Pileus Git - ~andy/linux/blob - fs/btrfs/extent-tree.c

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