fs/btrfs/extent-tree.c

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