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                 empty_cluster = 64 * 1024;
2876         }
2877
2878         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2879                 last_ptr = &root->fs_info->last_data_alloc;
2880
2881         if (last_ptr) {
2882                 if (*last_ptr) {
2883                         hint_byte = *last_ptr;
2884                         last_wanted = *last_ptr;
2885                 } else
2886                         empty_size += empty_cluster;
2887         } else {
2888                 empty_cluster = 0;
2889         }
2890         search_start = max(search_start, first_logical_byte(root, 0));
2891         search_start = max(search_start, hint_byte);
2892
2893         if (last_wanted && search_start != last_wanted) {
2894                 last_wanted = 0;
2895                 empty_size += empty_cluster;
2896         }
2897
2898         total_needed += empty_size;
2899         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2900         if (!block_group)
2901                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2902                                                              search_start);
2903         space_info = __find_space_info(root->fs_info, data);
2904
2905         down_read(&space_info->groups_sem);
2906         while (1) {
2907                 struct btrfs_free_space *free_space;
2908                 /*
2909                  * the only way this happens if our hint points to a block
2910                  * group thats not of the proper type, while looping this
2911                  * should never happen
2912                  */
2913                 if (empty_size)
2914                         extra_loop = 1;
2915
2916                 if (!block_group)
2917                         goto new_group_no_lock;
2918
2919                 if (unlikely(!block_group->cached)) {
2920                         mutex_lock(&block_group->cache_mutex);
2921                         ret = cache_block_group(root, block_group);
2922                         mutex_unlock(&block_group->cache_mutex);
2923                         if (ret)
2924                                 break;
2925                 }
2926
2927                 mutex_lock(&block_group->alloc_mutex);
2928                 if (unlikely(!block_group_bits(block_group, data)))
2929                         goto new_group;
2930
2931                 if (unlikely(block_group->ro))
2932                         goto new_group;
2933
2934                 free_space = btrfs_find_free_space(block_group, search_start,
2935                                                    total_needed);
2936                 if (free_space) {
2937                         u64 start = block_group->key.objectid;
2938                         u64 end = block_group->key.objectid +
2939                                 block_group->key.offset;
2940
2941                         search_start = stripe_align(root, free_space->offset);
2942
2943                         /* move on to the next group */
2944                         if (search_start + num_bytes >= search_end)
2945                                 goto new_group;
2946
2947                         /* move on to the next group */
2948                         if (search_start + num_bytes > end)
2949                                 goto new_group;
2950
2951                         if (last_wanted && search_start != last_wanted) {
2952                                 total_needed += empty_cluster;
2953                                 empty_size += empty_cluster;
2954                                 last_wanted = 0;
2955                                 /*
2956                                  * if search_start is still in this block group
2957                                  * then we just re-search this block group
2958                                  */
2959                                 if (search_start >= start &&
2960                                     search_start < end) {
2961                                         mutex_unlock(&block_group->alloc_mutex);
2962                                         continue;
2963                                 }
2964
2965                                 /* else we go to the next block group */
2966                                 goto new_group;
2967                         }
2968
2969                         if (exclude_nr > 0 &&
2970                             (search_start + num_bytes > exclude_start &&
2971                              search_start < exclude_start + exclude_nr)) {
2972                                 search_start = exclude_start + exclude_nr;
2973                                 /*
2974                                  * if search_start is still in this block group
2975                                  * then we just re-search this block group
2976                                  */
2977                                 if (search_start >= start &&
2978                                     search_start < end) {
2979                                         mutex_unlock(&block_group->alloc_mutex);
2980                                         last_wanted = 0;
2981                                         continue;
2982                                 }
2983
2984                                 /* else we go to the next block group */
2985                                 goto new_group;
2986                         }
2987
2988                         ins->objectid = search_start;
2989                         ins->offset = num_bytes;
2990
2991                         btrfs_remove_free_space_lock(block_group, search_start,
2992                                                      num_bytes);
2993                         /* we are all good, lets return */
2994                         mutex_unlock(&block_group->alloc_mutex);
2995                         break;
2996                 }
2997 new_group:
2998                 mutex_unlock(&block_group->alloc_mutex);
2999                 put_block_group(block_group);
3000                 block_group = NULL;
3001 new_group_no_lock:
3002                 /* don't try to compare new allocations against the
3003                  * last allocation any more
3004                  */
3005                 last_wanted = 0;
3006
3007                 /*
3008                  * Here's how this works.
3009                  * loop == 0: we were searching a block group via a hint
3010                  *              and didn't find anything, so we start at
3011                  *              the head of the block groups and keep searching
3012                  * loop == 1: we're searching through all of the block groups
3013                  *              if we hit the head again we have searched
3014                  *              all of the block groups for this space and we
3015                  *              need to try and allocate, if we cant error out.
3016                  * loop == 2: we allocated more space and are looping through
3017                  *              all of the block groups again.
3018                  */
3019                 if (loop == 0) {
3020                         head = &space_info->block_groups;
3021                         cur = head->next;
3022                         loop++;
3023                 } else if (loop == 1 && cur == head) {
3024                         int keep_going;
3025
3026                         /* at this point we give up on the empty_size
3027                          * allocations and just try to allocate the min
3028                          * space.
3029                          *
3030                          * The extra_loop field was set if an empty_size
3031                          * allocation was attempted above, and if this
3032                          * is try we need to try the loop again without
3033                          * the additional empty_size.
3034                          */
3035                         total_needed -= empty_size;
3036                         empty_size = 0;
3037                         keep_going = extra_loop;
3038                         loop++;
3039
3040                         if (allowed_chunk_alloc && !chunk_alloc_done) {
3041                                 up_read(&space_info->groups_sem);
3042                                 ret = do_chunk_alloc(trans, root, num_bytes +
3043                                                      2 * 1024 * 1024, data, 1);
3044                                 down_read(&space_info->groups_sem);
3045                                 if (ret < 0)
3046                                         goto loop_check;
3047                                 head = &space_info->block_groups;
3048                                 /*
3049                                  * we've allocated a new chunk, keep
3050                                  * trying
3051                                  */
3052                                 keep_going = 1;
3053                                 chunk_alloc_done = 1;
3054                         } else if (!allowed_chunk_alloc) {
3055                                 space_info->force_alloc = 1;
3056                         }
3057 loop_check:
3058                         if (keep_going) {
3059                                 cur = head->next;
3060                                 extra_loop = 0;
3061                         } else {
3062                                 break;
3063                         }
3064                 } else if (cur == head) {
3065                         break;
3066                 }
3067
3068                 block_group = list_entry(cur, struct btrfs_block_group_cache,
3069                                          list);
3070                 atomic_inc(&block_group->count);
3071
3072                 search_start = block_group->key.objectid;
3073                 cur = cur->next;
3074         }
3075
3076         /* we found what we needed */
3077         if (ins->objectid) {
3078                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3079                         trans->block_group = block_group->key.objectid;
3080
3081                 if (last_ptr)
3082                         *last_ptr = ins->objectid + ins->offset;
3083                 ret = 0;
3084         } else if (!ret) {
3085                 printk(KERN_ERR "btrfs searching for %llu bytes, "
3086                        "num_bytes %llu, loop %d, allowed_alloc %d\n",
3087                        (unsigned long long)total_needed,
3088                        (unsigned long long)num_bytes,
3089                        loop, allowed_chunk_alloc);
3090                 ret = -ENOSPC;
3091         }
3092         if (block_group)
3093                 put_block_group(block_group);
3094
3095         up_read(&space_info->groups_sem);
3096         return ret;
3097 }
3098
3099 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3100 {
3101         struct btrfs_block_group_cache *cache;
3102
3103         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3104                (unsigned long long)(info->total_bytes - info->bytes_used -
3105                                     info->bytes_pinned - info->bytes_reserved),
3106                (info->full) ? "" : "not ");
3107
3108         down_read(&info->groups_sem);
3109         list_for_each_entry(cache, &info->block_groups, list) {
3110                 spin_lock(&cache->lock);
3111                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3112                        "%llu pinned %llu reserved\n",
3113                        (unsigned long long)cache->key.objectid,
3114                        (unsigned long long)cache->key.offset,
3115                        (unsigned long long)btrfs_block_group_used(&cache->item),
3116                        (unsigned long long)cache->pinned,
3117                        (unsigned long long)cache->reserved);
3118                 btrfs_dump_free_space(cache, bytes);
3119                 spin_unlock(&cache->lock);
3120         }
3121         up_read(&info->groups_sem);
3122 }
3123
3124 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3125                                   struct btrfs_root *root,
3126                                   u64 num_bytes, u64 min_alloc_size,
3127                                   u64 empty_size, u64 hint_byte,
3128                                   u64 search_end, struct btrfs_key *ins,
3129                                   u64 data)
3130 {
3131         int ret;
3132         u64 search_start = 0;
3133         u64 alloc_profile;
3134         struct btrfs_fs_info *info = root->fs_info;
3135
3136         if (data) {
3137                 alloc_profile = info->avail_data_alloc_bits &
3138                         info->data_alloc_profile;
3139                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3140         } else if (root == root->fs_info->chunk_root) {
3141                 alloc_profile = info->avail_system_alloc_bits &
3142                         info->system_alloc_profile;
3143                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3144         } else {
3145                 alloc_profile = info->avail_metadata_alloc_bits &
3146                         info->metadata_alloc_profile;
3147                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3148         }
3149 again:
3150         data = btrfs_reduce_alloc_profile(root, data);
3151         /*
3152          * the only place that sets empty_size is btrfs_realloc_node, which
3153          * is not called recursively on allocations
3154          */
3155         if (empty_size || root->ref_cows) {
3156                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3157                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3158                                      2 * 1024 * 1024,
3159                                      BTRFS_BLOCK_GROUP_METADATA |
3160                                      (info->metadata_alloc_profile &
3161                                       info->avail_metadata_alloc_bits), 0);
3162                 }
3163                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3164                                      num_bytes + 2 * 1024 * 1024, data, 0);
3165         }
3166
3167         WARN_ON(num_bytes < root->sectorsize);
3168         ret = find_free_extent(trans, root, num_bytes, empty_size,
3169                                search_start, search_end, hint_byte, ins,
3170                                trans->alloc_exclude_start,
3171                                trans->alloc_exclude_nr, data);
3172
3173         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3174                 num_bytes = num_bytes >> 1;
3175                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3176                 num_bytes = max(num_bytes, min_alloc_size);
3177                 do_chunk_alloc(trans, root->fs_info->extent_root,
3178                                num_bytes, data, 1);
3179                 goto again;
3180         }
3181         if (ret) {
3182                 struct btrfs_space_info *sinfo;
3183
3184                 sinfo = __find_space_info(root->fs_info, data);
3185                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3186                        "wanted %llu\n", (unsigned long long)data,
3187                        (unsigned long long)num_bytes);
3188                 dump_space_info(sinfo, num_bytes);
3189                 BUG();
3190         }
3191
3192         return ret;
3193 }
3194
3195 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3196 {
3197         struct btrfs_block_group_cache *cache;
3198         int ret = 0;
3199
3200         cache = btrfs_lookup_block_group(root->fs_info, start);
3201         if (!cache) {
3202                 printk(KERN_ERR "Unable to find block group for %llu\n",
3203                        (unsigned long long)start);
3204                 return -ENOSPC;
3205         }
3206
3207         ret = btrfs_discard_extent(root, start, len);
3208
3209         btrfs_add_free_space(cache, start, len);
3210         put_block_group(cache);
3211         update_reserved_extents(root, start, len, 0);
3212
3213         return ret;
3214 }
3215
3216 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3217                                   struct btrfs_root *root,
3218                                   u64 num_bytes, u64 min_alloc_size,
3219                                   u64 empty_size, u64 hint_byte,
3220                                   u64 search_end, struct btrfs_key *ins,
3221                                   u64 data)
3222 {
3223         int ret;
3224         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3225                                      empty_size, hint_byte, search_end, ins,
3226                                      data);
3227         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3228         return ret;
3229 }
3230
3231 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3232                                          struct btrfs_root *root, u64 parent,
3233                                          u64 root_objectid, u64 ref_generation,
3234                                          u64 owner, struct btrfs_key *ins)
3235 {
3236         int ret;
3237         int pending_ret;
3238         u64 super_used;
3239         u64 root_used;
3240         u64 num_bytes = ins->offset;
3241         u32 sizes[2];
3242         struct btrfs_fs_info *info = root->fs_info;
3243         struct btrfs_root *extent_root = info->extent_root;
3244         struct btrfs_extent_item *extent_item;
3245         struct btrfs_extent_ref *ref;
3246         struct btrfs_path *path;
3247         struct btrfs_key keys[2];
3248
3249         if (parent == 0)
3250                 parent = ins->objectid;
3251
3252         /* block accounting for super block */
3253         spin_lock(&info->delalloc_lock);
3254         super_used = btrfs_super_bytes_used(&info->super_copy);
3255         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3256
3257         /* block accounting for root item */
3258         root_used = btrfs_root_used(&root->root_item);
3259         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3260         spin_unlock(&info->delalloc_lock);
3261
3262         if (root == extent_root) {
3263                 struct pending_extent_op *extent_op;
3264
3265                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3266                 BUG_ON(!extent_op);
3267
3268                 extent_op->type = PENDING_EXTENT_INSERT;
3269                 extent_op->bytenr = ins->objectid;
3270                 extent_op->num_bytes = ins->offset;
3271                 extent_op->parent = parent;
3272                 extent_op->orig_parent = 0;
3273                 extent_op->generation = ref_generation;
3274                 extent_op->orig_generation = 0;
3275                 extent_op->level = (int)owner;
3276                 INIT_LIST_HEAD(&extent_op->list);
3277                 extent_op->del = 0;
3278
3279                 mutex_lock(&root->fs_info->extent_ins_mutex);
3280                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3281                                 ins->objectid + ins->offset - 1,
3282                                 EXTENT_WRITEBACK, GFP_NOFS);
3283                 set_state_private(&root->fs_info->extent_ins,
3284                                   ins->objectid, (unsigned long)extent_op);
3285                 mutex_unlock(&root->fs_info->extent_ins_mutex);
3286                 goto update_block;
3287         }
3288
3289         memcpy(&keys[0], ins, sizeof(*ins));
3290         keys[1].objectid = ins->objectid;
3291         keys[1].type = BTRFS_EXTENT_REF_KEY;
3292         keys[1].offset = parent;
3293         sizes[0] = sizeof(*extent_item);
3294         sizes[1] = sizeof(*ref);
3295
3296         path = btrfs_alloc_path();
3297         BUG_ON(!path);
3298
3299         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3300                                        sizes, 2);
3301         BUG_ON(ret);
3302
3303         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3304                                      struct btrfs_extent_item);
3305         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3306         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3307                              struct btrfs_extent_ref);
3308
3309         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3310         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3311         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3312         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3313
3314         btrfs_mark_buffer_dirty(path->nodes[0]);
3315
3316         trans->alloc_exclude_start = 0;
3317         trans->alloc_exclude_nr = 0;
3318         btrfs_free_path(path);
3319         finish_current_insert(trans, extent_root, 0);
3320         pending_ret = del_pending_extents(trans, extent_root, 0);
3321
3322         if (ret)
3323                 goto out;
3324         if (pending_ret) {
3325                 ret = pending_ret;
3326                 goto out;
3327         }
3328
3329 update_block:
3330         ret = update_block_group(trans, root, ins->objectid,
3331                                  ins->offset, 1, 0);
3332         if (ret) {
3333                 printk(KERN_ERR "btrfs update block group failed for %llu "
3334                        "%llu\n", (unsigned long long)ins->objectid,
3335                        (unsigned long long)ins->offset);
3336                 BUG();
3337         }
3338 out:
3339         return ret;
3340 }
3341
3342 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3343                                 struct btrfs_root *root, u64 parent,
3344                                 u64 root_objectid, u64 ref_generation,
3345                                 u64 owner, struct btrfs_key *ins)
3346 {
3347         int ret;
3348
3349         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3350                 return 0;
3351         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3352                                             ref_generation, owner, ins);
3353         update_reserved_extents(root, ins->objectid, ins->offset, 0);
3354         return ret;
3355 }
3356
3357 /*
3358  * this is used by the tree logging recovery code.  It records that
3359  * an extent has been allocated and makes sure to clear the free
3360  * space cache bits as well
3361  */
3362 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3363                                 struct btrfs_root *root, u64 parent,
3364                                 u64 root_objectid, u64 ref_generation,
3365                                 u64 owner, struct btrfs_key *ins)
3366 {
3367         int ret;
3368         struct btrfs_block_group_cache *block_group;
3369
3370         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3371         mutex_lock(&block_group->cache_mutex);
3372         cache_block_group(root, block_group);
3373         mutex_unlock(&block_group->cache_mutex);
3374
3375         ret = btrfs_remove_free_space(block_group, ins->objectid,
3376                                       ins->offset);
3377         BUG_ON(ret);
3378         put_block_group(block_group);
3379         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3380                                             ref_generation, owner, ins);
3381         return ret;
3382 }
3383
3384 /*
3385  * finds a free extent and does all the dirty work required for allocation
3386  * returns the key for the extent through ins, and a tree buffer for
3387  * the first block of the extent through buf.
3388  *
3389  * returns 0 if everything worked, non-zero otherwise.
3390  */
3391 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3392                        struct btrfs_root *root,
3393                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3394                        u64 root_objectid, u64 ref_generation,
3395                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3396                        u64 search_end, struct btrfs_key *ins, u64 data)
3397 {
3398         int ret;
3399
3400         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3401                                      min_alloc_size, empty_size, hint_byte,
3402                                      search_end, ins, data);
3403         BUG_ON(ret);
3404         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3405                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3406                                         root_objectid, ref_generation,
3407                                         owner_objectid, ins);
3408                 BUG_ON(ret);
3409
3410         } else {
3411                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3412         }
3413         return ret;
3414 }
3415
3416 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3417                                             struct btrfs_root *root,
3418                                             u64 bytenr, u32 blocksize)
3419 {
3420         struct extent_buffer *buf;
3421
3422         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3423         if (!buf)
3424                 return ERR_PTR(-ENOMEM);
3425         btrfs_set_header_generation(buf, trans->transid);
3426         btrfs_tree_lock(buf);
3427         clean_tree_block(trans, root, buf);
3428
3429         btrfs_set_lock_blocking(buf);
3430         btrfs_set_buffer_uptodate(buf);
3431
3432         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3433                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3434                          buf->start + buf->len - 1, GFP_NOFS);
3435         } else {
3436                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3437                          buf->start + buf->len - 1, GFP_NOFS);
3438         }
3439         trans->blocks_used++;
3440         /* this returns a buffer locked for blocking */
3441         return buf;
3442 }
3443
3444 /*
3445  * helper function to allocate a block for a given tree
3446  * returns the tree buffer or NULL.
3447  */
3448 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3449                                              struct btrfs_root *root,
3450                                              u32 blocksize, u64 parent,
3451                                              u64 root_objectid,
3452                                              u64 ref_generation,
3453                                              int level,
3454                                              u64 hint,
3455                                              u64 empty_size)
3456 {
3457         struct btrfs_key ins;
3458         int ret;
3459         struct extent_buffer *buf;
3460
3461         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3462                                  root_objectid, ref_generation, level,
3463                                  empty_size, hint, (u64)-1, &ins, 0);
3464         if (ret) {
3465                 BUG_ON(ret > 0);
3466                 return ERR_PTR(ret);
3467         }
3468
3469         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3470         return buf;
3471 }
3472
3473 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3474                         struct btrfs_root *root, struct extent_buffer *leaf)
3475 {
3476         u64 leaf_owner;
3477         u64 leaf_generation;
3478         struct refsort *sorted;
3479         struct btrfs_key key;
3480         struct btrfs_file_extent_item *fi;
3481         int i;
3482         int nritems;
3483         int ret;
3484         int refi = 0;
3485         int slot;
3486
3487         BUG_ON(!btrfs_is_leaf(leaf));
3488         nritems = btrfs_header_nritems(leaf);
3489         leaf_owner = btrfs_header_owner(leaf);
3490         leaf_generation = btrfs_header_generation(leaf);
3491
3492         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3493         /* we do this loop twice.  The first time we build a list
3494          * of the extents we have a reference on, then we sort the list
3495          * by bytenr.  The second time around we actually do the
3496          * extent freeing.
3497          */
3498         for (i = 0; i < nritems; i++) {
3499                 u64 disk_bytenr;
3500                 cond_resched();
3501
3502                 btrfs_item_key_to_cpu(leaf, &key, i);
3503
3504                 /* only extents have references, skip everything else */
3505                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3506                         continue;
3507
3508                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3509
3510                 /* inline extents live in the btree, they don't have refs */
3511                 if (btrfs_file_extent_type(leaf, fi) ==
3512                     BTRFS_FILE_EXTENT_INLINE)
3513                         continue;
3514
3515                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3516
3517                 /* holes don't have refs */
3518                 if (disk_bytenr == 0)
3519                         continue;
3520
3521                 sorted[refi].bytenr = disk_bytenr;
3522                 sorted[refi].slot = i;
3523                 refi++;
3524         }
3525
3526         if (refi == 0)
3527                 goto out;
3528
3529         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3530
3531         for (i = 0; i < refi; i++) {
3532                 u64 disk_bytenr;
3533
3534                 disk_bytenr = sorted[i].bytenr;
3535                 slot = sorted[i].slot;
3536
3537                 cond_resched();
3538
3539                 btrfs_item_key_to_cpu(leaf, &key, slot);
3540                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3541                         continue;
3542
3543                 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3544
3545                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3546                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3547                                 leaf->start, leaf_owner, leaf_generation,
3548                                 key.objectid, 0);
3549                 BUG_ON(ret);
3550
3551                 atomic_inc(&root->fs_info->throttle_gen);
3552                 wake_up(&root->fs_info->transaction_throttle);
3553                 cond_resched();
3554         }
3555 out:
3556         kfree(sorted);
3557         return 0;
3558 }
3559
3560 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3561                                         struct btrfs_root *root,
3562                                         struct btrfs_leaf_ref *ref)
3563 {
3564         int i;
3565         int ret;
3566         struct btrfs_extent_info *info;
3567         struct refsort *sorted;
3568
3569         if (ref->nritems == 0)
3570                 return 0;
3571
3572         sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
3573         for (i = 0; i < ref->nritems; i++) {
3574                 sorted[i].bytenr = ref->extents[i].bytenr;
3575                 sorted[i].slot = i;
3576         }
3577         sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
3578
3579         /*
3580          * the items in the ref were sorted when the ref was inserted
3581          * into the ref cache, so this is already in order
3582          */
3583         for (i = 0; i < ref->nritems; i++) {
3584                 info = ref->extents + sorted[i].slot;
3585                 ret = __btrfs_free_extent(trans, root, info->bytenr,
3586                                           info->num_bytes, ref->bytenr,
3587                                           ref->owner, ref->generation,
3588                                           info->objectid, 0);
3589
3590                 atomic_inc(&root->fs_info->throttle_gen);
3591                 wake_up(&root->fs_info->transaction_throttle);
3592                 cond_resched();
3593
3594                 BUG_ON(ret);
3595                 info++;
3596         }
3597
3598         kfree(sorted);
3599         return 0;
3600 }
3601
3602 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3603                                      u64 len, u32 *refs)
3604 {
3605         int ret;
3606
3607         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3608         BUG_ON(ret);
3609
3610 #if 0 /* some debugging code in case we see problems here */
3611         /* if the refs count is one, it won't get increased again.  But
3612          * if the ref count is > 1, someone may be decreasing it at
3613          * the same time we are.
3614          */
3615         if (*refs != 1) {
3616                 struct extent_buffer *eb = NULL;
3617                 eb = btrfs_find_create_tree_block(root, start, len);
3618                 if (eb)
3619                         btrfs_tree_lock(eb);
3620
3621                 mutex_lock(&root->fs_info->alloc_mutex);
3622                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3623                 BUG_ON(ret);
3624                 mutex_unlock(&root->fs_info->alloc_mutex);
3625
3626                 if (eb) {
3627                         btrfs_tree_unlock(eb);
3628                         free_extent_buffer(eb);
3629                 }
3630                 if (*refs == 1) {
3631                         printk(KERN_ERR "btrfs block %llu went down to one "
3632                                "during drop_snap\n", (unsigned long long)start);
3633                 }
3634
3635         }
3636 #endif
3637
3638         cond_resched();
3639         return ret;
3640 }
3641
3642 /*
3643  * this is used while deleting old snapshots, and it drops the refs
3644  * on a whole subtree starting from a level 1 node.
3645  *
3646  * The idea is to sort all the leaf pointers, and then drop the
3647  * ref on all the leaves in order.  Most of the time the leaves
3648  * will have ref cache entries, so no leaf IOs will be required to
3649  * find the extents they have references on.
3650  *
3651  * For each leaf, any references it has are also dropped in order
3652  *
3653  * This ends up dropping the references in something close to optimal
3654  * order for reading and modifying the extent allocation tree.
3655  */
3656 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3657                                         struct btrfs_root *root,
3658                                         struct btrfs_path *path)
3659 {
3660         u64 bytenr;
3661         u64 root_owner;
3662         u64 root_gen;
3663         struct extent_buffer *eb = path->nodes[1];
3664         struct extent_buffer *leaf;
3665         struct btrfs_leaf_ref *ref;
3666         struct refsort *sorted = NULL;
3667         int nritems = btrfs_header_nritems(eb);
3668         int ret;
3669         int i;
3670         int refi = 0;
3671         int slot = path->slots[1];
3672         u32 blocksize = btrfs_level_size(root, 0);
3673         u32 refs;
3674
3675         if (nritems == 0)
3676                 goto out;
3677
3678         root_owner = btrfs_header_owner(eb);
3679         root_gen = btrfs_header_generation(eb);
3680         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3681
3682         /*
3683          * step one, sort all the leaf pointers so we don't scribble
3684          * randomly into the extent allocation tree
3685          */
3686         for (i = slot; i < nritems; i++) {
3687                 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
3688                 sorted[refi].slot = i;
3689                 refi++;
3690         }
3691
3692         /*
3693          * nritems won't be zero, but if we're picking up drop_snapshot
3694          * after a crash, slot might be > 0, so double check things
3695          * just in case.
3696          */
3697         if (refi == 0)
3698                 goto out;
3699
3700         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3701
3702         /*
3703          * the first loop frees everything the leaves point to
3704          */
3705         for (i = 0; i < refi; i++) {
3706                 u64 ptr_gen;
3707
3708                 bytenr = sorted[i].bytenr;
3709
3710                 /*
3711                  * check the reference count on this leaf.  If it is > 1
3712                  * we just decrement it below and don't update any
3713                  * of the refs the leaf points to.
3714                  */
3715                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3716                 BUG_ON(ret);
3717                 if (refs != 1)
3718                         continue;
3719
3720                 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
3721
3722                 /*
3723                  * the leaf only had one reference, which means the
3724                  * only thing pointing to this leaf is the snapshot
3725                  * we're deleting.  It isn't possible for the reference
3726                  * count to increase again later
3727                  *
3728                  * The reference cache is checked for the leaf,
3729                  * and if found we'll be able to drop any refs held by
3730                  * the leaf without needing to read it in.
3731                  */
3732                 ref = btrfs_lookup_leaf_ref(root, bytenr);
3733                 if (ref && ref->generation != ptr_gen) {
3734                         btrfs_free_leaf_ref(root, ref);
3735                         ref = NULL;
3736                 }
3737                 if (ref) {
3738                         ret = cache_drop_leaf_ref(trans, root, ref);
3739                         BUG_ON(ret);
3740                         btrfs_remove_leaf_ref(root, ref);
3741                         btrfs_free_leaf_ref(root, ref);
3742                 } else {
3743                         /*
3744                          * the leaf wasn't in the reference cache, so
3745                          * we have to read it.
3746                          */
3747                         leaf = read_tree_block(root, bytenr, blocksize,
3748                                                ptr_gen);
3749                         ret = btrfs_drop_leaf_ref(trans, root, leaf);
3750                         BUG_ON(ret);
3751                         free_extent_buffer(leaf);
3752                 }
3753                 atomic_inc(&root->fs_info->throttle_gen);
3754                 wake_up(&root->fs_info->transaction_throttle);
3755                 cond_resched();
3756         }
3757
3758         /*
3759          * run through the loop again to free the refs on the leaves.
3760          * This is faster than doing it in the loop above because
3761          * the leaves are likely to be clustered together.  We end up
3762          * working in nice chunks on the extent allocation tree.
3763          */
3764         for (i = 0; i < refi; i++) {
3765                 bytenr = sorted[i].bytenr;
3766                 ret = __btrfs_free_extent(trans, root, bytenr,
3767                                         blocksize, eb->start,
3768                                         root_owner, root_gen, 0, 1);
3769                 BUG_ON(ret);
3770
3771                 atomic_inc(&root->fs_info->throttle_gen);
3772                 wake_up(&root->fs_info->transaction_throttle);
3773                 cond_resched();
3774         }
3775 out:
3776         kfree(sorted);
3777
3778         /*
3779          * update the path to show we've processed the entire level 1
3780          * node.  This will get saved into the root's drop_snapshot_progress
3781          * field so these drops are not repeated again if this transaction
3782          * commits.
3783          */
3784         path->slots[1] = nritems;
3785         return 0;
3786 }
3787
3788 /*
3789  * helper function for drop_snapshot, this walks down the tree dropping ref
3790  * counts as it goes.
3791  */
3792 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3793                                    struct btrfs_root *root,
3794                                    struct btrfs_path *path, int *level)
3795 {
3796         u64 root_owner;
3797         u64 root_gen;
3798         u64 bytenr;
3799         u64 ptr_gen;
3800         struct extent_buffer *next;
3801         struct extent_buffer *cur;
3802         struct extent_buffer *parent;
3803         u32 blocksize;
3804         int ret;
3805         u32 refs;
3806
3807         WARN_ON(*level < 0);
3808         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3809         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3810                                 path->nodes[*level]->len, &refs);
3811         BUG_ON(ret);
3812         if (refs > 1)
3813                 goto out;
3814
3815         /*
3816          * walk down to the last node level and free all the leaves
3817          */
3818         while (*level >= 0) {
3819                 WARN_ON(*level < 0);
3820                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3821                 cur = path->nodes[*level];
3822
3823                 if (btrfs_header_level(cur) != *level)
3824                         WARN_ON(1);
3825
3826                 if (path->slots[*level] >=
3827                     btrfs_header_nritems(cur))
3828                         break;
3829
3830                 /* the new code goes down to level 1 and does all the
3831                  * leaves pointed to that node in bulk.  So, this check
3832                  * for level 0 will always be false.
3833                  *
3834                  * But, the disk format allows the drop_snapshot_progress
3835                  * field in the root to leave things in a state where
3836                  * a leaf will need cleaning up here.  If someone crashes
3837                  * with the old code and then boots with the new code,
3838                  * we might find a leaf here.
3839                  */
3840                 if (*level == 0) {
3841                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3842                         BUG_ON(ret);
3843                         break;
3844                 }
3845
3846                 /*
3847                  * once we get to level one, process the whole node
3848                  * at once, including everything below it.
3849                  */
3850                 if (*level == 1) {
3851                         ret = drop_level_one_refs(trans, root, path);
3852                         BUG_ON(ret);
3853                         break;
3854                 }
3855
3856                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3857                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3858                 blocksize = btrfs_level_size(root, *level - 1);
3859
3860                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3861                 BUG_ON(ret);
3862
3863                 /*
3864                  * if there is more than one reference, we don't need
3865                  * to read that node to drop any references it has.  We
3866                  * just drop the ref we hold on that node and move on to the
3867                  * next slot in this level.
3868                  */
3869                 if (refs != 1) {
3870                         parent = path->nodes[*level];
3871                         root_owner = btrfs_header_owner(parent);
3872                         root_gen = btrfs_header_generation(parent);
3873                         path->slots[*level]++;
3874
3875                         ret = __btrfs_free_extent(trans, root, bytenr,
3876                                                 blocksize, parent->start,
3877                                                 root_owner, root_gen,
3878                                                 *level - 1, 1);
3879                         BUG_ON(ret);
3880
3881                         atomic_inc(&root->fs_info->throttle_gen);
3882                         wake_up(&root->fs_info->transaction_throttle);
3883                         cond_resched();
3884
3885                         continue;
3886                 }
3887
3888                 /*
3889                  * we need to keep freeing things in the next level down.
3890                  * read the block and loop around to process it
3891                  */
3892                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3893                 WARN_ON(*level <= 0);
3894                 if (path->nodes[*level-1])
3895                         free_extent_buffer(path->nodes[*level-1]);
3896                 path->nodes[*level-1] = next;
3897                 *level = btrfs_header_level(next);
3898                 path->slots[*level] = 0;
3899                 cond_resched();
3900         }
3901 out:
3902         WARN_ON(*level < 0);
3903         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3904
3905         if (path->nodes[*level] == root->node) {
3906                 parent = path->nodes[*level];
3907                 bytenr = path->nodes[*level]->start;
3908         } else {
3909                 parent = path->nodes[*level + 1];
3910                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3911         }
3912
3913         blocksize = btrfs_level_size(root, *level);
3914         root_owner = btrfs_header_owner(parent);
3915         root_gen = btrfs_header_generation(parent);
3916
3917         /*
3918          * cleanup and free the reference on the last node
3919          * we processed
3920          */
3921         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3922                                   parent->start, root_owner, root_gen,
3923                                   *level, 1);
3924         free_extent_buffer(path->nodes[*level]);
3925         path->nodes[*level] = NULL;
3926
3927         *level += 1;
3928         BUG_ON(ret);
3929
3930         cond_resched();
3931         return 0;
3932 }
3933
3934 /*
3935  * helper function for drop_subtree, this function is similar to
3936  * walk_down_tree. The main difference is that it checks reference
3937  * counts while tree blocks are locked.
3938  */
3939 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3940                                       struct btrfs_root *root,
3941                                       struct btrfs_path *path, int *level)
3942 {
3943         struct extent_buffer *next;
3944         struct extent_buffer *cur;
3945         struct extent_buffer *parent;
3946         u64 bytenr;
3947         u64 ptr_gen;
3948         u32 blocksize;
3949         u32 refs;
3950         int ret;
3951
3952         cur = path->nodes[*level];
3953         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3954                                       &refs);
3955         BUG_ON(ret);
3956         if (refs > 1)
3957                 goto out;
3958
3959         while (*level >= 0) {
3960                 cur = path->nodes[*level];
3961                 if (*level == 0) {
3962                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3963                         BUG_ON(ret);
3964                         clean_tree_block(trans, root, cur);
3965                         break;
3966                 }
3967                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3968                         clean_tree_block(trans, root, cur);
3969                         break;
3970                 }
3971
3972                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3973                 blocksize = btrfs_level_size(root, *level - 1);
3974                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3975
3976                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3977                 btrfs_tree_lock(next);
3978                 btrfs_set_lock_blocking(next);
3979
3980                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3981                                               &refs);
3982                 BUG_ON(ret);
3983                 if (refs > 1) {
3984                         parent = path->nodes[*level];
3985                         ret = btrfs_free_extent(trans, root, bytenr,
3986                                         blocksize, parent->start,
3987                                         btrfs_header_owner(parent),
3988                                         btrfs_header_generation(parent),
3989                                         *level - 1, 1);
3990                         BUG_ON(ret);
3991                         path->slots[*level]++;
3992                         btrfs_tree_unlock(next);
3993                         free_extent_buffer(next);
3994                         continue;
3995                 }
3996
3997                 *level = btrfs_header_level(next);
3998                 path->nodes[*level] = next;
3999                 path->slots[*level] = 0;
4000                 path->locks[*level] = 1;
4001                 cond_resched();
4002         }
4003 out:
4004         parent = path->nodes[*level + 1];
4005         bytenr = path->nodes[*level]->start;
4006         blocksize = path->nodes[*level]->len;
4007
4008         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4009                         parent->start, btrfs_header_owner(parent),
4010                         btrfs_header_generation(parent), *level, 1);
4011         BUG_ON(ret);
4012
4013         if (path->locks[*level]) {
4014                 btrfs_tree_unlock(path->nodes[*level]);
4015                 path->locks[*level] = 0;
4016         }
4017         free_extent_buffer(path->nodes[*level]);
4018         path->nodes[*level] = NULL;
4019         *level += 1;
4020         cond_resched();
4021         return 0;
4022 }
4023
4024 /*
4025  * helper for dropping snapshots.  This walks back up the tree in the path
4026  * to find the first node higher up where we haven't yet gone through
4027  * all the slots
4028  */
4029 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
4030                                  struct btrfs_root *root,
4031                                  struct btrfs_path *path,
4032                                  int *level, int max_level)
4033 {
4034         u64 root_owner;
4035         u64 root_gen;
4036         struct btrfs_root_item *root_item = &root->root_item;
4037         int i;
4038         int slot;
4039         int ret;
4040
4041         for (i = *level; i < max_level && path->nodes[i]; i++) {
4042                 slot = path->slots[i];
4043                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
4044                         struct extent_buffer *node;
4045                         struct btrfs_disk_key disk_key;
4046
4047                         /*
4048                          * there is more work to do in this level.
4049                          * Update the drop_progress marker to reflect
4050                          * the work we've done so far, and then bump
4051                          * the slot number
4052                          */
4053                         node = path->nodes[i];
4054                         path->slots[i]++;
4055                         *level = i;
4056                         WARN_ON(*level == 0);
4057                         btrfs_node_key(node, &disk_key, path->slots[i]);
4058                         memcpy(&root_item->drop_progress,
4059                                &disk_key, sizeof(disk_key));
4060                         root_item->drop_level = i;
4061                         return 0;
4062                 } else {
4063                         struct extent_buffer *parent;
4064
4065                         /*
4066                          * this whole node is done, free our reference
4067                          * on it and go up one level
4068                          */
4069                         if (path->nodes[*level] == root->node)
4070                                 parent = path->nodes[*level];
4071                         else
4072                                 parent = path->nodes[*level + 1];
4073
4074                         root_owner = btrfs_header_owner(parent);
4075                         root_gen = btrfs_header_generation(parent);
4076
4077                         clean_tree_block(trans, root, path->nodes[*level]);
4078                         ret = btrfs_free_extent(trans, root,
4079                                                 path->nodes[*level]->start,
4080                                                 path->nodes[*level]->len,
4081                                                 parent->start, root_owner,
4082                                                 root_gen, *level, 1);
4083                         BUG_ON(ret);
4084                         if (path->locks[*level]) {
4085                                 btrfs_tree_unlock(path->nodes[*level]);
4086                                 path->locks[*level] = 0;
4087                         }
4088                         free_extent_buffer(path->nodes[*level]);
4089                         path->nodes[*level] = NULL;
4090                         *level = i + 1;
4091                 }
4092         }
4093         return 1;
4094 }
4095
4096 /*
4097  * drop the reference count on the tree rooted at 'snap'.  This traverses
4098  * the tree freeing any blocks that have a ref count of zero after being
4099  * decremented.
4100  */
4101 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
4102                         *root)
4103 {
4104         int ret = 0;
4105         int wret;
4106         int level;
4107         struct btrfs_path *path;
4108         int i;
4109         int orig_level;
4110         struct btrfs_root_item *root_item = &root->root_item;
4111
4112         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
4113         path = btrfs_alloc_path();
4114         BUG_ON(!path);
4115
4116         level = btrfs_header_level(root->node);
4117         orig_level = level;
4118         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
4119                 path->nodes[level] = root->node;
4120                 extent_buffer_get(root->node);
4121                 path->slots[level] = 0;
4122         } else {
4123                 struct btrfs_key key;
4124                 struct btrfs_disk_key found_key;
4125                 struct extent_buffer *node;
4126
4127                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
4128                 level = root_item->drop_level;
4129                 path->lowest_level = level;
4130                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4131                 if (wret < 0) {
4132                         ret = wret;
4133                         goto out;
4134                 }
4135                 node = path->nodes[level];
4136                 btrfs_node_key(node, &found_key, path->slots[level]);
4137                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
4138                                sizeof(found_key)));
4139                 /*
4140                  * unlock our path, this is safe because only this
4141                  * function is allowed to delete this snapshot
4142                  */
4143                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4144                         if (path->nodes[i] && path->locks[i]) {
4145                                 path->locks[i] = 0;
4146                                 btrfs_tree_unlock(path->nodes[i]);
4147                         }
4148                 }
4149         }
4150         while (1) {
4151                 wret = walk_down_tree(trans, root, path, &level);
4152                 if (wret > 0)
4153                         break;
4154                 if (wret < 0)
4155                         ret = wret;
4156
4157                 wret = walk_up_tree(trans, root, path, &level,
4158                                     BTRFS_MAX_LEVEL);
4159                 if (wret > 0)
4160                         break;
4161                 if (wret < 0)
4162                         ret = wret;
4163                 if (trans->transaction->in_commit) {
4164                         ret = -EAGAIN;
4165                         break;
4166                 }
4167                 atomic_inc(&root->fs_info->throttle_gen);
4168                 wake_up(&root->fs_info->transaction_throttle);
4169         }
4170         for (i = 0; i <= orig_level; i++) {
4171                 if (path->nodes[i]) {
4172                         free_extent_buffer(path->nodes[i]);
4173                         path->nodes[i] = NULL;
4174                 }
4175         }
4176 out:
4177         btrfs_free_path(path);
4178         return ret;
4179 }
4180
4181 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
4182                         struct btrfs_root *root,
4183                         struct extent_buffer *node,
4184                         struct extent_buffer *parent)
4185 {
4186         struct btrfs_path *path;
4187         int level;
4188         int parent_level;
4189         int ret = 0;
4190         int wret;
4191
4192         path = btrfs_alloc_path();
4193         BUG_ON(!path);
4194
4195         BUG_ON(!btrfs_tree_locked(parent));
4196         parent_level = btrfs_header_level(parent);
4197         extent_buffer_get(parent);
4198         path->nodes[parent_level] = parent;
4199         path->slots[parent_level] = btrfs_header_nritems(parent);
4200
4201         BUG_ON(!btrfs_tree_locked(node));
4202         level = btrfs_header_level(node);
4203         extent_buffer_get(node);
4204         path->nodes[level] = node;
4205         path->slots[level] = 0;
4206
4207         while (1) {
4208                 wret = walk_down_subtree(trans, root, path, &level);
4209                 if (wret < 0)
4210                         ret = wret;
4211                 if (wret != 0)
4212                         break;
4213
4214                 wret = walk_up_tree(trans, root, path, &level, parent_level);
4215                 if (wret < 0)
4216                         ret = wret;
4217                 if (wret != 0)
4218                         break;
4219         }
4220
4221         btrfs_free_path(path);
4222         return ret;
4223 }
4224
4225 static unsigned long calc_ra(unsigned long start, unsigned long last,
4226                              unsigned long nr)
4227 {
4228         return min(last, start + nr - 1);
4229 }
4230
4231 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
4232                                          u64 len)
4233 {
4234         u64 page_start;
4235         u64 page_end;
4236         unsigned long first_index;
4237         unsigned long last_index;
4238         unsigned long i;
4239         struct page *page;
4240         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4241         struct file_ra_state *ra;
4242         struct btrfs_ordered_extent *ordered;
4243         unsigned int total_read = 0;
4244         unsigned int total_dirty = 0;
4245         int ret = 0;
4246
4247         ra = kzalloc(sizeof(*ra), GFP_NOFS);
4248
4249         mutex_lock(&inode->i_mutex);
4250         first_index = start >> PAGE_CACHE_SHIFT;
4251         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
4252
4253         /* make sure the dirty trick played by the caller work */
4254         ret = invalidate_inode_pages2_range(inode->i_mapping,
4255                                             first_index, last_index);
4256         if (ret)
4257                 goto out_unlock;
4258
4259         file_ra_state_init(ra, inode->i_mapping);
4260
4261         for (i = first_index ; i <= last_index; i++) {
4262                 if (total_read % ra->ra_pages == 0) {
4263                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
4264                                        calc_ra(i, last_index, ra->ra_pages));
4265                 }
4266                 total_read++;
4267 again:
4268                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
4269                         BUG_ON(1);
4270                 page = grab_cache_page(inode->i_mapping, i);
4271                 if (!page) {
4272                         ret = -ENOMEM;
4273                         goto out_unlock;
4274                 }
4275                 if (!PageUptodate(page)) {
4276                         btrfs_readpage(NULL, page);
4277                         lock_page(page);
4278                         if (!PageUptodate(page)) {
4279                                 unlock_page(page);
4280                                 page_cache_release(page);
4281                                 ret = -EIO;
4282                                 goto out_unlock;
4283                         }
4284                 }
4285                 wait_on_page_writeback(page);
4286
4287                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
4288                 page_end = page_start + PAGE_CACHE_SIZE - 1;
4289                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4290
4291                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4292                 if (ordered) {
4293                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4294                         unlock_page(page);
4295                         page_cache_release(page);
4296                         btrfs_start_ordered_extent(inode, ordered, 1);
4297                         btrfs_put_ordered_extent(ordered);
4298                         goto again;
4299                 }
4300                 set_page_extent_mapped(page);
4301
4302                 if (i == first_index)
4303                         set_extent_bits(io_tree, page_start, page_end,
4304                                         EXTENT_BOUNDARY, GFP_NOFS);
4305                 btrfs_set_extent_delalloc(inode, page_start, page_end);
4306
4307                 set_page_dirty(page);
4308                 total_dirty++;
4309
4310                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4311                 unlock_page(page);
4312                 page_cache_release(page);
4313         }
4314
4315 out_unlock:
4316         kfree(ra);
4317         mutex_unlock(&inode->i_mutex);
4318         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4319         return ret;
4320 }
4321
4322 static noinline int relocate_data_extent(struct inode *reloc_inode,
4323                                          struct btrfs_key *extent_key,
4324                                          u64 offset)
4325 {
4326         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4327         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4328         struct extent_map *em;
4329         u64 start = extent_key->objectid - offset;
4330         u64 end = start + extent_key->offset - 1;
4331
4332         em = alloc_extent_map(GFP_NOFS);
4333         BUG_ON(!em || IS_ERR(em));
4334
4335         em->start = start;
4336         em->len = extent_key->offset;
4337         em->block_len = extent_key->offset;
4338         em->block_start = extent_key->objectid;
4339         em->bdev = root->fs_info->fs_devices->latest_bdev;
4340         set_bit(EXTENT_FLAG_PINNED, &em->flags);
4341
4342         /* setup extent map to cheat btrfs_readpage */
4343         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4344         while (1) {
4345                 int ret;
4346                 spin_lock(&em_tree->lock);
4347                 ret = add_extent_mapping(em_tree, em);
4348                 spin_unlock(&em_tree->lock);
4349                 if (ret != -EEXIST) {
4350                         free_extent_map(em);
4351                         break;
4352                 }
4353                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4354         }
4355         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4356
4357         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4358 }
4359
4360 struct btrfs_ref_path {
4361         u64 extent_start;
4362         u64 nodes[BTRFS_MAX_LEVEL];
4363         u64 root_objectid;
4364         u64 root_generation;
4365         u64 owner_objectid;
4366         u32 num_refs;
4367         int lowest_level;
4368         int current_level;
4369         int shared_level;
4370
4371         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4372         u64 new_nodes[BTRFS_MAX_LEVEL];
4373 };
4374
4375 struct disk_extent {
4376         u64 ram_bytes;
4377         u64 disk_bytenr;
4378         u64 disk_num_bytes;
4379         u64 offset;
4380         u64 num_bytes;
4381         u8 compression;
4382         u8 encryption;
4383         u16 other_encoding;
4384 };
4385
4386 static int is_cowonly_root(u64 root_objectid)
4387 {
4388         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4389             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4390             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4391             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4392             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4393             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4394                 return 1;
4395         return 0;
4396 }
4397
4398 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4399                                     struct btrfs_root *extent_root,
4400                                     struct btrfs_ref_path *ref_path,
4401                                     int first_time)
4402 {
4403         struct extent_buffer *leaf;
4404         struct btrfs_path *path;
4405         struct btrfs_extent_ref *ref;
4406         struct btrfs_key key;
4407         struct btrfs_key found_key;
4408         u64 bytenr;
4409         u32 nritems;
4410         int level;
4411         int ret = 1;
4412
4413         path = btrfs_alloc_path();
4414         if (!path)
4415                 return -ENOMEM;
4416
4417         if (first_time) {
4418                 ref_path->lowest_level = -1;
4419                 ref_path->current_level = -1;
4420                 ref_path->shared_level = -1;
4421                 goto walk_up;
4422         }
4423 walk_down:
4424         level = ref_path->current_level - 1;
4425         while (level >= -1) {
4426                 u64 parent;
4427                 if (level < ref_path->lowest_level)
4428                         break;
4429
4430                 if (level >= 0)
4431                         bytenr = ref_path->nodes[level];
4432                 else
4433                         bytenr = ref_path->extent_start;
4434                 BUG_ON(bytenr == 0);
4435
4436                 parent = ref_path->nodes[level + 1];
4437                 ref_path->nodes[level + 1] = 0;
4438                 ref_path->current_level = level;
4439                 BUG_ON(parent == 0);
4440
4441                 key.objectid = bytenr;
4442                 key.offset = parent + 1;
4443                 key.type = BTRFS_EXTENT_REF_KEY;
4444
4445                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4446                 if (ret < 0)
4447                         goto out;
4448                 BUG_ON(ret == 0);
4449
4450                 leaf = path->nodes[0];
4451                 nritems = btrfs_header_nritems(leaf);
4452                 if (path->slots[0] >= nritems) {
4453                         ret = btrfs_next_leaf(extent_root, path);
4454                         if (ret < 0)
4455                                 goto out;
4456                         if (ret > 0)
4457                                 goto next;
4458                         leaf = path->nodes[0];
4459                 }
4460
4461                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4462                 if (found_key.objectid == bytenr &&
4463                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4464                         if (level < ref_path->shared_level)
4465                                 ref_path->shared_level = level;
4466                         goto found;
4467                 }
4468 next:
4469                 level--;
4470                 btrfs_release_path(extent_root, path);
4471                 cond_resched();
4472         }
4473         /* reached lowest level */
4474         ret = 1;
4475         goto out;
4476 walk_up:
4477         level = ref_path->current_level;
4478         while (level < BTRFS_MAX_LEVEL - 1) {
4479                 u64 ref_objectid;
4480
4481                 if (level >= 0)
4482                         bytenr = ref_path->nodes[level];
4483                 else
4484                         bytenr = ref_path->extent_start;
4485
4486                 BUG_ON(bytenr == 0);
4487
4488                 key.objectid = bytenr;
4489                 key.offset = 0;
4490                 key.type = BTRFS_EXTENT_REF_KEY;
4491
4492                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4493                 if (ret < 0)
4494                         goto out;
4495
4496                 leaf = path->nodes[0];
4497                 nritems = btrfs_header_nritems(leaf);
4498                 if (path->slots[0] >= nritems) {
4499                         ret = btrfs_next_leaf(extent_root, path);
4500                         if (ret < 0)
4501                                 goto out;
4502                         if (ret > 0) {
4503                                 /* the extent was freed by someone */
4504                                 if (ref_path->lowest_level == level)
4505                                         goto out;
4506                                 btrfs_release_path(extent_root, path);
4507                                 goto walk_down;
4508                         }
4509                         leaf = path->nodes[0];
4510                 }
4511
4512                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4513                 if (found_key.objectid != bytenr ||
4514                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4515                         /* the extent was freed by someone */
4516                         if (ref_path->lowest_level == level) {
4517                                 ret = 1;
4518                                 goto out;
4519                         }
4520                         btrfs_release_path(extent_root, path);
4521                         goto walk_down;
4522                 }
4523 found:
4524                 ref = btrfs_item_ptr(leaf, path->slots[0],
4525                                 struct btrfs_extent_ref);
4526                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4527                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4528                         if (first_time) {
4529                                 level = (int)ref_objectid;
4530                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4531                                 ref_path->lowest_level = level;
4532                                 ref_path->current_level = level;
4533                                 ref_path->nodes[level] = bytenr;
4534                         } else {
4535                                 WARN_ON(ref_objectid != level);
4536                         }
4537                 } else {
4538                         WARN_ON(level != -1);
4539                 }
4540                 first_time = 0;
4541
4542                 if (ref_path->lowest_level == level) {
4543                         ref_path->owner_objectid = ref_objectid;
4544                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4545                 }
4546
4547                 /*
4548                  * the block is tree root or the block isn't in reference
4549                  * counted tree.
4550                  */
4551                 if (found_key.objectid == found_key.offset ||
4552                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4553                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4554                         ref_path->root_generation =
4555                                 btrfs_ref_generation(leaf, ref);
4556                         if (level < 0) {
4557                                 /* special reference from the tree log */
4558                                 ref_path->nodes[0] = found_key.offset;
4559                                 ref_path->current_level = 0;
4560                         }
4561                         ret = 0;
4562                         goto out;
4563                 }
4564
4565                 level++;
4566                 BUG_ON(ref_path->nodes[level] != 0);
4567                 ref_path->nodes[level] = found_key.offset;
4568                 ref_path->current_level = level;
4569
4570                 /*
4571                  * the reference was created in the running transaction,
4572                  * no need to continue walking up.
4573                  */
4574                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4575                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4576                         ref_path->root_generation =
4577                                 btrfs_ref_generation(leaf, ref);
4578                         ret = 0;
4579                         goto out;
4580                 }
4581
4582                 btrfs_release_path(extent_root, path);
4583                 cond_resched();
4584         }
4585         /* reached max tree level, but no tree root found. */
4586         BUG();
4587 out:
4588         btrfs_free_path(path);
4589         return ret;
4590 }
4591
4592 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4593                                 struct btrfs_root *extent_root,
4594                                 struct btrfs_ref_path *ref_path,
4595                                 u64 extent_start)
4596 {
4597         memset(ref_path, 0, sizeof(*ref_path));
4598         ref_path->extent_start = extent_start;
4599
4600         return __next_ref_path(trans, extent_root, ref_path, 1);
4601 }
4602
4603 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4604                                struct btrfs_root *extent_root,
4605                                struct btrfs_ref_path *ref_path)
4606 {
4607         return __next_ref_path(trans, extent_root, ref_path, 0);
4608 }
4609
4610 static noinline int get_new_locations(struct inode *reloc_inode,
4611                                       struct btrfs_key *extent_key,
4612                                       u64 offset, int no_fragment,
4613                                       struct disk_extent **extents,
4614                                       int *nr_extents)
4615 {
4616         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4617         struct btrfs_path *path;
4618         struct btrfs_file_extent_item *fi;
4619         struct extent_buffer *leaf;
4620         struct disk_extent *exts = *extents;
4621         struct btrfs_key found_key;
4622         u64 cur_pos;
4623         u64 last_byte;
4624         u32 nritems;
4625         int nr = 0;
4626         int max = *nr_extents;
4627         int ret;
4628
4629         WARN_ON(!no_fragment && *extents);
4630         if (!exts) {
4631                 max = 1;
4632                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4633                 if (!exts)
4634                         return -ENOMEM;
4635         }
4636
4637         path = btrfs_alloc_path();
4638         BUG_ON(!path);
4639
4640         cur_pos = extent_key->objectid - offset;
4641         last_byte = extent_key->objectid + extent_key->offset;
4642         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4643                                        cur_pos, 0);
4644         if (ret < 0)
4645                 goto out;
4646         if (ret > 0) {
4647                 ret = -ENOENT;
4648                 goto out;
4649         }
4650
4651         while (1) {
4652                 leaf = path->nodes[0];
4653                 nritems = btrfs_header_nritems(leaf);
4654                 if (path->slots[0] >= nritems) {
4655                         ret = btrfs_next_leaf(root, path);
4656                         if (ret < 0)
4657                                 goto out;
4658                         if (ret > 0)
4659                                 break;
4660                         leaf = path->nodes[0];
4661                 }
4662
4663                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4664                 if (found_key.offset != cur_pos ||
4665                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4666                     found_key.objectid != reloc_inode->i_ino)
4667                         break;
4668
4669                 fi = btrfs_item_ptr(leaf, path->slots[0],
4670                                     struct btrfs_file_extent_item);
4671                 if (btrfs_file_extent_type(leaf, fi) !=
4672                     BTRFS_FILE_EXTENT_REG ||
4673                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4674                         break;
4675
4676                 if (nr == max) {
4677                         struct disk_extent *old = exts;
4678                         max *= 2;
4679                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4680                         memcpy(exts, old, sizeof(*exts) * nr);
4681                         if (old != *extents)
4682                                 kfree(old);
4683                 }
4684
4685                 exts[nr].disk_bytenr =
4686                         btrfs_file_extent_disk_bytenr(leaf, fi);
4687                 exts[nr].disk_num_bytes =
4688                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4689                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4690                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4691                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4692                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4693                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4694                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4695                                                                            fi);
4696                 BUG_ON(exts[nr].offset > 0);
4697                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4698                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4699
4700                 cur_pos += exts[nr].num_bytes;
4701                 nr++;
4702
4703                 if (cur_pos + offset >= last_byte)
4704                         break;
4705
4706                 if (no_fragment) {
4707                         ret = 1;
4708                         goto out;
4709                 }
4710                 path->slots[0]++;
4711         }
4712
4713         BUG_ON(cur_pos + offset > last_byte);
4714         if (cur_pos + offset < last_byte) {
4715                 ret = -ENOENT;
4716                 goto out;
4717         }
4718         ret = 0;
4719 out:
4720         btrfs_free_path(path);
4721         if (ret) {
4722                 if (exts != *extents)
4723                         kfree(exts);
4724         } else {
4725                 *extents = exts;
4726                 *nr_extents = nr;
4727         }
4728         return ret;
4729 }
4730
4731 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4732                                         struct btrfs_root *root,
4733                                         struct btrfs_path *path,
4734                                         struct btrfs_key *extent_key,
4735                                         struct btrfs_key *leaf_key,
4736                                         struct btrfs_ref_path *ref_path,
4737                                         struct disk_extent *new_extents,
4738                                         int nr_extents)
4739 {
4740         struct extent_buffer *leaf;
4741         struct btrfs_file_extent_item *fi;
4742         struct inode *inode = NULL;
4743         struct btrfs_key key;
4744         u64 lock_start = 0;
4745         u64 lock_end = 0;
4746         u64 num_bytes;
4747         u64 ext_offset;
4748         u64 search_end = (u64)-1;
4749         u32 nritems;
4750         int nr_scaned = 0;
4751         int extent_locked = 0;
4752         int extent_type;
4753         int ret;
4754
4755         memcpy(&key, leaf_key, sizeof(key));
4756         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4757                 if (key.objectid < ref_path->owner_objectid ||
4758                     (key.objectid == ref_path->owner_objectid &&
4759                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4760                         key.objectid = ref_path->owner_objectid;
4761                         key.type = BTRFS_EXTENT_DATA_KEY;
4762                         key.offset = 0;
4763                 }
4764         }
4765
4766         while (1) {
4767                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4768                 if (ret < 0)
4769                         goto out;
4770
4771                 leaf = path->nodes[0];
4772                 nritems = btrfs_header_nritems(leaf);
4773 next:
4774                 if (extent_locked && ret > 0) {
4775                         /*
4776                          * the file extent item was modified by someone
4777                          * before the extent got locked.
4778                          */
4779                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4780                                       lock_end, GFP_NOFS);
4781                         extent_locked = 0;
4782                 }
4783
4784                 if (path->slots[0] >= nritems) {
4785                         if (++nr_scaned > 2)
4786                                 break;
4787
4788                         BUG_ON(extent_locked);
4789                         ret = btrfs_next_leaf(root, path);
4790                         if (ret < 0)
4791                                 goto out;
4792                         if (ret > 0)
4793                                 break;
4794                         leaf = path->nodes[0];
4795                         nritems = btrfs_header_nritems(leaf);
4796                 }
4797
4798                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4799
4800                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4801                         if ((key.objectid > ref_path->owner_objectid) ||
4802                             (key.objectid == ref_path->owner_objectid &&
4803                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4804                             key.offset >= search_end)
4805                                 break;
4806                 }
4807
4808                 if (inode && key.objectid != inode->i_ino) {
4809                         BUG_ON(extent_locked);
4810                         btrfs_release_path(root, path);
4811                         mutex_unlock(&inode->i_mutex);
4812                         iput(inode);
4813                         inode = NULL;
4814                         continue;
4815                 }
4816
4817                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4818                         path->slots[0]++;
4819                         ret = 1;
4820                         goto next;
4821                 }
4822                 fi = btrfs_item_ptr(leaf, path->slots[0],
4823                                     struct btrfs_file_extent_item);
4824                 extent_type = btrfs_file_extent_type(leaf, fi);
4825                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4826                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4827                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4828                      extent_key->objectid)) {
4829                         path->slots[0]++;
4830                         ret = 1;
4831                         goto next;
4832                 }
4833
4834                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4835                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4836
4837                 if (search_end == (u64)-1) {
4838                         search_end = key.offset - ext_offset +
4839                                 btrfs_file_extent_ram_bytes(leaf, fi);
4840                 }
4841
4842                 if (!extent_locked) {
4843                         lock_start = key.offset;
4844                         lock_end = lock_start + num_bytes - 1;
4845                 } else {
4846                         if (lock_start > key.offset ||
4847                             lock_end + 1 < key.offset + num_bytes) {
4848                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4849                                               lock_start, lock_end, GFP_NOFS);
4850                                 extent_locked = 0;
4851                         }
4852                 }
4853
4854                 if (!inode) {
4855                         btrfs_release_path(root, path);
4856
4857                         inode = btrfs_iget_locked(root->fs_info->sb,
4858                                                   key.objectid, root);
4859                         if (inode->i_state & I_NEW) {
4860                                 BTRFS_I(inode)->root = root;
4861                                 BTRFS_I(inode)->location.objectid =
4862                                         key.objectid;
4863                                 BTRFS_I(inode)->location.type =
4864                                         BTRFS_INODE_ITEM_KEY;
4865                                 BTRFS_I(inode)->location.offset = 0;
4866                                 btrfs_read_locked_inode(inode);
4867                                 unlock_new_inode(inode);
4868                         }
4869                         /*
4870                          * some code call btrfs_commit_transaction while
4871                          * holding the i_mutex, so we can't use mutex_lock
4872                          * here.
4873                          */
4874                         if (is_bad_inode(inode) ||
4875                             !mutex_trylock(&inode->i_mutex)) {
4876                                 iput(inode);
4877                                 inode = NULL;
4878                                 key.offset = (u64)-1;
4879                                 goto skip;
4880                         }
4881                 }
4882
4883                 if (!extent_locked) {
4884                         struct btrfs_ordered_extent *ordered;
4885
4886                         btrfs_release_path(root, path);
4887
4888                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4889                                     lock_end, GFP_NOFS);
4890                         ordered = btrfs_lookup_first_ordered_extent(inode,
4891                                                                     lock_end);
4892                         if (ordered &&
4893                             ordered->file_offset <= lock_end &&
4894                             ordered->file_offset + ordered->len > lock_start) {
4895                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4896                                               lock_start, lock_end, GFP_NOFS);
4897                                 btrfs_start_ordered_extent(inode, ordered, 1);
4898                                 btrfs_put_ordered_extent(ordered);
4899                                 key.offset += num_bytes;
4900                                 goto skip;
4901                         }
4902                         if (ordered)
4903                                 btrfs_put_ordered_extent(ordered);
4904
4905                         extent_locked = 1;
4906                         continue;
4907                 }
4908
4909                 if (nr_extents == 1) {
4910                         /* update extent pointer in place */
4911                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4912                                                 new_extents[0].disk_bytenr);
4913                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4914                                                 new_extents[0].disk_num_bytes);
4915                         btrfs_mark_buffer_dirty(leaf);
4916
4917                         btrfs_drop_extent_cache(inode, key.offset,
4918                                                 key.offset + num_bytes - 1, 0);
4919
4920                         ret = btrfs_inc_extent_ref(trans, root,
4921                                                 new_extents[0].disk_bytenr,
4922                                                 new_extents[0].disk_num_bytes,
4923                                                 leaf->start,
4924                                                 root->root_key.objectid,
4925                                                 trans->transid,
4926                                                 key.objectid);
4927                         BUG_ON(ret);
4928
4929                         ret = btrfs_free_extent(trans, root,
4930                                                 extent_key->objectid,
4931                                                 extent_key->offset,
4932                                                 leaf->start,
4933                                                 btrfs_header_owner(leaf),
4934                                                 btrfs_header_generation(leaf),
4935                                                 key.objectid, 0);
4936                         BUG_ON(ret);
4937
4938                         btrfs_release_path(root, path);
4939                         key.offset += num_bytes;
4940                 } else {
4941                         BUG_ON(1);
4942 #if 0
4943                         u64 alloc_hint;
4944                         u64 extent_len;
4945                         int i;
4946                         /*
4947                          * drop old extent pointer at first, then insert the
4948                          * new pointers one bye one
4949                          */
4950                         btrfs_release_path(root, path);
4951                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4952                                                  key.offset + num_bytes,
4953                                                  key.offset, &alloc_hint);
4954                         BUG_ON(ret);
4955
4956                         for (i = 0; i < nr_extents; i++) {
4957                                 if (ext_offset >= new_extents[i].num_bytes) {
4958                                         ext_offset -= new_extents[i].num_bytes;
4959                                         continue;
4960                                 }
4961                                 extent_len = min(new_extents[i].num_bytes -
4962                                                  ext_offset, num_bytes);
4963
4964                                 ret = btrfs_insert_empty_item(trans, root,
4965                                                               path, &key,
4966                                                               sizeof(*fi));
4967                                 BUG_ON(ret);
4968
4969                                 leaf = path->nodes[0];
4970                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4971                                                 struct btrfs_file_extent_item);
4972                                 btrfs_set_file_extent_generation(leaf, fi,
4973                                                         trans->transid);
4974                                 btrfs_set_file_extent_type(leaf, fi,
4975                                                         BTRFS_FILE_EXTENT_REG);
4976                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4977                                                 new_extents[i].disk_bytenr);
4978                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4979                                                 new_extents[i].disk_num_bytes);
4980                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4981                                                 new_extents[i].ram_bytes);
4982
4983                                 btrfs_set_file_extent_compression(leaf, fi,
4984                                                 new_extents[i].compression);
4985                                 btrfs_set_file_extent_encryption(leaf, fi,
4986                                                 new_extents[i].encryption);
4987                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4988                                                 new_extents[i].other_encoding);
4989
4990                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4991                                                         extent_len);
4992                                 ext_offset += new_extents[i].offset;
4993                                 btrfs_set_file_extent_offset(leaf, fi,
4994                                                         ext_offset);
4995                                 btrfs_mark_buffer_dirty(leaf);
4996
4997                                 btrfs_drop_extent_cache(inode, key.offset,
4998                                                 key.offset + extent_len - 1, 0);
4999
5000                                 ret = btrfs_inc_extent_ref(trans, root,
5001                                                 new_extents[i].disk_bytenr,
5002                                                 new_extents[i].disk_num_bytes,
5003                                                 leaf->start,
5004                                                 root->root_key.objectid,
5005                                                 trans->transid, key.objectid);
5006                                 BUG_ON(ret);
5007                                 btrfs_release_path(root, path);
5008
5009                                 inode_add_bytes(inode, extent_len);
5010
5011                                 ext_offset = 0;
5012                                 num_bytes -= extent_len;
5013                                 key.offset += extent_len;
5014
5015                                 if (num_bytes == 0)
5016                                         break;
5017                         }
5018                         BUG_ON(i >= nr_extents);
5019 #endif
5020                 }
5021
5022                 if (extent_locked) {
5023                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5024                                       lock_end, GFP_NOFS);
5025                         extent_locked = 0;
5026                 }
5027 skip:
5028                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
5029                     key.offset >= search_end)
5030                         break;
5031
5032                 cond_resched();
5033         }
5034         ret = 0;
5035 out:
5036         btrfs_release_path(root, path);
5037         if (inode) {
5038                 mutex_unlock(&inode->i_mutex);
5039                 if (extent_locked) {
5040                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5041                                       lock_end, GFP_NOFS);
5042                 }
5043                 iput(inode);
5044         }
5045         return ret;
5046 }
5047
5048 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
5049                                struct btrfs_root *root,
5050                                struct extent_buffer *buf, u64 orig_start)
5051 {
5052         int level;
5053         int ret;
5054
5055         BUG_ON(btrfs_header_generation(buf) != trans->transid);
5056         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5057
5058         level = btrfs_header_level(buf);
5059         if (level == 0) {
5060                 struct btrfs_leaf_ref *ref;
5061                 struct btrfs_leaf_ref *orig_ref;
5062
5063                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
5064                 if (!orig_ref)
5065                         return -ENOENT;
5066
5067                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
5068                 if (!ref) {
5069                         btrfs_free_leaf_ref(root, orig_ref);
5070                         return -ENOMEM;
5071                 }
5072
5073                 ref->nritems = orig_ref->nritems;
5074                 memcpy(ref->extents, orig_ref->extents,
5075                         sizeof(ref->extents[0]) * ref->nritems);
5076
5077                 btrfs_free_leaf_ref(root, orig_ref);
5078
5079                 ref->root_gen = trans->transid;
5080                 ref->bytenr = buf->start;
5081                 ref->owner = btrfs_header_owner(buf);
5082                 ref->generation = btrfs_header_generation(buf);
5083
5084                 ret = btrfs_add_leaf_ref(root, ref, 0);
5085                 WARN_ON(ret);
5086                 btrfs_free_leaf_ref(root, ref);
5087         }
5088         return 0;
5089 }
5090
5091 static noinline int invalidate_extent_cache(struct btrfs_root *root,
5092                                         struct extent_buffer *leaf,
5093                                         struct btrfs_block_group_cache *group,
5094                                         struct btrfs_root *target_root)
5095 {
5096         struct btrfs_key key;
5097         struct inode *inode = NULL;
5098         struct btrfs_file_extent_item *fi;
5099         u64 num_bytes;
5100         u64 skip_objectid = 0;
5101         u32 nritems;
5102         u32 i;
5103
5104         nritems = btrfs_header_nritems(leaf);
5105         for (i = 0; i < nritems; i++) {
5106                 btrfs_item_key_to_cpu(leaf, &key, i);
5107                 if (key.objectid == skip_objectid ||
5108                     key.type != BTRFS_EXTENT_DATA_KEY)
5109                         continue;
5110                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5111                 if (btrfs_file_extent_type(leaf, fi) ==
5112                     BTRFS_FILE_EXTENT_INLINE)
5113                         continue;
5114                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5115                         continue;
5116                 if (!inode || inode->i_ino != key.objectid) {
5117                         iput(inode);
5118                         inode = btrfs_ilookup(target_root->fs_info->sb,
5119                                               key.objectid, target_root, 1);
5120                 }
5121                 if (!inode) {
5122                         skip_objectid = key.objectid;
5123                         continue;
5124                 }
5125                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5126
5127                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5128                             key.offset + num_bytes - 1, GFP_NOFS);
5129                 btrfs_drop_extent_cache(inode, key.offset,
5130                                         key.offset + num_bytes - 1, 1);
5131                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5132                               key.offset + num_bytes - 1, GFP_NOFS);
5133                 cond_resched();
5134         }
5135         iput(inode);
5136         return 0;
5137 }
5138
5139 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
5140                                         struct btrfs_root *root,
5141                                         struct extent_buffer *leaf,
5142                                         struct btrfs_block_group_cache *group,
5143                                         struct inode *reloc_inode)
5144 {
5145         struct btrfs_key key;
5146         struct btrfs_key extent_key;
5147         struct btrfs_file_extent_item *fi;
5148         struct btrfs_leaf_ref *ref;
5149         struct disk_extent *new_extent;
5150         u64 bytenr;
5151         u64 num_bytes;
5152         u32 nritems;
5153         u32 i;
5154         int ext_index;
5155         int nr_extent;
5156         int ret;
5157
5158         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
5159         BUG_ON(!new_extent);
5160
5161         ref = btrfs_lookup_leaf_ref(root, leaf->start);
5162         BUG_ON(!ref);
5163
5164         ext_index = -1;
5165         nritems = btrfs_header_nritems(leaf);
5166         for (i = 0; i < nritems; i++) {
5167                 btrfs_item_key_to_cpu(leaf, &key, i);
5168                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
5169                         continue;
5170                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5171                 if (btrfs_file_extent_type(leaf, fi) ==
5172                     BTRFS_FILE_EXTENT_INLINE)
5173                         continue;
5174                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5175                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
5176                 if (bytenr == 0)
5177                         continue;
5178
5179                 ext_index++;
5180                 if (bytenr >= group->key.objectid + group->key.offset ||
5181                     bytenr + num_bytes <= group->key.objectid)
5182                         continue;
5183
5184                 extent_key.objectid = bytenr;
5185                 extent_key.offset = num_bytes;
5186                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
5187                 nr_extent = 1;
5188                 ret = get_new_locations(reloc_inode, &extent_key,
5189                                         group->key.objectid, 1,
5190                                         &new_extent, &nr_extent);
5191                 if (ret > 0)
5192                         continue;
5193                 BUG_ON(ret < 0);
5194
5195                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
5196                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
5197                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
5198                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
5199
5200                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5201                                                 new_extent->disk_bytenr);
5202                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5203                                                 new_extent->disk_num_bytes);
5204                 btrfs_mark_buffer_dirty(leaf);
5205
5206                 ret = btrfs_inc_extent_ref(trans, root,
5207                                         new_extent->disk_bytenr,
5208                                         new_extent->disk_num_bytes,
5209                                         leaf->start,
5210                                         root->root_key.objectid,
5211                                         trans->transid, key.objectid);
5212                 BUG_ON(ret);
5213                 ret = btrfs_free_extent(trans, root,
5214                                         bytenr, num_bytes, leaf->start,
5215                                         btrfs_header_owner(leaf),
5216                                         btrfs_header_generation(leaf),
5217                                         key.objectid, 0);
5218                 BUG_ON(ret);
5219                 cond_resched();
5220         }
5221         kfree(new_extent);
5222         BUG_ON(ext_index + 1 != ref->nritems);
5223         btrfs_free_leaf_ref(root, ref);
5224         return 0;
5225 }
5226
5227 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
5228                           struct btrfs_root *root)
5229 {
5230         struct btrfs_root *reloc_root;
5231         int ret;
5232
5233         if (root->reloc_root) {
5234                 reloc_root = root->reloc_root;
5235                 root->reloc_root = NULL;
5236                 list_add(&reloc_root->dead_list,
5237                          &root->fs_info->dead_reloc_roots);
5238
5239                 btrfs_set_root_bytenr(&reloc_root->root_item,
5240                                       reloc_root->node->start);
5241                 btrfs_set_root_level(&root->root_item,
5242                                      btrfs_header_level(reloc_root->node));
5243                 memset(&reloc_root->root_item.drop_progress, 0,
5244                         sizeof(struct btrfs_disk_key));
5245                 reloc_root->root_item.drop_level = 0;
5246
5247                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
5248                                         &reloc_root->root_key,
5249                                         &reloc_root->root_item);
5250                 BUG_ON(ret);
5251         }
5252         return 0;
5253 }
5254
5255 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
5256 {
5257         struct btrfs_trans_handle *trans;
5258         struct btrfs_root *reloc_root;
5259         struct btrfs_root *prev_root = NULL;
5260         struct list_head dead_roots;
5261         int ret;
5262         unsigned long nr;
5263
5264         INIT_LIST_HEAD(&dead_roots);
5265         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
5266
5267         while (!list_empty(&dead_roots)) {
5268                 reloc_root = list_entry(dead_roots.prev,
5269                                         struct btrfs_root, dead_list);
5270                 list_del_init(&reloc_root->dead_list);
5271
5272                 BUG_ON(reloc_root->commit_root != NULL);
5273                 while (1) {
5274                         trans = btrfs_join_transaction(root, 1);
5275                         BUG_ON(!trans);
5276
5277                         mutex_lock(&root->fs_info->drop_mutex);
5278                         ret = btrfs_drop_snapshot(trans, reloc_root);
5279                         if (ret != -EAGAIN)
5280                                 break;
5281                         mutex_unlock(&root->fs_info->drop_mutex);
5282
5283                         nr = trans->blocks_used;
5284                         ret = btrfs_end_transaction(trans, root);
5285                         BUG_ON(ret);
5286                         btrfs_btree_balance_dirty(root, nr);
5287                 }
5288
5289                 free_extent_buffer(reloc_root->node);
5290
5291                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
5292                                      &reloc_root->root_key);
5293                 BUG_ON(ret);
5294                 mutex_unlock(&root->fs_info->drop_mutex);
5295
5296                 nr = trans->blocks_used;
5297                 ret = btrfs_end_transaction(trans, root);
5298                 BUG_ON(ret);
5299                 btrfs_btree_balance_dirty(root, nr);
5300
5301                 kfree(prev_root);
5302                 prev_root = reloc_root;
5303         }
5304         if (prev_root) {
5305                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5306                 kfree(prev_root);
5307         }
5308         return 0;
5309 }
5310
5311 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5312 {
5313         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5314         return 0;
5315 }
5316
5317 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5318 {
5319         struct btrfs_root *reloc_root;
5320         struct btrfs_trans_handle *trans;
5321         struct btrfs_key location;
5322         int found;
5323         int ret;
5324
5325         mutex_lock(&root->fs_info->tree_reloc_mutex);
5326         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5327         BUG_ON(ret);
5328         found = !list_empty(&root->fs_info->dead_reloc_roots);
5329         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5330
5331         if (found) {
5332                 trans = btrfs_start_transaction(root, 1);
5333                 BUG_ON(!trans);
5334                 ret = btrfs_commit_transaction(trans, root);
5335                 BUG_ON(ret);
5336         }
5337
5338         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5339         location.offset = (u64)-1;
5340         location.type = BTRFS_ROOT_ITEM_KEY;
5341
5342         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5343         BUG_ON(!reloc_root);
5344         btrfs_orphan_cleanup(reloc_root);
5345         return 0;
5346 }
5347
5348 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5349                                     struct btrfs_root *root)
5350 {
5351         struct btrfs_root *reloc_root;
5352         struct extent_buffer *eb;
5353         struct btrfs_root_item *root_item;
5354         struct btrfs_key root_key;
5355         int ret;
5356
5357         BUG_ON(!root->ref_cows);
5358         if (root->reloc_root)
5359                 return 0;
5360
5361         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5362         BUG_ON(!root_item);
5363
5364         ret = btrfs_copy_root(trans, root, root->commit_root,
5365                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5366         BUG_ON(ret);
5367
5368         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5369         root_key.offset = root->root_key.objectid;
5370         root_key.type = BTRFS_ROOT_ITEM_KEY;
5371
5372         memcpy(root_item, &root->root_item, sizeof(root_item));
5373         btrfs_set_root_refs(root_item, 0);
5374         btrfs_set_root_bytenr(root_item, eb->start);
5375         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5376         btrfs_set_root_generation(root_item, trans->transid);
5377
5378         btrfs_tree_unlock(eb);
5379         free_extent_buffer(eb);
5380
5381         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5382                                 &root_key, root_item);
5383         BUG_ON(ret);
5384         kfree(root_item);
5385
5386         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5387                                                  &root_key);
5388         BUG_ON(!reloc_root);
5389         reloc_root->last_trans = trans->transid;
5390         reloc_root->commit_root = NULL;
5391         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5392
5393         root->reloc_root = reloc_root;
5394         return 0;
5395 }
5396
5397 /*
5398  * Core function of space balance.
5399  *
5400  * The idea is using reloc trees to relocate tree blocks in reference
5401  * counted roots. There is one reloc tree for each subvol, and all
5402  * reloc trees share same root key objectid. Reloc trees are snapshots
5403  * of the latest committed roots of subvols (root->commit_root).
5404  *
5405  * To relocate a tree block referenced by a subvol, there are two steps.
5406  * COW the block through subvol's reloc tree, then update block pointer
5407  * in the subvol to point to the new block. Since all reloc trees share
5408  * same root key objectid, doing special handing for tree blocks owned
5409  * by them is easy. Once a tree block has been COWed in one reloc tree,
5410  * we can use the resulting new block directly when the same block is
5411  * required to COW again through other reloc trees. By this way, relocated
5412  * tree blocks are shared between reloc trees, so they are also shared
5413  * between subvols.
5414  */
5415 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5416                                       struct btrfs_root *root,
5417                                       struct btrfs_path *path,
5418                                       struct btrfs_key *first_key,
5419                                       struct btrfs_ref_path *ref_path,
5420                                       struct btrfs_block_group_cache *group,
5421                                       struct inode *reloc_inode)
5422 {
5423         struct btrfs_root *reloc_root;
5424         struct extent_buffer *eb = NULL;
5425         struct btrfs_key *keys;
5426         u64 *nodes;
5427         int level;
5428         int shared_level;
5429         int lowest_level = 0;
5430         int ret;
5431
5432         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5433                 lowest_level = ref_path->owner_objectid;
5434
5435         if (!root->ref_cows) {
5436                 path->lowest_level = lowest_level;
5437                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5438                 BUG_ON(ret < 0);
5439                 path->lowest_level = 0;
5440                 btrfs_release_path(root, path);
5441                 return 0;
5442         }
5443
5444         mutex_lock(&root->fs_info->tree_reloc_mutex);
5445         ret = init_reloc_tree(trans, root);
5446         BUG_ON(ret);
5447         reloc_root = root->reloc_root;
5448
5449         shared_level = ref_path->shared_level;
5450         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5451
5452         keys = ref_path->node_keys;
5453         nodes = ref_path->new_nodes;
5454         memset(&keys[shared_level + 1], 0,
5455                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5456         memset(&nodes[shared_level + 1], 0,
5457                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5458
5459         if (nodes[lowest_level] == 0) {
5460                 path->lowest_level = lowest_level;
5461                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5462                                         0, 1);
5463                 BUG_ON(ret);
5464                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5465                         eb = path->nodes[level];
5466                         if (!eb || eb == reloc_root->node)
5467                                 break;
5468                         nodes[level] = eb->start;
5469                         if (level == 0)
5470                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5471                         else
5472                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5473                 }
5474                 if (nodes[0] &&
5475                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5476                         eb = path->nodes[0];
5477                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5478                                                       group, reloc_inode);
5479                         BUG_ON(ret);
5480                 }
5481                 btrfs_release_path(reloc_root, path);
5482         } else {
5483                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5484                                        lowest_level);
5485                 BUG_ON(ret);
5486         }
5487
5488         /*
5489          * replace tree blocks in the fs tree with tree blocks in
5490          * the reloc tree.
5491          */
5492         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5493         BUG_ON(ret < 0);
5494
5495         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5496                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5497                                         0, 0);
5498                 BUG_ON(ret);
5499                 extent_buffer_get(path->nodes[0]);
5500                 eb = path->nodes[0];
5501                 btrfs_release_path(reloc_root, path);
5502                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5503                 BUG_ON(ret);
5504                 free_extent_buffer(eb);
5505         }
5506
5507         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5508         path->lowest_level = 0;
5509         return 0;
5510 }
5511
5512 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5513                                         struct btrfs_root *root,
5514                                         struct btrfs_path *path,
5515                                         struct btrfs_key *first_key,
5516                                         struct btrfs_ref_path *ref_path)
5517 {
5518         int ret;
5519
5520         ret = relocate_one_path(trans, root, path, first_key,
5521                                 ref_path, NULL, NULL);
5522         BUG_ON(ret);
5523
5524         if (root == root->fs_info->extent_root)
5525                 btrfs_extent_post_op(trans, root);
5526
5527         return 0;
5528 }
5529
5530 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5531                                     struct btrfs_root *extent_root,
5532                                     struct btrfs_path *path,
5533                                     struct btrfs_key *extent_key)
5534 {
5535         int ret;
5536
5537         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5538         if (ret)
5539                 goto out;
5540         ret = btrfs_del_item(trans, extent_root, path);
5541 out:
5542         btrfs_release_path(extent_root, path);
5543         return ret;
5544 }
5545
5546 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5547                                                 struct btrfs_ref_path *ref_path)
5548 {
5549         struct btrfs_key root_key;
5550
5551         root_key.objectid = ref_path->root_objectid;
5552         root_key.type = BTRFS_ROOT_ITEM_KEY;
5553         if (is_cowonly_root(ref_path->root_objectid))
5554                 root_key.offset = 0;
5555         else
5556                 root_key.offset = (u64)-1;
5557
5558         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5559 }
5560
5561 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5562                                         struct btrfs_path *path,
5563                                         struct btrfs_key *extent_key,
5564                                         struct btrfs_block_group_cache *group,
5565                                         struct inode *reloc_inode, int pass)
5566 {
5567         struct btrfs_trans_handle *trans;
5568         struct btrfs_root *found_root;
5569         struct btrfs_ref_path *ref_path = NULL;
5570         struct disk_extent *new_extents = NULL;
5571         int nr_extents = 0;
5572         int loops;
5573         int ret;
5574         int level;
5575         struct btrfs_key first_key;
5576         u64 prev_block = 0;
5577
5578
5579         trans = btrfs_start_transaction(extent_root, 1);
5580         BUG_ON(!trans);
5581
5582         if (extent_key->objectid == 0) {
5583                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5584                 goto out;
5585         }
5586
5587         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5588         if (!ref_path) {
5589                 ret = -ENOMEM;
5590                 goto out;
5591         }
5592
5593         for (loops = 0; ; loops++) {
5594                 if (loops == 0) {
5595                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5596                                                    extent_key->objectid);
5597                 } else {
5598                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5599                 }
5600                 if (ret < 0)
5601                         goto out;
5602                 if (ret > 0)
5603                         break;
5604
5605                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5606                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5607                         continue;
5608
5609                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5610                 BUG_ON(!found_root);
5611                 /*
5612                  * for reference counted tree, only process reference paths
5613                  * rooted at the latest committed root.
5614                  */
5615                 if (found_root->ref_cows &&
5616                     ref_path->root_generation != found_root->root_key.offset)
5617                         continue;
5618
5619                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5620                         if (pass == 0) {
5621                                 /*
5622                                  * copy data extents to new locations
5623                                  */
5624                                 u64 group_start = group->key.objectid;
5625                                 ret = relocate_data_extent(reloc_inode,
5626                                                            extent_key,
5627                                                            group_start);
5628                                 if (ret < 0)
5629                                         goto out;
5630                                 break;
5631                         }
5632                         level = 0;
5633                 } else {
5634                         level = ref_path->owner_objectid;
5635                 }
5636
5637                 if (prev_block != ref_path->nodes[level]) {
5638                         struct extent_buffer *eb;
5639                         u64 block_start = ref_path->nodes[level];
5640                         u64 block_size = btrfs_level_size(found_root, level);
5641
5642                         eb = read_tree_block(found_root, block_start,
5643                                              block_size, 0);
5644                         btrfs_tree_lock(eb);
5645                         BUG_ON(level != btrfs_header_level(eb));
5646
5647                         if (level == 0)
5648                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5649                         else
5650                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5651
5652                         btrfs_tree_unlock(eb);
5653                         free_extent_buffer(eb);
5654                         prev_block = block_start;
5655                 }
5656
5657                 btrfs_record_root_in_trans(found_root);
5658                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5659                         /*
5660                          * try to update data extent references while
5661                          * keeping metadata shared between snapshots.
5662                          */
5663                         if (pass == 1) {
5664                                 ret = relocate_one_path(trans, found_root,
5665                                                 path, &first_key, ref_path,
5666                                                 group, reloc_inode);
5667                                 if (ret < 0)
5668                                         goto out;
5669                                 continue;
5670                         }
5671                         /*
5672                          * use fallback method to process the remaining
5673                          * references.
5674                          */
5675                         if (!new_extents) {
5676                                 u64 group_start = group->key.objectid;
5677                                 new_extents = kmalloc(sizeof(*new_extents),
5678                                                       GFP_NOFS);
5679                                 nr_extents = 1;
5680                                 ret = get_new_locations(reloc_inode,
5681                                                         extent_key,
5682                                                         group_start, 1,
5683                                                         &new_extents,
5684                                                         &nr_extents);
5685                                 if (ret)
5686                                         goto out;
5687                         }
5688                         ret = replace_one_extent(trans, found_root,
5689                                                 path, extent_key,
5690                                                 &first_key, ref_path,
5691                                                 new_extents, nr_extents);
5692                 } else {
5693                         ret = relocate_tree_block(trans, found_root, path,
5694                                                   &first_key, ref_path);
5695                 }
5696                 if (ret < 0)
5697                         goto out;
5698         }
5699         ret = 0;
5700 out:
5701         btrfs_end_transaction(trans, extent_root);
5702         kfree(new_extents);
5703         kfree(ref_path);
5704         return ret;
5705 }
5706
5707 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5708 {
5709         u64 num_devices;
5710         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5711                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5712
5713         num_devices = root->fs_info->fs_devices->rw_devices;
5714         if (num_devices == 1) {
5715                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5716                 stripped = flags & ~stripped;
5717
5718                 /* turn raid0 into single device chunks */
5719                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5720                         return stripped;
5721
5722                 /* turn mirroring into duplication */
5723                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5724                              BTRFS_BLOCK_GROUP_RAID10))
5725                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5726                 return flags;
5727         } else {
5728                 /* they already had raid on here, just return */
5729                 if (flags & stripped)
5730                         return flags;
5731
5732                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5733                 stripped = flags & ~stripped;
5734
5735                 /* switch duplicated blocks with raid1 */
5736                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5737                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5738
5739                 /* turn single device chunks into raid0 */
5740                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5741         }
5742         return flags;
5743 }
5744
5745 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5746                      struct btrfs_block_group_cache *shrink_block_group,
5747                      int force)
5748 {
5749         struct btrfs_trans_handle *trans;
5750         u64 new_alloc_flags;
5751         u64 calc;
5752
5753         spin_lock(&shrink_block_group->lock);
5754         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5755                 spin_unlock(&shrink_block_group->lock);
5756
5757                 trans = btrfs_start_transaction(root, 1);
5758                 spin_lock(&shrink_block_group->lock);
5759
5760                 new_alloc_flags = update_block_group_flags(root,
5761                                                    shrink_block_group->flags);
5762                 if (new_alloc_flags != shrink_block_group->flags) {
5763                         calc =
5764                              btrfs_block_group_used(&shrink_block_group->item);
5765                 } else {
5766                         calc = shrink_block_group->key.offset;
5767                 }
5768                 spin_unlock(&shrink_block_group->lock);
5769
5770                 do_chunk_alloc(trans, root->fs_info->extent_root,
5771                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5772
5773                 btrfs_end_transaction(trans, root);
5774         } else
5775                 spin_unlock(&shrink_block_group->lock);
5776         return 0;
5777 }
5778
5779 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5780                                  struct btrfs_root *root,
5781                                  u64 objectid, u64 size)
5782 {
5783         struct btrfs_path *path;
5784         struct btrfs_inode_item *item;
5785         struct extent_buffer *leaf;
5786         int ret;
5787
5788         path = btrfs_alloc_path();
5789         if (!path)
5790                 return -ENOMEM;
5791
5792         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5793         if (ret)
5794                 goto out;
5795
5796         leaf = path->nodes[0];
5797         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5798         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5799         btrfs_set_inode_generation(leaf, item, 1);
5800         btrfs_set_inode_size(leaf, item, size);
5801         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5802         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5803         btrfs_mark_buffer_dirty(leaf);
5804         btrfs_release_path(root, path);
5805 out:
5806         btrfs_free_path(path);
5807         return ret;
5808 }
5809
5810 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5811                                         struct btrfs_block_group_cache *group)
5812 {
5813         struct inode *inode = NULL;
5814         struct btrfs_trans_handle *trans;
5815         struct btrfs_root *root;
5816         struct btrfs_key root_key;
5817         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5818         int err = 0;
5819
5820         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5821         root_key.type = BTRFS_ROOT_ITEM_KEY;
5822         root_key.offset = (u64)-1;
5823         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5824         if (IS_ERR(root))
5825                 return ERR_CAST(root);
5826
5827         trans = btrfs_start_transaction(root, 1);
5828         BUG_ON(!trans);
5829
5830         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5831         if (err)
5832                 goto out;
5833
5834         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5835         BUG_ON(err);
5836
5837         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5838                                        group->key.offset, 0, group->key.offset,
5839                                        0, 0, 0);
5840         BUG_ON(err);
5841
5842         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5843         if (inode->i_state & I_NEW) {
5844                 BTRFS_I(inode)->root = root;
5845                 BTRFS_I(inode)->location.objectid = objectid;
5846                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5847                 BTRFS_I(inode)->location.offset = 0;
5848                 btrfs_read_locked_inode(inode);
5849                 unlock_new_inode(inode);
5850                 BUG_ON(is_bad_inode(inode));
5851         } else {
5852                 BUG_ON(1);
5853         }
5854         BTRFS_I(inode)->index_cnt = group->key.objectid;
5855
5856         err = btrfs_orphan_add(trans, inode);
5857 out:
5858         btrfs_end_transaction(trans, root);
5859         if (err) {
5860                 if (inode)
5861                         iput(inode);
5862                 inode = ERR_PTR(err);
5863         }
5864         return inode;
5865 }
5866
5867 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5868 {
5869
5870         struct btrfs_ordered_sum *sums;
5871         struct btrfs_sector_sum *sector_sum;
5872         struct btrfs_ordered_extent *ordered;
5873         struct btrfs_root *root = BTRFS_I(inode)->root;
5874         struct list_head list;
5875         size_t offset;
5876         int ret;
5877         u64 disk_bytenr;
5878
5879         INIT_LIST_HEAD(&list);
5880
5881         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5882         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5883
5884         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5885         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5886                                        disk_bytenr + len - 1, &list);
5887
5888         while (!list_empty(&list)) {
5889                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5890                 list_del_init(&sums->list);
5891
5892                 sector_sum = sums->sums;
5893                 sums->bytenr = ordered->start;
5894
5895                 offset = 0;
5896                 while (offset < sums->len) {
5897                         sector_sum->bytenr += ordered->start - disk_bytenr;
5898                         sector_sum++;
5899                         offset += root->sectorsize;
5900                 }
5901
5902                 btrfs_add_ordered_sum(inode, ordered, sums);
5903         }
5904         btrfs_put_ordered_extent(ordered);
5905         return 0;
5906 }
5907
5908 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5909 {
5910         struct btrfs_trans_handle *trans;
5911         struct btrfs_path *path;
5912         struct btrfs_fs_info *info = root->fs_info;
5913         struct extent_buffer *leaf;
5914         struct inode *reloc_inode;
5915         struct btrfs_block_group_cache *block_group;
5916         struct btrfs_key key;
5917         u64 skipped;
5918         u64 cur_byte;
5919         u64 total_found;
5920         u32 nritems;
5921         int ret;
5922         int progress;
5923         int pass = 0;
5924
5925         root = root->fs_info->extent_root;
5926
5927         block_group = btrfs_lookup_block_group(info, group_start);
5928         BUG_ON(!block_group);
5929
5930         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5931                (unsigned long long)block_group->key.objectid,
5932                (unsigned long long)block_group->flags);
5933
5934         path = btrfs_alloc_path();
5935         BUG_ON(!path);
5936
5937         reloc_inode = create_reloc_inode(info, block_group);
5938         BUG_ON(IS_ERR(reloc_inode));
5939
5940         __alloc_chunk_for_shrink(root, block_group, 1);
5941         set_block_group_readonly(block_group);
5942
5943         btrfs_start_delalloc_inodes(info->tree_root);
5944         btrfs_wait_ordered_extents(info->tree_root, 0);
5945 again:
5946         skipped = 0;
5947         total_found = 0;
5948         progress = 0;
5949         key.objectid = block_group->key.objectid;
5950         key.offset = 0;
5951         key.type = 0;
5952         cur_byte = key.objectid;
5953
5954         trans = btrfs_start_transaction(info->tree_root, 1);
5955         btrfs_commit_transaction(trans, info->tree_root);
5956
5957         mutex_lock(&root->fs_info->cleaner_mutex);
5958         btrfs_clean_old_snapshots(info->tree_root);
5959         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5960         mutex_unlock(&root->fs_info->cleaner_mutex);
5961
5962         while (1) {
5963                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5964                 if (ret < 0)
5965                         goto out;
5966 next:
5967                 leaf = path->nodes[0];
5968                 nritems = btrfs_header_nritems(leaf);
5969                 if (path->slots[0] >= nritems) {
5970                         ret = btrfs_next_leaf(root, path);
5971                         if (ret < 0)
5972                                 goto out;
5973                         if (ret == 1) {
5974                                 ret = 0;
5975                                 break;
5976                         }
5977                         leaf = path->nodes[0];
5978                         nritems = btrfs_header_nritems(leaf);
5979                 }
5980
5981                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5982
5983                 if (key.objectid >= block_group->key.objectid +
5984                     block_group->key.offset)
5985                         break;
5986
5987                 if (progress && need_resched()) {
5988                         btrfs_release_path(root, path);
5989                         cond_resched();
5990                         progress = 0;
5991                         continue;
5992                 }
5993                 progress = 1;
5994
5995                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5996                     key.objectid + key.offset <= cur_byte) {
5997                         path->slots[0]++;
5998                         goto next;
5999                 }
6000
6001                 total_found++;
6002                 cur_byte = key.objectid + key.offset;
6003                 btrfs_release_path(root, path);
6004
6005                 __alloc_chunk_for_shrink(root, block_group, 0);
6006                 ret = relocate_one_extent(root, path, &key, block_group,
6007                                           reloc_inode, pass);
6008                 BUG_ON(ret < 0);
6009                 if (ret > 0)
6010                         skipped++;
6011
6012                 key.objectid = cur_byte;
6013                 key.type = 0;
6014                 key.offset = 0;
6015         }
6016
6017         btrfs_release_path(root, path);
6018
6019         if (pass == 0) {
6020                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
6021                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
6022         }
6023
6024         if (total_found > 0) {
6025                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
6026                        (unsigned long long)total_found, pass);
6027                 pass++;
6028                 if (total_found == skipped && pass > 2) {
6029                         iput(reloc_inode);
6030                         reloc_inode = create_reloc_inode(info, block_group);
6031                         pass = 0;
6032                 }
6033                 goto again;
6034         }
6035
6036         /* delete reloc_inode */
6037         iput(reloc_inode);
6038
6039         /* unpin extents in this range */
6040         trans = btrfs_start_transaction(info->tree_root, 1);
6041         btrfs_commit_transaction(trans, info->tree_root);
6042
6043         spin_lock(&block_group->lock);
6044         WARN_ON(block_group->pinned > 0);
6045         WARN_ON(block_group->reserved > 0);
6046         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
6047         spin_unlock(&block_group->lock);
6048         put_block_group(block_group);
6049         ret = 0;
6050 out:
6051         btrfs_free_path(path);
6052         return ret;
6053 }
6054
6055 static int find_first_block_group(struct btrfs_root *root,
6056                 struct btrfs_path *path, struct btrfs_key *key)
6057 {
6058         int ret = 0;
6059         struct btrfs_key found_key;
6060         struct extent_buffer *leaf;
6061         int slot;
6062
6063         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6064         if (ret < 0)
6065                 goto out;
6066
6067         while (1) {
6068                 slot = path->slots[0];
6069                 leaf = path->nodes[0];
6070                 if (slot >= btrfs_header_nritems(leaf)) {
6071                         ret = btrfs_next_leaf(root, path);
6072                         if (ret == 0)
6073                                 continue;
6074                         if (ret < 0)
6075                                 goto out;
6076                         break;
6077                 }
6078                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6079
6080                 if (found_key.objectid >= key->objectid &&
6081                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6082                         ret = 0;
6083                         goto out;
6084                 }
6085                 path->slots[0]++;
6086         }
6087         ret = -ENOENT;
6088 out:
6089         return ret;
6090 }
6091
6092 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6093 {
6094         struct btrfs_block_group_cache *block_group;
6095         struct rb_node *n;
6096
6097         spin_lock(&info->block_group_cache_lock);
6098         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6099                 block_group = rb_entry(n, struct btrfs_block_group_cache,
6100                                        cache_node);
6101                 rb_erase(&block_group->cache_node,
6102                          &info->block_group_cache_tree);
6103                 spin_unlock(&info->block_group_cache_lock);
6104
6105                 btrfs_remove_free_space_cache(block_group);
6106                 down_write(&block_group->space_info->groups_sem);
6107                 list_del(&block_group->list);
6108                 up_write(&block_group->space_info->groups_sem);
6109
6110                 WARN_ON(atomic_read(&block_group->count) != 1);
6111                 kfree(block_group);
6112
6113                 spin_lock(&info->block_group_cache_lock);
6114         }
6115         spin_unlock(&info->block_group_cache_lock);
6116         return 0;
6117 }
6118
6119 int btrfs_read_block_groups(struct btrfs_root *root)
6120 {
6121         struct btrfs_path *path;
6122         int ret;
6123         struct btrfs_block_group_cache *cache;
6124         struct btrfs_fs_info *info = root->fs_info;
6125         struct btrfs_space_info *space_info;
6126         struct btrfs_key key;
6127         struct btrfs_key found_key;
6128         struct extent_buffer *leaf;
6129
6130         root = info->extent_root;
6131         key.objectid = 0;
6132         key.offset = 0;
6133         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
6134         path = btrfs_alloc_path();
6135         if (!path)
6136                 return -ENOMEM;
6137
6138         while (1) {
6139                 ret = find_first_block_group(root, path, &key);
6140                 if (ret > 0) {
6141                         ret = 0;
6142                         goto error;
6143                 }
6144                 if (ret != 0)
6145                         goto error;
6146
6147                 leaf = path->nodes[0];
6148                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6149                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6150                 if (!cache) {
6151                         ret = -ENOMEM;
6152                         break;
6153                 }
6154
6155                 atomic_set(&cache->count, 1);
6156                 spin_lock_init(&cache->lock);
6157                 mutex_init(&cache->alloc_mutex);
6158                 mutex_init(&cache->cache_mutex);
6159                 INIT_LIST_HEAD(&cache->list);
6160                 read_extent_buffer(leaf, &cache->item,
6161                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
6162                                    sizeof(cache->item));
6163                 memcpy(&cache->key, &found_key, sizeof(found_key));
6164
6165                 key.objectid = found_key.objectid + found_key.offset;
6166                 btrfs_release_path(root, path);
6167                 cache->flags = btrfs_block_group_flags(&cache->item);
6168
6169                 ret = update_space_info(info, cache->flags, found_key.offset,
6170                                         btrfs_block_group_used(&cache->item),
6171                                         &space_info);
6172                 BUG_ON(ret);
6173                 cache->space_info = space_info;
6174                 down_write(&space_info->groups_sem);
6175                 list_add_tail(&cache->list, &space_info->block_groups);
6176                 up_write(&space_info->groups_sem);
6177
6178                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
6179                 BUG_ON(ret);
6180
6181                 set_avail_alloc_bits(root->fs_info, cache->flags);
6182                 if (btrfs_chunk_readonly(root, cache->key.objectid))
6183                         set_block_group_readonly(cache);
6184         }
6185         ret = 0;
6186 error:
6187         btrfs_free_path(path);
6188         return ret;
6189 }
6190
6191 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
6192                            struct btrfs_root *root, u64 bytes_used,
6193                            u64 type, u64 chunk_objectid, u64 chunk_offset,
6194                            u64 size)
6195 {
6196         int ret;
6197         struct btrfs_root *extent_root;
6198         struct btrfs_block_group_cache *cache;
6199
6200         extent_root = root->fs_info->extent_root;
6201
6202         root->fs_info->last_trans_new_blockgroup = trans->transid;
6203
6204         cache = kzalloc(sizeof(*cache), GFP_NOFS);
6205         if (!cache)
6206                 return -ENOMEM;
6207
6208         cache->key.objectid = chunk_offset;
6209         cache->key.offset = size;
6210         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
6211         atomic_set(&cache->count, 1);
6212         spin_lock_init(&cache->lock);
6213         mutex_init(&cache->alloc_mutex);
6214         mutex_init(&cache->cache_mutex);
6215         INIT_LIST_HEAD(&cache->list);
6216
6217         btrfs_set_block_group_used(&cache->item, bytes_used);
6218         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
6219         cache->flags = type;
6220         btrfs_set_block_group_flags(&cache->item, type);
6221
6222         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
6223                                 &cache->space_info);
6224         BUG_ON(ret);
6225         down_write(&cache->space_info->groups_sem);
6226         list_add_tail(&cache->list, &cache->space_info->block_groups);
6227         up_write(&cache->space_info->groups_sem);
6228
6229         ret = btrfs_add_block_group_cache(root->fs_info, cache);
6230         BUG_ON(ret);
6231
6232         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
6233                                 sizeof(cache->item));
6234         BUG_ON(ret);
6235
6236         finish_current_insert(trans, extent_root, 0);
6237         ret = del_pending_extents(trans, extent_root, 0);
6238         BUG_ON(ret);
6239         set_avail_alloc_bits(extent_root->fs_info, type);
6240
6241         return 0;
6242 }
6243
6244 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
6245                              struct btrfs_root *root, u64 group_start)
6246 {
6247         struct btrfs_path *path;
6248         struct btrfs_block_group_cache *block_group;
6249         struct btrfs_key key;
6250         int ret;
6251
6252         root = root->fs_info->extent_root;
6253
6254         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
6255         BUG_ON(!block_group);
6256         BUG_ON(!block_group->ro);
6257
6258         memcpy(&key, &block_group->key, sizeof(key));
6259
6260         path = btrfs_alloc_path();
6261         BUG_ON(!path);
6262
6263         spin_lock(&root->fs_info->block_group_cache_lock);
6264         rb_erase(&block_group->cache_node,
6265                  &root->fs_info->block_group_cache_tree);
6266         spin_unlock(&root->fs_info->block_group_cache_lock);
6267         btrfs_remove_free_space_cache(block_group);
6268         down_write(&block_group->space_info->groups_sem);
6269         list_del(&block_group->list);
6270         up_write(&block_group->space_info->groups_sem);
6271
6272         spin_lock(&block_group->space_info->lock);
6273         block_group->space_info->total_bytes -= block_group->key.offset;
6274         block_group->space_info->bytes_readonly -= block_group->key.offset;
6275         spin_unlock(&block_group->space_info->lock);
6276         block_group->space_info->full = 0;
6277
6278         put_block_group(block_group);
6279         put_block_group(block_group);
6280
6281         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6282         if (ret > 0)
6283                 ret = -EIO;
6284         if (ret < 0)
6285                 goto out;
6286
6287         ret = btrfs_del_item(trans, root, path);
6288 out:
6289         btrfs_free_path(path);
6290         return ret;
6291 }