*level] >=
3828                     btrfs_header_nritems(cur))
3829                         break;
3830
3831                 /* the new code goes down to level 1 and does all the
3832                  * leaves pointed to that node in bulk.  So, this check
3833                  * for level 0 will always be false.
3834                  *
3835                  * But, the disk format allows the drop_snapshot_progress
3836                  * field in the root to leave things in a state where
3837                  * a leaf will need cleaning up here.  If someone crashes
3838                  * with the old code and then boots with the new code,
3839                  * we might find a leaf here.
3840                  */
3841                 if (*level == 0) {
3842                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3843                         BUG_ON(ret);
3844                         break;
3845                 }
3846
3847                 /*
3848                  * once we get to level one, process the whole node
3849                  * at once, including everything below it.
3850                  */
3851                 if (*level == 1) {
3852                         ret = drop_level_one_refs(trans, root, path);
3853                         BUG_ON(ret);
3854                         break;
3855                 }
3856
3857                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3858                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3859                 blocksize = btrfs_level_size(root, *level - 1);
3860
3861                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3862                 BUG_ON(ret);
3863
3864                 /*
3865                  * if there is more than one reference, we don't need
3866                  * to read that node to drop any references it has.  We
3867                  * just drop the ref we hold on that node and move on to the
3868                  * next slot in this level.
3869                  */
3870                 if (refs != 1) {
3871                         parent = path->nodes[*level];
3872                         root_owner = btrfs_header_owner(parent);
3873                         root_gen = btrfs_header_generation(parent);
3874                         path->slots[*level]++;
3875
3876                         ret = __btrfs_free_extent(trans, root, bytenr,
3877                                                 blocksize, parent->start,
3878                                                 root_owner, root_gen,
3879                                                 *level - 1, 1);
3880                         BUG_ON(ret);
3881
3882                         atomic_inc(&root->fs_info->throttle_gen);
3883                         wake_up(&root->fs_info->transaction_throttle);
3884                         cond_resched();
3885
3886                         continue;
3887                 }
3888
3889                 /*
3890                  * we need to keep freeing things in the next level down.
3891                  * read the block and loop around to process it
3892                  */
3893                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3894                 WARN_ON(*level <= 0);
3895                 if (path->nodes[*level-1])
3896                         free_extent_buffer(path->nodes[*level-1]);
3897                 path->nodes[*level-1] = next;
3898                 *level = btrfs_header_level(next);
3899                 path->slots[*level] = 0;
3900                 cond_resched();
3901         }
3902 out:
3903         WARN_ON(*level < 0);
3904         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3905
3906         if (path->nodes[*level] == root->node) {
3907                 parent = path->nodes[*level];
3908                 bytenr = path->nodes[*level]->start;
3909         } else {
3910                 parent = path->nodes[*level + 1];
3911                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3912         }
3913
3914         blocksize = btrfs_level_size(root, *level);
3915         root_owner = btrfs_header_owner(parent);
3916         root_gen = btrfs_header_generation(parent);
3917
3918         /*
3919          * cleanup and free the reference on the last node
3920          * we processed
3921          */
3922         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3923                                   parent->start, root_owner, root_gen,
3924                                   *level, 1);
3925         free_extent_buffer(path->nodes[*level]);
3926         path->nodes[*level] = NULL;
3927
3928         *level += 1;
3929         BUG_ON(ret);
3930
3931         cond_resched();
3932         return 0;
3933 }
3934
3935 /*
3936  * helper function for drop_subtree, this function is similar to
3937  * walk_down_tree. The main difference is that it checks reference
3938  * counts while tree blocks are locked.
3939  */
3940 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3941                                       struct btrfs_root *root,
3942                                       struct btrfs_path *path, int *level)
3943 {
3944         struct extent_buffer *next;
3945         struct extent_buffer *cur;
3946         struct extent_buffer *parent;
3947         u64 bytenr;
3948         u64 ptr_gen;
3949         u32 blocksize;
3950         u32 refs;
3951         int ret;
3952
3953         cur = path->nodes[*level];
3954         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3955                                       &refs);
3956         BUG_ON(ret);
3957         if (refs > 1)
3958                 goto out;
3959
3960         while (*level >= 0) {
3961                 cur = path->nodes[*level];
3962                 if (*level == 0) {
3963                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3964                         BUG_ON(ret);
3965                         clean_tree_block(trans, root, cur);
3966                         break;
3967                 }
3968                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3969                         clean_tree_block(trans, root, cur);
3970                         break;
3971                 }
3972
3973                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3974                 blocksize = btrfs_level_size(root, *level - 1);
3975                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3976
3977                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3978                 btrfs_tree_lock(next);
3979                 btrfs_set_lock_blocking(next);
3980
3981                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3982                                               &refs);
3983                 BUG_ON(ret);
3984                 if (refs > 1) {
3985                         parent = path->nodes[*level];
3986                         ret = btrfs_free_extent(trans, root, bytenr,
3987                                         blocksize, parent->start,
3988                                         btrfs_header_owner(parent),
3989                                         btrfs_header_generation(parent),
3990                                         *level - 1, 1);
3991                         BUG_ON(ret);
3992                         path->slots[*level]++;
3993                         btrfs_tree_unlock(next);
3994                         free_extent_buffer(next);
3995                         continue;
3996                 }
3997
3998                 *level = btrfs_header_level(next);
3999                 path->nodes[*level] = next;
4000                 path->slots[*level] = 0;
4001                 path->locks[*level] = 1;
4002                 cond_resched();
4003         }
4004 out:
4005         parent = path->nodes[*level + 1];
4006         bytenr = path->nodes[*level]->start;
4007         blocksize = path->nodes[*level]->len;
4008
4009         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4010                         parent->start, btrfs_header_owner(parent),
4011                         btrfs_header_generation(parent), *level, 1);
4012         BUG_ON(ret);
4013
4014         if (path->locks[*level]) {
4015                 btrfs_tree_unlock(path->nodes[*level]);
4016                 path->locks[*level] = 0;
4017         }
4018         free_extent_buffer(path->nodes[*level]);
4019         path->nodes[*level] = NULL;
4020         *level += 1;
4021         cond_resched();
4022         return 0;
4023 }
4024
4025 /*
4026  * helper for dropping snapshots.  This walks back up the tree in the path
4027  * to find the first node higher up where we haven't yet gone through
4028  * all the slots
4029  */
4030 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
4031                                  struct btrfs_root *root,
4032                                  struct btrfs_path *path,
4033                                  int *level, int max_level)
4034 {
4035         u64 root_owner;
4036         u64 root_gen;
4037         struct btrfs_root_item *root_item = &root->root_item;
4038         int i;
4039         int slot;
4040         int ret;
4041
4042         for (i = *level; i < max_level && path->nodes[i]; i++) {
4043                 slot = path->slots[i];
4044                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
4045                         struct extent_buffer *node;
4046                         struct btrfs_disk_key disk_key;
4047
4048                         /*
4049                          * there is more work to do in this level.
4050                          * Update the drop_progress marker to reflect
4051                          * the work we've done so far, and then bump
4052                          * the slot number
4053                          */
4054                         node = path->nodes[i];
4055                         path->slots[i]++;
4056                         *level = i;
4057                         WARN_ON(*level == 0);
4058                         btrfs_node_key(node, &disk_key, path->slots[i]);
4059                         memcpy(&root_item->drop_progress,
4060                                &disk_key, sizeof(disk_key));
4061                         root_item->drop_level = i;
4062                         return 0;
4063                 } else {
4064                         struct extent_buffer *parent;
4065
4066                         /*
4067                          * this whole node is done, free our reference
4068                          * on it and go up one level
4069                          */
4070                         if (path->nodes[*level] == root->node)
4071                                 parent = path->nodes[*level];
4072                         else
4073                                 parent = path->nodes[*level + 1];
4074
4075                         root_owner = btrfs_header_owner(parent);
4076                         root_gen = btrfs_header_generation(parent);
4077
4078                         clean_tree_block(trans, root, path->nodes[*level]);
4079                         ret = btrfs_free_extent(trans, root,
4080                                                 path->nodes[*level]->start,
4081                                                 path->nodes[*level]->len,
4082                                                 parent->start, root_owner,
4083                                                 root_gen, *level, 1);
4084                         BUG_ON(ret);
4085                         if (path->locks[*level]) {
4086                                 btrfs_tree_unlock(path->nodes[*level]);
4087                                 path->locks[*level] = 0;
4088                         }
4089                         free_extent_buffer(path->nodes[*level]);
4090                         path->nodes[*level] = NULL;
4091                         *level = i + 1;
4092                 }
4093         }
4094         return 1;
4095 }
4096
4097 /*
4098  * drop the reference count on the tree rooted at 'snap'.  This traverses
4099  * the tree freeing any blocks that have a ref count of zero after being
4100  * decremented.
4101  */
4102 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
4103                         *root)
4104 {
4105         int ret = 0;
4106         int wret;
4107         int level;
4108         struct btrfs_path *path;
4109         int i;
4110         int orig_level;
4111         struct btrfs_root_item *root_item = &root->root_item;
4112
4113         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
4114         path = btrfs_alloc_path();
4115         BUG_ON(!path);
4116
4117         level = btrfs_header_level(root->node);
4118         orig_level = level;
4119         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
4120                 path->nodes[level] = root->node;
4121                 extent_buffer_get(root->node);
4122                 path->slots[level] = 0;
4123         } else {
4124                 struct btrfs_key key;
4125                 struct btrfs_disk_key found_key;
4126                 struct extent_buffer *node;
4127
4128                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
4129                 level = root_item->drop_level;
4130                 path->lowest_level = level;
4131                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4132                 if (wret < 0) {
4133                         ret = wret;
4134                         goto out;
4135                 }
4136                 node = path->nodes[level];
4137                 btrfs_node_key(node, &found_key, path->slots[level]);
4138                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
4139                                sizeof(found_key)));
4140                 /*
4141                  * unlock our path, this is safe because only this
4142                  * function is allowed to delete this snapshot
4143                  */
4144                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4145                         if (path->nodes[i] && path->locks[i]) {
4146                                 path->locks[i] = 0;
4147                                 btrfs_tree_unlock(path->nodes[i]);
4148                         }
4149                 }
4150         }
4151         while (1) {
4152                 wret = walk_down_tree(trans, root, path, &level);
4153                 if (wret > 0)
4154                         break;
4155                 if (wret < 0)
4156                         ret = wret;
4157
4158                 wret = walk_up_tree(trans, root, path, &level,
4159                                     BTRFS_MAX_LEVEL);
4160                 if (wret > 0)
4161                         break;
4162                 if (wret < 0)
4163                         ret = wret;
4164                 if (trans->transaction->in_commit) {
4165                         ret = -EAGAIN;
4166                         break;
4167                 }
4168                 atomic_inc(&root->fs_info->throttle_gen);
4169                 wake_up(&root->fs_info->transaction_throttle);
4170         }
4171         for (i = 0; i <= orig_level; i++) {
4172                 if (path->nodes[i]) {
4173                         free_extent_buffer(path->nodes[i]);
4174                         path->nodes[i] = NULL;
4175                 }
4176         }
4177 out:
4178         btrfs_free_path(path);
4179         return ret;
4180 }
4181
4182 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
4183                         struct btrfs_root *root,
4184                         struct extent_buffer *node,
4185                         struct extent_buffer *parent)
4186 {
4187         struct btrfs_path *path;
4188         int level;
4189         int parent_level;
4190         int ret = 0;
4191         int wret;
4192
4193         path = btrfs_alloc_path();
4194         BUG_ON(!path);
4195
4196         BUG_ON(!btrfs_tree_locked(parent));
4197         parent_level = btrfs_header_level(parent);
4198         extent_buffer_get(parent);
4199         path->nodes[parent_level] = parent;
4200         path->slots[parent_level] = btrfs_header_nritems(parent);
4201
4202         BUG_ON(!btrfs_tree_locked(node));
4203         level = btrfs_header_level(node);
4204         extent_buffer_get(node);
4205         path->nodes[level] = node;
4206         path->slots[level] = 0;
4207
4208         while (1) {
4209                 wret = walk_down_subtree(trans, root, path, &level);
4210                 if (wret < 0)
4211                         ret = wret;
4212                 if (wret != 0)
4213                         break;
4214
4215                 wret = walk_up_tree(trans, root, path, &level, parent_level);
4216                 if (wret < 0)
4217                         ret = wret;
4218                 if (wret != 0)
4219                         break;
4220         }
4221
4222         btrfs_free_path(path);
4223         return ret;
4224 }
4225
4226 static unsigned long calc_ra(unsigned long start, unsigned long last,
4227                              unsigned long nr)
4228 {
4229         return min(last, start + nr - 1);
4230 }
4231
4232 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
4233                                          u64 len)
4234 {
4235         u64 page_start;
4236         u64 page_end;
4237         unsigned long first_index;
4238         unsigned long last_index;
4239         unsigned long i;
4240         struct page *page;
4241         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4242         struct file_ra_state *ra;
4243         struct btrfs_ordered_extent *ordered;
4244         unsigned int total_read = 0;
4245         unsigned int total_dirty = 0;
4246         int ret = 0;
4247
4248         ra = kzalloc(sizeof(*ra), GFP_NOFS);
4249
4250         mutex_lock(&inode->i_mutex);
4251         first_index = start >> PAGE_CACHE_SHIFT;
4252         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
4253
4254         /* make sure the dirty trick played by the caller work */
4255         ret = invalidate_inode_pages2_range(inode->i_mapping,
4256                                             first_index, last_index);
4257         if (ret)
4258                 goto out_unlock;
4259
4260         file_ra_state_init(ra, inode->i_mapping);
4261
4262         for (i = first_index ; i <= last_index; i++) {
4263                 if (total_read % ra->ra_pages == 0) {
4264                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
4265                                        calc_ra(i, last_index, ra->ra_pages));
4266                 }
4267                 total_read++;
4268 again:
4269                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
4270                         BUG_ON(1);
4271                 page = grab_cache_page(inode->i_mapping, i);
4272                 if (!page) {
4273                         ret = -ENOMEM;
4274                         goto out_unlock;
4275                 }
4276                 if (!PageUptodate(page)) {
4277                         btrfs_readpage(NULL, page);
4278                         lock_page(page);
4279                         if (!PageUptodate(page)) {
4280                                 unlock_page(page);
4281                                 page_cache_release(page);
4282                                 ret = -EIO;
4283                                 goto out_unlock;
4284                         }
4285                 }
4286                 wait_on_page_writeback(page);
4287
4288                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
4289                 page_end = page_start + PAGE_CACHE_SIZE - 1;
4290                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4291
4292                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4293                 if (ordered) {
4294                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4295                         unlock_page(page);
4296                         page_cache_release(page);
4297                         btrfs_start_ordered_extent(inode, ordered, 1);
4298                         btrfs_put_ordered_extent(ordered);
4299                         goto again;
4300                 }
4301                 set_page_extent_mapped(page);
4302
4303                 if (i == first_index)
4304                         set_extent_bits(io_tree, page_start, page_end,
4305                                         EXTENT_BOUNDARY, GFP_NOFS);
4306                 btrfs_set_extent_delalloc(inode, page_start, page_end);
4307
4308                 set_page_dirty(page);
4309                 total_dirty++;
4310
4311                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4312                 unlock_page(page);
4313                 page_cache_release(page);
4314         }
4315
4316 out_unlock:
4317         kfree(ra);
4318         mutex_unlock(&inode->i_mutex);
4319         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4320         return ret;
4321 }
4322
4323 static noinline int relocate_data_extent(struct inode *reloc_inode,
4324                                          struct btrfs_key *extent_key,
4325                                          u64 offset)
4326 {
4327         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4328         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4329         struct extent_map *em;
4330         u64 start = extent_key->objectid - offset;
4331         u64 end = start + extent_key->offset - 1;
4332
4333         em = alloc_extent_map(GFP_NOFS);
4334         BUG_ON(!em || IS_ERR(em));
4335
4336         em->start = start;
4337         em->len = extent_key->offset;
4338         em->block_len = extent_key->offset;
4339         em->block_start = extent_key->objectid;
4340         em->bdev = root->fs_info->fs_devices->latest_bdev;
4341         set_bit(EXTENT_FLAG_PINNED, &em->flags);
4342
4343         /* setup extent map to cheat btrfs_readpage */
4344         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4345         while (1) {
4346                 int ret;
4347                 spin_lock(&em_tree->lock);
4348                 ret = add_extent_mapping(em_tree, em);
4349                 spin_unlock(&em_tree->lock);
4350                 if (ret != -EEXIST) {
4351                         free_extent_map(em);
4352                         break;
4353                 }
4354                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4355         }
4356         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4357
4358         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4359 }
4360
4361 struct btrfs_ref_path {
4362         u64 extent_start;
4363         u64 nodes[BTRFS_MAX_LEVEL];
4364         u64 root_objectid;
4365         u64 root_generation;
4366         u64 owner_objectid;
4367         u32 num_refs;
4368         int lowest_level;
4369         int current_level;
4370         int shared_level;
4371
4372         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4373         u64 new_nodes[BTRFS_MAX_LEVEL];
4374 };
4375
4376 struct disk_extent {
4377         u64 ram_bytes;
4378         u64 disk_bytenr;
4379         u64 disk_num_bytes;
4380         u64 offset;
4381         u64 num_bytes;
4382         u8 compression;
4383         u8 encryption;
4384         u16 other_encoding;
4385 };
4386
4387 static int is_cowonly_root(u64 root_objectid)
4388 {
4389         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4390             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4391             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4392             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4393             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4394             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4395                 return 1;
4396         return 0;
4397 }
4398
4399 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4400                                     struct btrfs_root *extent_root,
4401                                     struct btrfs_ref_path *ref_path,
4402                                     int first_time)
4403 {
4404         struct extent_buffer *leaf;
4405         struct btrfs_path *path;
4406         struct btrfs_extent_ref *ref;
4407         struct btrfs_key key;
4408         struct btrfs_key found_key;
4409         u64 bytenr;
4410         u32 nritems;
4411         int level;
4412         int ret = 1;
4413
4414         path = btrfs_alloc_path();
4415         if (!path)
4416                 return -ENOMEM;
4417
4418         if (first_time) {
4419                 ref_path->lowest_level = -1;
4420                 ref_path->current_level = -1;
4421                 ref_path->shared_level = -1;
4422                 goto walk_up;
4423         }
4424 walk_down:
4425         level = ref_path->current_level - 1;
4426         while (level >= -1) {
4427                 u64 parent;
4428                 if (level < ref_path->lowest_level)
4429                         break;
4430
4431                 if (level >= 0)
4432                         bytenr = ref_path->nodes[level];
4433                 else
4434                         bytenr = ref_path->extent_start;
4435                 BUG_ON(bytenr == 0);
4436
4437                 parent = ref_path->nodes[level + 1];
4438                 ref_path->nodes[level + 1] = 0;
4439                 ref_path->current_level = level;
4440                 BUG_ON(parent == 0);
4441
4442                 key.objectid = bytenr;
4443                 key.offset = parent + 1;
4444                 key.type = BTRFS_EXTENT_REF_KEY;
4445
4446                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4447                 if (ret < 0)
4448                         goto out;
4449                 BUG_ON(ret == 0);
4450
4451                 leaf = path->nodes[0];
4452                 nritems = btrfs_header_nritems(leaf);
4453                 if (path->slots[0] >= nritems) {
4454                         ret = btrfs_next_leaf(extent_root, path);
4455                         if (ret < 0)
4456                                 goto out;
4457                         if (ret > 0)
4458                                 goto next;
4459                         leaf = path->nodes[0];
4460                 }
4461
4462                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4463                 if (found_key.objectid == bytenr &&
4464                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4465                         if (level < ref_path->shared_level)
4466                                 ref_path->shared_level = level;
4467                         goto found;
4468                 }
4469 next:
4470                 level--;
4471                 btrfs_release_path(extent_root, path);
4472                 cond_resched();
4473         }
4474         /* reached lowest level */
4475         ret = 1;
4476         goto out;
4477 walk_up:
4478         level = ref_path->current_level;
4479         while (level < BTRFS_MAX_LEVEL - 1) {
4480                 u64 ref_objectid;
4481
4482                 if (level >= 0)
4483                         bytenr = ref_path->nodes[level];
4484                 else
4485                         bytenr = ref_path->extent_start;
4486
4487                 BUG_ON(bytenr == 0);
4488
4489                 key.objectid = bytenr;
4490                 key.offset = 0;
4491                 key.type = BTRFS_EXTENT_REF_KEY;
4492
4493                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4494                 if (ret < 0)
4495                         goto out;
4496
4497                 leaf = path->nodes[0];
4498                 nritems = btrfs_header_nritems(leaf);
4499                 if (path->slots[0] >= nritems) {
4500                         ret = btrfs_next_leaf(extent_root, path);
4501                         if (ret < 0)
4502                                 goto out;
4503                         if (ret > 0) {
4504                                 /* the extent was freed by someone */
4505                                 if (ref_path->lowest_level == level)
4506                                         goto out;
4507                                 btrfs_release_path(extent_root, path);
4508                                 goto walk_down;
4509                         }
4510                         leaf = path->nodes[0];
4511                 }
4512
4513                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4514                 if (found_key.objectid != bytenr ||
4515                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4516                         /* the extent was freed by someone */
4517                         if (ref_path->lowest_level == level) {
4518                                 ret = 1;
4519                                 goto out;
4520                         }
4521                         btrfs_release_path(extent_root, path);
4522                         goto walk_down;
4523                 }
4524 found:
4525                 ref = btrfs_item_ptr(leaf, path->slots[0],
4526                                 struct btrfs_extent_ref);
4527                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4528                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4529                         if (first_time) {
4530                                 level = (int)ref_objectid;
4531                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4532                                 ref_path->lowest_level = level;
4533                                 ref_path->current_level = level;
4534                                 ref_path->nodes[level] = bytenr;
4535                         } else {
4536                                 WARN_ON(ref_objectid != level);
4537                         }
4538                 } else {
4539                         WARN_ON(level != -1);
4540                 }
4541                 first_time = 0;
4542
4543                 if (ref_path->lowest_level == level) {
4544                         ref_path->owner_objectid = ref_objectid;
4545                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4546                 }
4547
4548                 /*
4549                  * the block is tree root or the block isn't in reference
4550                  * counted tree.
4551                  */
4552                 if (found_key.objectid == found_key.offset ||
4553                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4554                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4555                         ref_path->root_generation =
4556                                 btrfs_ref_generation(leaf, ref);
4557                         if (level < 0) {
4558                                 /* special reference from the tree log */
4559                                 ref_path->nodes[0] = found_key.offset;
4560                                 ref_path->current_level = 0;
4561                         }
4562                         ret = 0;
4563                         goto out;
4564                 }
4565
4566                 level++;
4567                 BUG_ON(ref_path->nodes[level] != 0);
4568                 ref_path->nodes[level] = found_key.offset;
4569                 ref_path->current_level = level;
4570
4571                 /*
4572                  * the reference was created in the running transaction,
4573                  * no need to continue walking up.
4574                  */
4575                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4576                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4577                         ref_path->root_generation =
4578                                 btrfs_ref_generation(leaf, ref);
4579                         ret = 0;
4580                         goto out;
4581                 }
4582
4583                 btrfs_release_path(extent_root, path);
4584                 cond_resched();
4585         }
4586         /* reached max tree level, but no tree root found. */
4587         BUG();
4588 out:
4589         btrfs_free_path(path);
4590         return ret;
4591 }
4592
4593 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4594                                 struct btrfs_root *extent_root,
4595                                 struct btrfs_ref_path *ref_path,
4596                                 u64 extent_start)
4597 {
4598         memset(ref_path, 0, sizeof(*ref_path));
4599         ref_path->extent_start = extent_start;
4600
4601         return __next_ref_path(trans, extent_root, ref_path, 1);
4602 }
4603
4604 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4605                                struct btrfs_root *extent_root,
4606                                struct btrfs_ref_path *ref_path)
4607 {
4608         return __next_ref_path(trans, extent_root, ref_path, 0);
4609 }
4610
4611 static noinline int get_new_locations(struct inode *reloc_inode,
4612                                       struct btrfs_key *extent_key,
4613                                       u64 offset, int no_fragment,
4614                                       struct disk_extent **extents,
4615                                       int *nr_extents)
4616 {
4617         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4618         struct btrfs_path *path;
4619         struct btrfs_file_extent_item *fi;
4620         struct extent_buffer *leaf;
4621         struct disk_extent *exts = *extents;
4622         struct btrfs_key found_key;
4623         u64 cur_pos;
4624         u64 last_byte;
4625         u32 nritems;
4626         int nr = 0;
4627         int max = *nr_extents;
4628         int ret;
4629
4630         WARN_ON(!no_fragment && *extents);
4631         if (!exts) {
4632                 max = 1;
4633                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4634                 if (!exts)
4635                         return -ENOMEM;
4636         }
4637
4638         path = btrfs_alloc_path();
4639         BUG_ON(!path);
4640
4641         cur_pos = extent_key->objectid - offset;
4642         last_byte = extent_key->objectid + extent_key->offset;
4643         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4644                                        cur_pos, 0);
4645         if (ret < 0)
4646                 goto out;
4647         if (ret > 0) {
4648                 ret = -ENOENT;
4649                 goto out;
4650         }
4651
4652         while (1) {
4653                 leaf = path->nodes[0];
4654                 nritems = btrfs_header_nritems(leaf);
4655                 if (path->slots[0] >= nritems) {
4656                         ret = btrfs_next_leaf(root, path);
4657                         if (ret < 0)
4658                                 goto out;
4659                         if (ret > 0)
4660                                 break;
4661                         leaf = path->nodes[0];
4662                 }
4663
4664                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4665                 if (found_key.offset != cur_pos ||
4666                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4667                     found_key.objectid != reloc_inode->i_ino)
4668                         break;
4669
4670                 fi = btrfs_item_ptr(leaf, path->slots[0],
4671                                     struct btrfs_file_extent_item);
4672                 if (btrfs_file_extent_type(leaf, fi) !=
4673                     BTRFS_FILE_EXTENT_REG ||
4674                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4675                         break;
4676
4677                 if (nr == max) {
4678                         struct disk_extent *old = exts;
4679                         max *= 2;
4680                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4681                         memcpy(exts, old, sizeof(*exts) * nr);
4682                         if (old != *extents)
4683                                 kfree(old);
4684                 }
4685
4686                 exts[nr].disk_bytenr =
4687                         btrfs_file_extent_disk_bytenr(leaf, fi);
4688                 exts[nr].disk_num_bytes =
4689                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4690                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4691                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4692                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4693                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4694                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4695                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4696                                                                            fi);
4697                 BUG_ON(exts[nr].offset > 0);
4698                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4699                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4700
4701                 cur_pos += exts[nr].num_bytes;
4702                 nr++;
4703
4704                 if (cur_pos + offset >= last_byte)
4705                         break;
4706
4707                 if (no_fragment) {
4708                         ret = 1;
4709                         goto out;
4710                 }
4711                 path->slots[0]++;
4712         }
4713
4714         BUG_ON(cur_pos + offset > last_byte);
4715         if (cur_pos + offset < last_byte) {
4716                 ret = -ENOENT;
4717                 goto out;
4718         }
4719         ret = 0;
4720 out:
4721         btrfs_free_path(path);
4722         if (ret) {
4723                 if (exts != *extents)
4724                         kfree(exts);
4725         } else {
4726                 *extents = exts;
4727                 *nr_extents = nr;
4728         }
4729         return ret;
4730 }
4731
4732 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4733                                         struct btrfs_root *root,
4734                                         struct btrfs_path *path,
4735                                         struct btrfs_key *extent_key,
4736                                         struct btrfs_key *leaf_key,
4737                                         struct btrfs_ref_path *ref_path,
4738                                         struct disk_extent *new_extents,
4739                                         int nr_extents)
4740 {
4741         struct extent_buffer *leaf;
4742         struct btrfs_file_extent_item *fi;
4743         struct inode *inode = NULL;
4744         struct btrfs_key key;
4745         u64 lock_start = 0;
4746         u64 lock_end = 0;
4747         u64 num_bytes;
4748         u64 ext_offset;
4749         u64 search_end = (u64)-1;
4750         u32 nritems;
4751         int nr_scaned = 0;
4752         int extent_locked = 0;
4753         int extent_type;
4754         int ret;
4755
4756         memcpy(&key, leaf_key, sizeof(key));
4757         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4758                 if (key.objectid < ref_path->owner_objectid ||
4759                     (key.objectid == ref_path->owner_objectid &&
4760                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4761                         key.objectid = ref_path->owner_objectid;
4762                         key.type = BTRFS_EXTENT_DATA_KEY;
4763                         key.offset = 0;
4764                 }
4765         }
4766
4767         while (1) {
4768                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4769                 if (ret < 0)
4770                         goto out;
4771
4772                 leaf = path->nodes[0];
4773                 nritems = btrfs_header_nritems(leaf);
4774 next:
4775                 if (extent_locked && ret > 0) {
4776                         /*
4777                          * the file extent item was modified by someone
4778                          * before the extent got locked.
4779                          */
4780                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4781                                       lock_end, GFP_NOFS);
4782                         extent_locked = 0;
4783                 }
4784
4785                 if (path->slots[0] >= nritems) {
4786                         if (++nr_scaned > 2)
4787                                 break;
4788
4789                         BUG_ON(extent_locked);
4790                         ret = btrfs_next_leaf(root, path);
4791                         if (ret < 0)
4792                                 goto out;
4793                         if (ret > 0)
4794                                 break;
4795                         leaf = path->nodes[0];
4796                         nritems = btrfs_header_nritems(leaf);
4797                 }
4798
4799                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4800
4801                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4802                         if ((key.objectid > ref_path->owner_objectid) ||
4803                             (key.objectid == ref_path->owner_objectid &&
4804                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4805                             key.offset >= search_end)
4806                                 break;
4807                 }
4808
4809                 if (inode && key.objectid != inode->i_ino) {
4810                         BUG_ON(extent_locked);
4811                         btrfs_release_path(root, path);
4812                         mutex_unlock(&inode->i_mutex);
4813                         iput(inode);
4814                         inode = NULL;
4815                         continue;
4816                 }
4817
4818                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4819                         path->slots[0]++;
4820                         ret = 1;
4821                         goto next;
4822                 }
4823                 fi = btrfs_item_ptr(leaf, path->slots[0],
4824                                     struct btrfs_file_extent_item);
4825                 extent_type = btrfs_file_extent_type(leaf, fi);
4826                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4827                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4828                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4829                      extent_key->objectid)) {
4830                         path->slots[0]++;
4831                         ret = 1;
4832                         goto next;
4833                 }
4834
4835                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4836                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4837
4838                 if (search_end == (u64)-1) {
4839                         search_end = key.offset - ext_offset +
4840                                 btrfs_file_extent_ram_bytes(leaf, fi);
4841                 }
4842
4843                 if (!extent_locked) {
4844                         lock_start = key.offset;
4845                         lock_end = lock_start + num_bytes - 1;
4846                 } else {
4847                         if (lock_start > key.offset ||
4848                             lock_end + 1 < key.offset + num_bytes) {
4849                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4850                                               lock_start, lock_end, GFP_NOFS);
4851                                 extent_locked = 0;
4852                         }
4853                 }
4854
4855                 if (!inode) {
4856                         btrfs_release_path(root, path);
4857
4858                         inode = btrfs_iget_locked(root->fs_info->sb,
4859                                                   key.objectid, root);
4860                         if (inode->i_state & I_NEW) {
4861                                 BTRFS_I(inode)->root = root;
4862                                 BTRFS_I(inode)->location.objectid =
4863                                         key.objectid;
4864                                 BTRFS_I(inode)->location.type =
4865                                         BTRFS_INODE_ITEM_KEY;
4866                                 BTRFS_I(inode)->location.offset = 0;
4867                                 btrfs_read_locked_inode(inode);
4868                                 unlock_new_inode(inode);
4869                         }
4870                         /*
4871                          * some code call btrfs_commit_transaction while
4872                          * holding the i_mutex, so we can't use mutex_lock
4873                          * here.
4874                          */
4875                         if (is_bad_inode(inode) ||
4876                             !mutex_trylock(&inode->i_mutex)) {
4877                                 iput(inode);
4878                                 inode = NULL;
4879                                 key.offset = (u64)-1;
4880                                 goto skip;
4881                         }
4882                 }
4883
4884                 if (!extent_locked) {
4885                         struct btrfs_ordered_extent *ordered;
4886
4887                         btrfs_release_path(root, path);
4888
4889                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4890                                     lock_end, GFP_NOFS);
4891                         ordered = btrfs_lookup_first_ordered_extent(inode,
4892                                                                     lock_end);
4893                         if (ordered &&
4894                             ordered->file_offset <= lock_end &&
4895                             ordered->file_offset + ordered->len > lock_start) {
4896                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4897                                               lock_start, lock_end, GFP_NOFS);
4898                                 btrfs_start_ordered_extent(inode, ordered, 1);
4899                                 btrfs_put_ordered_extent(ordered);
4900                                 key.offset += num_bytes;
4901                                 goto skip;
4902                         }
4903                         if (ordered)
4904                                 btrfs_put_ordered_extent(ordered);
4905
4906                         extent_locked = 1;
4907                         continue;
4908                 }
4909
4910                 if (nr_extents == 1) {
4911                         /* update extent pointer in place */
4912                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4913                                                 new_extents[0].disk_bytenr);
4914                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4915                                                 new_extents[0].disk_num_bytes);
4916                         btrfs_mark_buffer_dirty(leaf);
4917
4918                         btrfs_drop_extent_cache(inode, key.offset,
4919                                                 key.offset + num_bytes - 1, 0);
4920
4921                         ret = btrfs_inc_extent_ref(trans, root,
4922                                                 new_extents[0].disk_bytenr,
4923                                                 new_extents[0].disk_num_bytes,
4924                                                 leaf->start,
4925                                                 root->root_key.objectid,
4926                                                 trans->transid,
4927                                                 key.objectid);
4928                         BUG_ON(ret);
4929
4930                         ret = btrfs_free_extent(trans, root,
4931                                                 extent_key->objectid,
4932                                                 extent_key->offset,
4933                                                 leaf->start,
4934                                                 btrfs_header_owner(leaf),
4935                                                 btrfs_header_generation(leaf),
4936                                                 key.objectid, 0);
4937                         BUG_ON(ret);
4938
4939                         btrfs_release_path(root, path);
4940                         key.offset += num_bytes;
4941                 } else {
4942                         BUG_ON(1);
4943 #if 0
4944                         u64 alloc_hint;
4945                         u64 extent_len;
4946                         int i;
4947                         /*
4948                          * drop old extent pointer at first, then insert the
4949                          * new pointers one bye one
4950                          */
4951                         btrfs_release_path(root, path);
4952                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4953                                                  key.offset + num_bytes,
4954                                                  key.offset, &alloc_hint);
4955                         BUG_ON(ret);
4956
4957                         for (i = 0; i < nr_extents; i++) {
4958                                 if (ext_offset >= new_extents[i].num_bytes) {
4959                                         ext_offset -= new_extents[i].num_bytes;
4960                                         continue;
4961                                 }
4962                                 extent_len = min(new_extents[i].num_bytes -
4963                                                  ext_offset, num_bytes);
4964
4965                                 ret = btrfs_insert_empty_item(trans, root,
4966                                                               path, &key,
4967                                                               sizeof(*fi));
4968                                 BUG_ON(ret);
4969
4970                                 leaf = path->nodes[0];
4971                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4972                                                 struct btrfs_file_extent_item);
4973                                 btrfs_set_file_extent_generation(leaf, fi,
4974                                                         trans->transid);
4975                                 btrfs_set_file_extent_type(leaf, fi,
4976                                                         BTRFS_FILE_EXTENT_REG);
4977                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4978                                                 new_extents[i].disk_bytenr);
4979                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4980                                                 new_extents[i].disk_num_bytes);
4981                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4982                                                 new_extents[i].ram_bytes);
4983
4984                                 btrfs_set_file_extent_compression(leaf, fi,
4985                                                 new_extents[i].compression);
4986                                 btrfs_set_file_extent_encryption(leaf, fi,
4987                                                 new_extents[i].encryption);
4988                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4989                                                 new_extents[i].other_encoding);
4990
4991                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4992                                                         extent_len);
4993                                 ext_offset += new_extents[i].offset;
4994                                 btrfs_set_file_extent_offset(leaf, fi,
4995                                                         ext_offset);
4996                                 btrfs_mark_buffer_dirty(leaf);
4997
4998                                 btrfs_drop_extent_cache(inode, key.offset,
4999                                                 key.offset + extent_len - 1, 0);
5000
5001                                 ret = btrfs_inc_extent_ref(trans, root,
5002                                                 new_extents[i].disk_bytenr,
5003                                                 new_extents[i].disk_num_bytes,
5004                                                 leaf->start,
5005                                                 root->root_key.objectid,
5006                                                 trans->transid, key.objectid);
5007                                 BUG_ON(ret);
5008                                 btrfs_release_path(root, path);
5009
5010                                 inode_add_bytes(inode, extent_len);
5011
5012                                 ext_offset = 0;
5013                                 num_bytes -= extent_len;
5014                                 key.offset += extent_len;
5015
5016                                 if (num_bytes == 0)
5017                                         break;
5018                         }
5019                         BUG_ON(i >= nr_extents);
5020 #endif
5021                 }
5022
5023                 if (extent_locked) {
5024                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5025                                       lock_end, GFP_NOFS);
5026                         extent_locked = 0;
5027                 }
5028 skip:
5029                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
5030                     key.offset >= search_end)
5031                         break;
5032
5033                 cond_resched();
5034         }
5035         ret = 0;
5036 out:
5037         btrfs_release_path(root, path);
5038         if (inode) {
5039                 mutex_unlock(&inode->i_mutex);
5040                 if (extent_locked) {
5041                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5042                                       lock_end, GFP_NOFS);
5043                 }
5044                 iput(inode);
5045         }
5046         return ret;
5047 }
5048
5049 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
5050                                struct btrfs_root *root,
5051                                struct extent_buffer *buf, u64 orig_start)
5052 {
5053         int level;
5054         int ret;
5055
5056         BUG_ON(btrfs_header_generation(buf) != trans->transid);
5057         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5058
5059         level = btrfs_header_level(buf);
5060         if (level == 0) {
5061                 struct btrfs_leaf_ref *ref;
5062                 struct btrfs_leaf_ref *orig_ref;
5063
5064                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
5065                 if (!orig_ref)
5066                         return -ENOENT;
5067
5068                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
5069                 if (!ref) {
5070                         btrfs_free_leaf_ref(root, orig_ref);
5071                         return -ENOMEM;
5072                 }
5073
5074                 ref->nritems = orig_ref->nritems;
5075                 memcpy(ref->extents, orig_ref->extents,
5076                         sizeof(ref->extents[0]) * ref->nritems);
5077
5078                 btrfs_free_leaf_ref(root, orig_ref);
5079
5080                 ref->root_gen = trans->transid;
5081                 ref->bytenr = buf->start;
5082                 ref->owner = btrfs_header_owner(buf);
5083                 ref->generation = btrfs_header_generation(buf);
5084
5085                 ret = btrfs_add_leaf_ref(root, ref, 0);
5086                 WARN_ON(ret);
5087                 btrfs_free_leaf_ref(root, ref);
5088         }
5089         return 0;
5090 }
5091
5092 static noinline int invalidate_extent_cache(struct btrfs_root *root,
5093                                         struct extent_buffer *leaf,
5094                                         struct btrfs_block_group_cache *group,
5095                                         struct btrfs_root *target_root)
5096 {
5097         struct btrfs_key key;
5098         struct inode *inode = NULL;
5099         struct btrfs_file_extent_item *fi;
5100         u64 num_bytes;
5101         u64 skip_objectid = 0;
5102         u32 nritems;
5103         u32 i;
5104
5105         nritems = btrfs_header_nritems(leaf);
5106         for (i = 0; i < nritems; i++) {
5107                 btrfs_item_key_to_cpu(leaf, &key, i);
5108                 if (key.objectid == skip_objectid ||
5109                     key.type != BTRFS_EXTENT_DATA_KEY)
5110                         continue;
5111                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5112                 if (btrfs_file_extent_type(leaf, fi) ==
5113                     BTRFS_FILE_EXTENT_INLINE)
5114                         continue;
5115                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5116                         continue;
5117                 if (!inode || inode->i_ino != key.objectid) {
5118                         iput(inode);
5119                         inode = btrfs_ilookup(target_root->fs_info->sb,
5120                                               key.objectid, target_root, 1);
5121                 }
5122                 if (!inode) {
5123                         skip_objectid = key.objectid;
5124                         continue;
5125                 }
5126                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5127
5128                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5129                             key.offset + num_bytes - 1, GFP_NOFS);
5130                 btrfs_drop_extent_cache(inode, key.offset,
5131                                         key.offset + num_bytes - 1, 1);
5132                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5133                               key.offset + num_bytes - 1, GFP_NOFS);
5134                 cond_resched();
5135         }
5136         iput(inode);
5137         return 0;
5138 }
5139
5140 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
5141                                         struct btrfs_root *root,
5142                                         struct extent_buffer *leaf,
5143                                         struct btrfs_block_group_cache *group,
5144                                         struct inode *reloc_inode)
5145 {
5146         struct btrfs_key key;
5147         struct btrfs_key extent_key;
5148         struct btrfs_file_extent_item *fi;
5149         struct btrfs_leaf_ref *ref;
5150         struct disk_extent *new_extent;
5151         u64 bytenr;
5152         u64 num_bytes;
5153         u32 nritems;
5154         u32 i;
5155         int ext_index;
5156         int nr_extent;
5157         int ret;
5158
5159         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
5160         BUG_ON(!new_extent);
5161
5162         ref = btrfs_lookup_leaf_ref(root, leaf->start);
5163         BUG_ON(!ref);
5164
5165         ext_index = -1;
5166         nritems = btrfs_header_nritems(leaf);
5167         for (i = 0; i < nritems; i++) {
5168                 btrfs_item_key_to_cpu(leaf, &key, i);
5169                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
5170                         continue;
5171                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5172                 if (btrfs_file_extent_type(leaf, fi) ==
5173                     BTRFS_FILE_EXTENT_INLINE)
5174                         continue;
5175                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5176                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
5177                 if (bytenr == 0)
5178                         continue;
5179
5180                 ext_index++;
5181                 if (bytenr >= group->key.objectid + group->key.offset ||
5182                     bytenr + num_bytes <= group->key.objectid)
5183                         continue;
5184
5185                 extent_key.objectid = bytenr;
5186                 extent_key.offset = num_bytes;
5187                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
5188                 nr_extent = 1;
5189                 ret = get_new_locations(reloc_inode, &extent_key,
5190                                         group->key.objectid, 1,
5191                                         &new_extent, &nr_extent);
5192                 if (ret > 0)
5193                         continue;
5194                 BUG_ON(ret < 0);
5195
5196                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
5197                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
5198                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
5199                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
5200
5201                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5202                                                 new_extent->disk_bytenr);
5203                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5204                                                 new_extent->disk_num_bytes);
5205                 btrfs_mark_buffer_dirty(leaf);
5206
5207                 ret = btrfs_inc_extent_ref(trans, root,
5208                                         new_extent->disk_bytenr,
5209                                         new_extent->disk_num_bytes,
5210                                         leaf->start,
5211                                         root->root_key.objectid,
5212                                         trans->transid, key.objectid);
5213                 BUG_ON(ret);
5214                 ret = btrfs_free_extent(trans, root,
5215                                         bytenr, num_bytes, leaf->start,
5216                                         btrfs_header_owner(leaf),
5217                                         btrfs_header_generation(leaf),
5218                                         key.objectid, 0);
5219                 BUG_ON(ret);
5220                 cond_resched();
5221         }
5222         kfree(new_extent);
5223         BUG_ON(ext_index + 1 != ref->nritems);
5224         btrfs_free_leaf_ref(root, ref);
5225         return 0;
5226 }
5227
5228 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
5229                           struct btrfs_root *root)
5230 {
5231         struct btrfs_root *reloc_root;
5232         int ret;
5233
5234         if (root->reloc_root) {
5235                 reloc_root = root->reloc_root;
5236                 root->reloc_root = NULL;
5237                 list_add(&reloc_root->dead_list,
5238                          &root->fs_info->dead_reloc_roots);
5239
5240                 btrfs_set_root_bytenr(&reloc_root->root_item,
5241                                       reloc_root->node->start);
5242                 btrfs_set_root_level(&root->root_item,
5243                                      btrfs_header_level(reloc_root->node));
5244                 memset(&reloc_root->root_item.drop_progress, 0,
5245                         sizeof(struct btrfs_disk_key));
5246                 reloc_root->root_item.drop_level = 0;
5247
5248                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
5249                                         &reloc_root->root_key,
5250                                         &reloc_root->root_item);
5251                 BUG_ON(ret);
5252         }
5253         return 0;
5254 }
5255
5256 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
5257 {
5258         struct btrfs_trans_handle *trans;
5259         struct btrfs_root *reloc_root;
5260         struct btrfs_root *prev_root = NULL;
5261         struct list_head dead_roots;
5262         int ret;
5263         unsigned long nr;
5264
5265         INIT_LIST_HEAD(&dead_roots);
5266         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
5267
5268         while (!list_empty(&dead_roots)) {
5269                 reloc_root = list_entry(dead_roots.prev,
5270                                         struct btrfs_root, dead_list);
5271                 list_del_init(&reloc_root->dead_list);
5272
5273                 BUG_ON(reloc_root->commit_root != NULL);
5274                 while (1) {
5275                         trans = btrfs_join_transaction(root, 1);
5276                         BUG_ON(!trans);
5277
5278                         mutex_lock(&root->fs_info->drop_mutex);
5279                         ret = btrfs_drop_snapshot(trans, reloc_root);
5280                         if (ret != -EAGAIN)
5281                                 break;
5282                         mutex_unlock(&root->fs_info->drop_mutex);
5283
5284                         nr = trans->blocks_used;
5285                         ret = btrfs_end_transaction(trans, root);
5286                         BUG_ON(ret);
5287                         btrfs_btree_balance_dirty(root, nr);
5288                 }
5289
5290                 free_extent_buffer(reloc_root->node);
5291
5292                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
5293                                      &reloc_root->root_key);
5294                 BUG_ON(ret);
5295                 mutex_unlock(&root->fs_info->drop_mutex);
5296
5297                 nr = trans->blocks_used;
5298                 ret = btrfs_end_transaction(trans, root);
5299                 BUG_ON(ret);
5300                 btrfs_btree_balance_dirty(root, nr);
5301
5302                 kfree(prev_root);
5303                 prev_root = reloc_root;
5304         }
5305         if (prev_root) {
5306                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5307                 kfree(prev_root);
5308         }
5309         return 0;
5310 }
5311
5312 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5313 {
5314         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5315         return 0;
5316 }
5317
5318 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5319 {
5320         struct btrfs_root *reloc_root;
5321         struct btrfs_trans_handle *trans;
5322         struct btrfs_key location;
5323         int found;
5324         int ret;
5325
5326         mutex_lock(&root->fs_info->tree_reloc_mutex);
5327         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5328         BUG_ON(ret);
5329         found = !list_empty(&root->fs_info->dead_reloc_roots);
5330         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5331
5332         if (found) {
5333                 trans = btrfs_start_transaction(root, 1);
5334                 BUG_ON(!trans);
5335                 ret = btrfs_commit_transaction(trans, root);
5336                 BUG_ON(ret);
5337         }
5338
5339         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5340         location.offset = (u64)-1;
5341         location.type = BTRFS_ROOT_ITEM_KEY;
5342
5343         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5344         BUG_ON(!reloc_root);
5345         btrfs_orphan_cleanup(reloc_root);
5346         return 0;
5347 }
5348
5349 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5350                                     struct btrfs_root *root)
5351 {
5352         struct btrfs_root *reloc_root;
5353         struct extent_buffer *eb;
5354         struct btrfs_root_item *root_item;
5355         struct btrfs_key root_key;
5356         int ret;
5357
5358         BUG_ON(!root->ref_cows);
5359         if (root->reloc_root)
5360                 return 0;
5361
5362         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5363         BUG_ON(!root_item);
5364
5365         ret = btrfs_copy_root(trans, root, root->commit_root,
5366                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5367         BUG_ON(ret);
5368
5369         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5370         root_key.offset = root->root_key.objectid;
5371         root_key.type = BTRFS_ROOT_ITEM_KEY;
5372
5373         memcpy(root_item, &root->root_item, sizeof(root_item));
5374         btrfs_set_root_refs(root_item, 0);
5375         btrfs_set_root_bytenr(root_item, eb->start);
5376         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5377         btrfs_set_root_generation(root_item, trans->transid);
5378
5379         btrfs_tree_unlock(eb);
5380         free_extent_buffer(eb);
5381
5382         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5383                                 &root_key, root_item);
5384         BUG_ON(ret);
5385         kfree(root_item);
5386
5387         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5388                                                  &root_key);
5389         BUG_ON(!reloc_root);
5390         reloc_root->last_trans = trans->transid;
5391         reloc_root->commit_root = NULL;
5392         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5393
5394         root->reloc_root = reloc_root;
5395         return 0;
5396 }
5397
5398 /*
5399  * Core function of space balance.
5400  *
5401  * The idea is using reloc trees to relocate tree blocks in reference
5402  * counted roots. There is one reloc tree for each subvol, and all
5403  * reloc trees share same root key objectid. Reloc trees are snapshots
5404  * of the latest committed roots of subvols (root->commit_root).
5405  *
5406  * To relocate a tree block referenced by a subvol, there are two steps.
5407  * COW the block through subvol's reloc tree, then update block pointer
5408  * in the subvol to point to the new block. Since all reloc trees share
5409  * same root key objectid, doing special handing for tree blocks owned
5410  * by them is easy. Once a tree block has been COWed in one reloc tree,
5411  * we can use the resulting new block directly when the same block is
5412  * required to COW again through other reloc trees. By this way, relocated
5413  * tree blocks are shared between reloc trees, so they are also shared
5414  * between subvols.
5415  */
5416 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5417                                       struct btrfs_root *root,
5418                                       struct btrfs_path *path,
5419                                       struct btrfs_key *first_key,
5420                                       struct btrfs_ref_path *ref_path,
5421                                       struct btrfs_block_group_cache *group,
5422                                       struct inode *reloc_inode)
5423 {
5424         struct btrfs_root *reloc_root;
5425         struct extent_buffer *eb = NULL;
5426         struct btrfs_key *keys;
5427         u64 *nodes;
5428         int level;
5429         int shared_level;
5430         int lowest_level = 0;
5431         int ret;
5432
5433         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5434                 lowest_level = ref_path->owner_objectid;
5435
5436         if (!root->ref_cows) {
5437                 path->lowest_level = lowest_level;
5438                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5439                 BUG_ON(ret < 0);
5440                 path->lowest_level = 0;
5441                 btrfs_release_path(root, path);
5442                 return 0;
5443         }
5444
5445         mutex_lock(&root->fs_info->tree_reloc_mutex);
5446         ret = init_reloc_tree(trans, root);
5447         BUG_ON(ret);
5448         reloc_root = root->reloc_root;
5449
5450         shared_level = ref_path->shared_level;
5451         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5452
5453         keys = ref_path->node_keys;
5454         nodes = ref_path->new_nodes;
5455         memset(&keys[shared_level + 1], 0,
5456                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5457         memset(&nodes[shared_level + 1], 0,
5458                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5459
5460         if (nodes[lowest_level] == 0) {
5461                 path->lowest_level = lowest_level;
5462                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5463                                         0, 1);
5464                 BUG_ON(ret);
5465                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5466                         eb = path->nodes[level];
5467                         if (!eb || eb == reloc_root->node)
5468                                 break;
5469                         nodes[level] = eb->start;
5470                         if (level == 0)
5471                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5472                         else
5473                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5474                 }
5475                 if (nodes[0] &&
5476                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5477                         eb = path->nodes[0];
5478                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5479                                                       group, reloc_inode);
5480                         BUG_ON(ret);
5481                 }
5482                 btrfs_release_path(reloc_root, path);
5483         } else {
5484                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5485                                        lowest_level);
5486                 BUG_ON(ret);
5487         }
5488
5489         /*
5490          * replace tree blocks in the fs tree with tree blocks in
5491          * the reloc tree.
5492          */
5493         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5494         BUG_ON(ret < 0);
5495
5496         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5497                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5498                                         0, 0);
5499                 BUG_ON(ret);
5500                 extent_buffer_get(path->nodes[0]);
5501                 eb = path->nodes[0];
5502                 btrfs_release_path(reloc_root, path);
5503                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5504                 BUG_ON(ret);
5505                 free_extent_buffer(eb);
5506         }
5507
5508         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5509         path->lowest_level = 0;
5510         return 0;
5511 }
5512
5513 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5514                                         struct btrfs_root *root,
5515                                         struct btrfs_path *path,
5516                                         struct btrfs_key *first_key,
5517                                         struct btrfs_ref_path *ref_path)
5518 {
5519         int ret;
5520
5521         ret = relocate_one_path(trans, root, path, first_key,
5522                                 ref_path, NULL, NULL);
5523         BUG_ON(ret);
5524
5525         if (root == root->fs_info->extent_root)
5526                 btrfs_extent_post_op(trans, root);
5527
5528         return 0;
5529 }
5530
5531 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5532                                     struct btrfs_root *extent_root,
5533                                     struct btrfs_path *path,
5534                                     struct btrfs_key *extent_key)
5535 {
5536         int ret;
5537
5538         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5539         if (ret)
5540                 goto out;
5541         ret = btrfs_del_item(trans, extent_root, path);
5542 out:
5543         btrfs_release_path(extent_root, path);
5544         return ret;
5545 }
5546
5547 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5548                                                 struct btrfs_ref_path *ref_path)
5549 {
5550         struct btrfs_key root_key;
5551
5552         root_key.objectid = ref_path->root_objectid;
5553         root_key.type = BTRFS_ROOT_ITEM_KEY;
5554         if (is_cowonly_root(ref_path->root_objectid))
5555                 root_key.offset = 0;
5556         else
5557                 root_key.offset = (u64)-1;
5558
5559         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5560 }
5561
5562 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5563                                         struct btrfs_path *path,
5564                                         struct btrfs_key *extent_key,
5565                                         struct btrfs_block_group_cache *group,
5566                                         struct inode *reloc_inode, int pass)
5567 {
5568         struct btrfs_trans_handle *trans;
5569         struct btrfs_root *found_root;
5570         struct btrfs_ref_path *ref_path = NULL;
5571         struct disk_extent *new_extents = NULL;
5572         int nr_extents = 0;
5573         int loops;
5574         int ret;
5575         int level;
5576         struct btrfs_key first_key;
5577         u64 prev_block = 0;
5578
5579
5580         trans = btrfs_start_transaction(extent_root, 1);
5581         BUG_ON(!trans);
5582
5583         if (extent_key->objectid == 0) {
5584                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5585                 goto out;
5586         }
5587
5588         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5589         if (!ref_path) {
5590                 ret = -ENOMEM;
5591                 goto out;
5592         }
5593
5594         for (loops = 0; ; loops++) {
5595                 if (loops == 0) {
5596                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5597                                                    extent_key->objectid);
5598                 } else {
5599                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5600                 }
5601                 if (ret < 0)
5602                         goto out;
5603                 if (ret > 0)
5604                         break;
5605
5606                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5607                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5608                         continue;
5609
5610                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5611                 BUG_ON(!found_root);
5612                 /*
5613                  * for reference counted tree, only process reference paths
5614                  * rooted at the latest committed root.
5615                  */
5616                 if (found_root->ref_cows &&
5617                     ref_path->root_generation != found_root->root_key.offset)
5618                         continue;
5619
5620                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5621                         if (pass == 0) {
5622                                 /*
5623                                  * copy data extents to new locations
5624                                  */
5625                                 u64 group_start = group->key.objectid;
5626                                 ret = relocate_data_extent(reloc_inode,
5627                                                            extent_key,
5628                                                            group_start);
5629                                 if (ret < 0)
5630                                         goto out;
5631                                 break;
5632                         }
5633                         level = 0;
5634                 } else {
5635                         level = ref_path->owner_objectid;
5636                 }
5637
5638                 if (prev_block != ref_path->nodes[level]) {
5639                         struct extent_buffer *eb;
5640                         u64 block_start = ref_path->nodes[level];
5641                         u64 block_size = btrfs_level_size(found_root, level);
5642
5643                         eb = read_tree_block(found_root, block_start,
5644                                              block_size, 0);
5645                         btrfs_tree_lock(eb);
5646                         BUG_ON(level != btrfs_header_level(eb));
5647
5648                         if (level == 0)
5649                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5650                         else
5651                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5652
5653                         btrfs_tree_unlock(eb);
5654                         free_extent_buffer(eb);
5655                         prev_block = block_start;
5656                 }
5657
5658                 btrfs_record_root_in_trans(found_root);
5659                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5660                         /*
5661                          * try to update data extent references while
5662                          * keeping metadata shared between snapshots.
5663                          */
5664                         if (pass == 1) {
5665                                 ret = relocate_one_path(trans, found_root,
5666                                                 path, &first_key, ref_path,
5667                                                 group, reloc_inode);
5668                                 if (ret < 0)
5669                                         goto out;
5670                                 continue;
5671                         }
5672                         /*
5673                          * use fallback method to process the remaining
5674                          * references.
5675                          */
5676                         if (!new_extents) {
5677                                 u64 group_start = group->key.objectid;
5678                                 new_extents = kmalloc(sizeof(*new_extents),
5679                                                       GFP_NOFS);
5680                                 nr_extents = 1;
5681                                 ret = get_new_locations(reloc_inode,
5682                                                         extent_key,
5683                                                         group_start, 1,
5684                                                         &new_extents,
5685                                                         &nr_extents);
5686                                 if (ret)
5687                                         goto out;
5688                         }
5689                         ret = replace_one_extent(trans, found_root,
5690                                                 path, extent_key,
5691                                                 &first_key, ref_path,
5692                                                 new_extents, nr_extents);
5693                 } else {
5694                         ret = relocate_tree_block(trans, found_root, path,
5695                                                   &first_key, ref_path);
5696                 }
5697                 if (ret < 0)
5698                         goto out;
5699         }
5700         ret = 0;
5701 out:
5702         btrfs_end_transaction(trans, extent_root);
5703         kfree(new_extents);
5704         kfree(ref_path);
5705         return ret;
5706 }
5707
5708 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5709 {
5710         u64 num_devices;
5711         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5712                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5713
5714         num_devices = root->fs_info->fs_devices->rw_devices;
5715         if (num_devices == 1) {
5716                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5717                 stripped = flags & ~stripped;
5718
5719                 /* turn raid0 into single device chunks */
5720                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5721                         return stripped;
5722
5723                 /* turn mirroring into duplication */
5724                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5725                              BTRFS_BLOCK_GROUP_RAID10))
5726                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5727                 return flags;
5728         } else {
5729                 /* they already had raid on here, just return */
5730                 if (flags & stripped)
5731                         return flags;
5732
5733                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5734                 stripped = flags & ~stripped;
5735
5736                 /* switch duplicated blocks with raid1 */
5737                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5738                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5739
5740                 /* turn single device chunks into raid0 */
5741                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5742         }
5743         return flags;
5744 }
5745
5746 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5747                      struct btrfs_block_group_cache *shrink_block_group,
5748                      int force)
5749 {
5750         struct btrfs_trans_handle *trans;
5751         u64 new_alloc_flags;
5752         u64 calc;
5753
5754         spin_lock(&shrink_block_group->lock);
5755         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5756                 spin_unlock(&shrink_block_group->lock);
5757
5758                 trans = btrfs_start_transaction(root, 1);
5759                 spin_lock(&shrink_block_group->lock);
5760
5761                 new_alloc_flags = update_block_group_flags(root,
5762                                                    shrink_block_group->flags);
5763                 if (new_alloc_flags != shrink_block_group->flags) {
5764                         calc =
5765                              btrfs_block_group_used(&shrink_block_group->item);
5766                 } else {
5767                         calc = shrink_block_group->key.offset;
5768                 }
5769                 spin_unlock(&shrink_block_group->lock);
5770
5771                 do_chunk_alloc(trans, root->fs_info->extent_root,
5772                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5773
5774                 btrfs_end_transaction(trans, root);
5775         } else
5776                 spin_unlock(&shrink_block_group->lock);
5777         return 0;
5778 }
5779
5780 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5781                                  struct btrfs_root *root,
5782                                  u64 objectid, u64 size)
5783 {
5784         struct btrfs_path *path;
5785         struct btrfs_inode_item *item;
5786         struct extent_buffer *leaf;
5787         int ret;
5788
5789         path = btrfs_alloc_path();
5790         if (!path)
5791                 return -ENOMEM;
5792
5793         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5794         if (ret)
5795                 goto out;
5796
5797         leaf = path->nodes[0];
5798         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5799         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5800         btrfs_set_inode_generation(leaf, item, 1);
5801         btrfs_set_inode_size(leaf, item, size);
5802         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5803         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5804         btrfs_mark_buffer_dirty(leaf);
5805         btrfs_release_path(root, path);
5806 out:
5807         btrfs_free_path(path);
5808         return ret;
5809 }
5810
5811 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5812                                         struct btrfs_block_group_cache *group)
5813 {
5814         struct inode *inode = NULL;
5815         struct btrfs_trans_handle *trans;
5816         struct btrfs_root *root;
5817         struct btrfs_key root_key;
5818         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5819         int err = 0;
5820
5821         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5822         root_key.type = BTRFS_ROOT_ITEM_KEY;
5823         root_key.offset = (u64)-1;
5824         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5825         if (IS_ERR(root))
5826                 return ERR_CAST(root);
5827
5828         trans = btrfs_start_transaction(root, 1);
5829         BUG_ON(!trans);
5830
5831         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5832         if (err)
5833                 goto out;
5834
5835         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5836         BUG_ON(err);
5837
5838         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5839                                        group->key.offset, 0, group->key.offset,
5840                                        0, 0, 0);
5841         BUG_ON(err);
5842
5843         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5844         if (inode->i_state & I_NEW) {
5845                 BTRFS_I(inode)->root = root;
5846                 BTRFS_I(inode)->location.objectid = objectid;
5847                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5848                 BTRFS_I(inode)->location.offset = 0;
5849                 btrfs_read_locked_inode(inode);
5850                 unlock_new_inode(inode);
5851                 BUG_ON(is_bad_inode(inode));
5852         } else {
5853                 BUG_ON(1);
5854         }
5855         BTRFS_I(inode)->index_cnt = group->key.objectid;
5856
5857         err = btrfs_orphan_add(trans, inode);
5858 out:
5859         btrfs_end_transaction(trans, root);
5860         if (err) {
5861                 if (inode)
5862                         iput(inode);
5863                 inode = ERR_PTR(err);
5864         }
5865         return inode;
5866 }
5867
5868 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5869 {
5870
5871         struct btrfs_ordered_sum *sums;
5872         struct btrfs_sector_sum *sector_sum;
5873         struct btrfs_ordered_extent *ordered;
5874         struct btrfs_root *root = BTRFS_I(inode)->root;
5875         struct list_head list;
5876         size_t offset;
5877         int ret;
5878         u64 disk_bytenr;
5879
5880         INIT_LIST_HEAD(&list);
5881
5882         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5883         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5884
5885         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5886         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5887                                        disk_bytenr + len - 1, &list);
5888
5889         while (!list_empty(&list)) {
5890                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5891                 list_del_init(&sums->list);
5892
5893                 sector_sum = sums->sums;
5894                 sums->bytenr = ordered->start;
5895
5896                 offset = 0;
5897                 while (offset < sums->len) {
5898                         sector_sum->bytenr += ordered->start - disk_bytenr;
5899                         sector_sum++;
5900                         offset += root->sectorsize;
5901                 }
5902
5903                 btrfs_add_ordered_sum(inode, ordered, sums);
5904         }
5905         btrfs_put_ordered_extent(ordered);
5906         return 0;
5907 }
5908
5909 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5910 {
5911         struct btrfs_trans_handle *trans;
5912         struct btrfs_path *path;
5913         struct btrfs_fs_info *info = root->fs_info;
5914         struct extent_buffer *leaf;
5915         struct inode *reloc_inode;
5916         struct btrfs_block_group_cache *block_group;
5917         struct btrfs_key key;
5918         u64 skipped;
5919         u64 cur_byte;
5920         u64 total_found;
5921         u32 nritems;
5922         int ret;
5923         int progress;
5924         int pass = 0;
5925
5926         root = root->fs_info->extent_root;
5927
5928         block_group = btrfs_lookup_block_group(info, group_start);
5929         BUG_ON(!block_group);
5930
5931         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5932                (unsigned long long)block_group->key.objectid,
5933                (unsigned long long)block_group->flags);
5934
5935         path = btrfs_alloc_path();
5936         BUG_ON(!path);
5937
5938         reloc_inode = create_reloc_inode(info, block_group);
5939         BUG_ON(IS_ERR(reloc_inode));
5940
5941         __alloc_chunk_for_shrink(root, block_group, 1);
5942         set_block_group_readonly(block_group);
5943
5944         btrfs_start_delalloc_inodes(info->tree_root);
5945         btrfs_wait_ordered_extents(info->tree_root, 0);
5946 again:
5947         skipped = 0;
5948         total_found = 0;
5949         progress = 0;
5950         key.objectid = block_group->key.objectid;
5951         key.offset = 0;
5952         key.type = 0;
5953         cur_byte = key.objectid;
5954
5955         trans = btrfs_start_transaction(info->tree_root, 1);
5956         btrfs_commit_transaction(trans, info->tree_root);
5957
5958         mutex_lock(&root->fs_info->cleaner_mutex);
5959         btrfs_clean_old_snapshots(info->tree_root);
5960         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5961         mutex_unlock(&root->fs_info->cleaner_mutex);
5962
5963         while (1) {
5964                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5965                 if (ret < 0)
5966                         goto out;
5967 next:
5968                 leaf = path->nodes[0];
5969                 nritems = btrfs_header_nritems(leaf);
5970                 if (path->slots[0] >= nritems) {
5971                         ret = btrfs_next_leaf(root, path);
5972                         if (ret < 0)
5973                                 goto out;
5974                         if (ret == 1) {
5975                                 ret = 0;
5976                                 break;
5977                         }
5978                         leaf = path->nodes[0];
5979                         nritems = btrfs_header_nritems(leaf);
5980                 }
5981
5982                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5983
5984                 if (key.objectid >= block_group->key.objectid +
5985                     block_group->key.offset)
5986                         break;
5987
5988                 if (progress && need_resched()) {
5989                         btrfs_release_path(root, path);
5990                         cond_resched();
5991                         progress = 0;
5992                         continue;
5993                 }
5994                 progress = 1;
5995
5996                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5997                     key.objectid + key.offset <= cur_byte) {
5998                         path->slots[0]++;
5999                         goto next;
6000                 }
6001
6002                 total_found++;
6003                 cur_byte = key.objectid + key.offset;
6004                 btrfs_release_path(root, path);
6005
6006                 __alloc_chunk_for_shrink(root, block_group, 0);
6007                 ret = relocate_one_extent(root, path, &key, block_group,
6008                                           reloc_inode, pass);
6009                 BUG_ON(ret < 0);
6010                 if (ret > 0)
6011                         skipped++;
6012
6013                 key.objectid = cur_byte;
6014                 key.type = 0;
6015                 key.offset = 0;
6016         }
6017
6018         btrfs_release_path(root, path);
6019
6020         if (pass == 0) {
6021                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
6022                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
6023         }
6024
6025         if (total_found > 0) {
6026                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
6027                        (unsigned long long)total_found, pass);
6028                 pass++;
6029                 if (total_found == skipped && pass > 2) {
6030                         iput(reloc_inode);
6031                         reloc_inode = create_reloc_inode(info, block_group);
6032                         pass = 0;
6033                 }
6034                 goto again;
6035         }
6036
6037         /* delete reloc_inode */
6038         iput(reloc_inode);
6039
6040         /* unpin extents in this range */
6041         trans = btrfs_start_transaction(info->tree_root, 1);
6042         btrfs_commit_transaction(trans, info->tree_root);
6043
6044         spin_lock(&block_group->lock);
6045         WARN_ON(block_group->pinned > 0);
6046         WARN_ON(block_group->reserved > 0);
6047         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
6048         spin_unlock(&block_group->lock);
6049         put_block_group(block_group);
6050         ret = 0;
6051 out:
6052         btrfs_free_path(path);
6053         return ret;
6054 }
6055
6056 static int find_first_block_group(struct btrfs_root *root,
6057                 struct btrfs_path *path, struct btrfs_key *key)
6058 {
6059         int ret = 0;
6060         struct btrfs_key found_key;
6061         struct extent_buffer *leaf;
6062         int slot;
6063
6064         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6065         if (ret < 0)
6066                 goto out;
6067
6068         while (1) {
6069                 slot = path->slots[0];
6070                 leaf = path->nodes[0];
6071                 if (slot >= btrfs_header_nritems(leaf)) {
6072                         ret = btrfs_next_leaf(root, path);
6073                         if (ret == 0)
6074                                 continue;
6075                         if (ret < 0)
6076                                 goto out;
6077                         break;
6078                 }
6079                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6080
6081                 if (found_key.objectid >= key->objectid &&
6082                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6083                         ret = 0;
6084                         goto out;
6085                 }
6086                 path->slots[0]++;
6087         }
6088         ret = -ENOENT;
6089 out:
6090         return ret;
6091 }
6092
6093 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6094 {
6095         struct btrfs_block_group_cache *block_group;
6096         struct rb_node *n;
6097
6098         spin_lock(&info->block_group_cache_lock);
6099         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6100                 block_group = rb_entry(n, struct btrfs_block_group_cache,
6101                                        cache_node);
6102                 rb_erase(&block_group->cache_node,
6103                          &info->block_group_cache_tree);
6104                 spin_unlock(&info->block_group_cache_lock);
6105
6106                 btrfs_remove_free_space_cache(block_group);
6107                 down_write(&block_group->space_info->groups_sem);
6108                 list_del(&block_group->list);
6109                 up_write(&block_group->space_info->groups_sem);
6110
6111                 WARN_ON(atomic_read(&block_group->count) != 1);
6112                 kfree(block_group);
6113
6114                 spin_lock(&info->block_group_cache_lock);
6115         }
6116         spin_unlock(&info->block_group_cache_lock);
6117         return 0;
6118 }
6119
6120 int btrfs_read_block_groups(struct btrfs_root *root)
6121 {
6122         struct btrfs_path *path;
6123         int ret;
6124         struct btrfs_block_group_cache *cache;
6125         struct btrfs_fs_info *info = root->fs_info;
6126         struct btrfs_space_info *space_info;
6127         struct btrfs_key key;
6128         struct btrfs_key found_key;
6129         struct extent_buffer *leaf;
6130
6131         root = info->extent_root;
6132         key.objectid = 0;
6133         key.offset = 0;
6134         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
6135         path = btrfs_alloc_path();
6136         if (!path)
6137                 return -ENOMEM;
6138
6139         while (1) {
6140                 ret = find_first_block_group(root, path, &key);
6141                 if (ret > 0) {
6142                         ret = 0;
6143                         goto error;
6144                 }
6145                 if (ret != 0)
6146                         goto error;
6147
6148                 leaf = path->nodes[0];
6149                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6150                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6151                 if (!cache) {
6152                         ret = -ENOMEM;
6153                         break;
6154                 }
6155
6156                 atomic_set(&cache->count, 1);
6157                 spin_lock_init(&cache->lock);
6158                 mutex_init(&cache->alloc_mutex);
6159                 mutex_init(&cache->cache_mutex);
6160                 INIT_LIST_HEAD(&cache->list);
6161                 read_extent_buffer(leaf, &cache->item,
6162                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
6163                                    sizeof(cache->item));
6164                 memcpy(&cache->key, &found_key, sizeof(found_key));
6165
6166                 key.objectid = found_key.objectid + found_key.offset;
6167                 btrfs_release_path(root, path);
6168                 cache->flags = btrfs_block_group_flags(&cache->item);
6169
6170                 ret = update_space_info(info, cache->flags, found_key.offset,
6171                                         btrfs_block_group_used(&cache->item),
6172                                         &space_info);
6173                 BUG_ON(ret);
6174                 cache->space_info = space_info;
6175                 down_write(&space_info->groups_sem);
6176                 list_add_tail(&cache->list, &space_info->block_groups);
6177                 up_write(&space_info->groups_sem);
6178
6179                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
6180                 BUG_ON(ret);
6181
6182                 set_avail_alloc_bits(root->fs_info, cache->flags);
6183                 if (btrfs_chunk_readonly(root, cache->key.objectid))
6184                         set_block_group_readonly(cache);
6185         }
6186         ret = 0;
6187 error:
6188         btrfs_free_path(path);
6189         return ret;
6190 }
6191
6192 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
6193                            struct btrfs_root *root, u64 bytes_used,
6194                            u64 type, u64 chunk_objectid, u64 chunk_offset,
6195                            u64 size)
6196 {
6197         int ret;
6198         struct btrfs_root *extent_root;
6199         struct btrfs_block_group_cache *cache;
6200
6201         extent_root = root->fs_info->extent_root;
6202
6203         root->fs_info->last_trans_new_blockgroup = trans->transid;
6204
6205         cache = kzalloc(sizeof(*cache), GFP_NOFS);
6206         if (!cache)
6207                 return -ENOMEM;
6208
6209         cache->key.objectid = chunk_offset;
6210         cache->key.offset = size;
6211         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
6212         atomic_set(&cache->count, 1);
6213         spin_lock_init(&cache->lock);
6214         mutex_init(&cache->alloc_mutex);
6215         mutex_init(&cache->cache_mutex);
6216         INIT_LIST_HEAD(&cache->list);
6217
6218         btrfs_set_block_group_used(&cache->item, bytes_used);
6219         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
6220         cache->flags = type;
6221         btrfs_set_block_group_flags(&cache->item, type);
6222
6223         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
6224                                 &cache->space_info);
6225         BUG_ON(ret);
6226         down_write(&cache->space_info->groups_sem);
6227         list_add_tail(&cache->list, &cache->space_info->block_groups);
6228         up_write(&cache->space_info->groups_sem);
6229
6230         ret = btrfs_add_block_group_cache(root->fs_info, cache);
6231         BUG_ON(ret);
6232
6233         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
6234                                 sizeof(cache->item));
6235         BUG_ON(ret);
6236
6237         finish_current_insert(trans, extent_root, 0);
6238         ret = del_pending_extents(trans, extent_root, 0);
6239         BUG_ON(ret);
6240         set_avail_alloc_bits(extent_root->fs_info, type);
6241
6242         return 0;
6243 }
6244
6245 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
6246                              struct btrfs_root *root, u64 group_start)
6247 {
6248         struct btrfs_path *path;
6249         struct btrfs_block_group_cache *block_group;
6250         struct btrfs_key key;
6251         int ret;
6252
6253         root = root->fs_info->extent_root;
6254
6255         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
6256         BUG_ON(!block_group);
6257         BUG_ON(!block_group->ro);
6258
6259         memcpy(&key, &block_group->key, sizeof(key));
6260
6261         path = btrfs_alloc_path();
6262         BUG_ON(!path);
6263
6264         spin_lock(&root->fs_info->block_group_cache_lock);
6265         rb_erase(&block_group->cache_node,
6266                  &root->fs_info->block_group_cache_tree);
6267         spin_unlock(&root->fs_info->block_group_cache_lock);
6268         btrfs_remove_free_space_cache(block_group);
6269         down_write(&block_group->space_info->groups_sem);
6270         list_del(&block_group->list);
6271         up_write(&block_group->space_info->groups_sem);
6272
6273         spin_lock(&block_group->space_info->lock);
6274         block_group->space_info->total_bytes -= block_group->key.offset;
6275         block_group->space_info->bytes_readonly -= block_group->key.offset;
6276         spin_unlock(&block_group->space_info->lock);
6277         block_group->space_info->full = 0;
6278
6279         put_block_group(block_group);
6280         put_block_group(block_group);
6281
6282         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6283         if (ret > 0)
6284                 ret = -EIO;
6285         if (ret < 0)
6286                 goto out;
6287
6288         ret = btrfs_del_item(trans, root, path);
6289 out:
6290         btrfs_free_path(path);
6291         return ret;
6292 }