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