fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include "hash.h"
  23 #include "crc32c.h"
  24 #include "ctree.h"
  25 #include "disk-io.h"
  26 #include "print-tree.h"
  27 #include "transaction.h"
  28 #include "volumes.h"
  29 #include "locking.h"
  30 #include "ref-cache.h"
  31
  32 #define PENDING_EXTENT_INSERT 0
  33 #define PENDING_EXTENT_DELETE 1
  34 #define PENDING_BACKREF_UPDATE 2
  35
  36 struct pending_extent_op {
  37         int type;
  38         u64 bytenr;
  39         u64 num_bytes;
  40         u64 parent;
  41         u64 orig_parent;
  42         u64 generation;
  43         u64 orig_generation;
  44         int level;
  45         struct list_head list;
  46         int del;
  47 };
  48
  49 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  50                                  btrfs_root *extent_root, int all);
  51 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  52                                btrfs_root *extent_root, int all);
  53 static struct btrfs_block_group_cache *
  54 __btrfs_find_block_group(struct btrfs_root *root,
  55                          struct btrfs_block_group_cache *hint,
  56                          u64 search_start, int data, int owner);
  57 static int pin_down_bytes(struct btrfs_trans_handle *trans,
  58                           struct btrfs_root *root,
  59                           u64 bytenr, u64 num_bytes, int is_data);
  60 static int update_block_group(struct btrfs_trans_handle *trans,
  61                               struct btrfs_root *root,
  62                               u64 bytenr, u64 num_bytes, int alloc,
  63                               int mark_free);
  64
  65 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  66 {
  67         return (cache->flags & bits) == bits;
  68 }
  69
  70 /*
  71  * this adds the block group to the fs_info rb tree for the block group
  72  * cache
  73  */
  74 int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  75                                 struct btrfs_block_group_cache *block_group)
  76 {
  77         struct rb_node **p;
  78         struct rb_node *parent = NULL;
  79         struct btrfs_block_group_cache *cache;
  80
  81         spin_lock(&info->block_group_cache_lock);
  82         p = &info->block_group_cache_tree.rb_node;
  83
  84         while (*p) {
  85                 parent = *p;
  86                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  87                                  cache_node);
  88                 if (block_group->key.objectid < cache->key.objectid) {
  89                         p = &(*p)->rb_left;
  90                 } else if (block_group->key.objectid > cache->key.objectid) {
  91                         p = &(*p)->rb_right;
  92                 } else {
  93                         spin_unlock(&info->block_group_cache_lock);
  94                         return -EEXIST;
  95                 }
  96         }
  97
  98         rb_link_node(&block_group->cache_node, parent, p);
  99         rb_insert_color(&block_group->cache_node,
 100                         &info->block_group_cache_tree);
 101         spin_unlock(&info->block_group_cache_lock);
 102
 103         return 0;
 104 }
 105
 106 /*
 107  * This will return the block group at or after bytenr if contains is 0, else
 108  * it will return the block group that contains the bytenr
 109  */
 110 static struct btrfs_block_group_cache *
 111 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 112                               int contains)
 113 {
 114         struct btrfs_block_group_cache *cache, *ret = NULL;
 115         struct rb_node *n;
 116         u64 end, start;
 117
 118         spin_lock(&info->block_group_cache_lock);
 119         n = info->block_group_cache_tree.rb_node;
 120
 121         while (n) {
 122                 cache = rb_entry(n, struct btrfs_block_group_cache,
 123                                  cache_node);
 124                 end = cache->key.objectid + cache->key.offset - 1;
 125                 start = cache->key.objectid;
 126
 127                 if (bytenr < start) {
 128                         if (!contains && (!ret || start < ret->key.objectid))
 129                                 ret = cache;
 130                         n = n->rb_left;
 131                 } else if (bytenr > start) {
 132                         if (contains && bytenr <= end) {
 133                                 ret = cache;
 134                                 break;
 135                         }
 136                         n = n->rb_right;
 137                 } else {
 138                         ret = cache;
 139                         break;
 140                 }
 141         }
 142         spin_unlock(&info->block_group_cache_lock);
 143
 144         return ret;
 145 }
 146
 147 /*
 148  * this is only called by cache_block_group, since we could have freed extents
 149  * we need to check the pinned_extents for any extents that can't be used yet
 150  * since their free space will be released as soon as the transaction commits.
 151  */
 152 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 153                               struct btrfs_fs_info *info, u64 start, u64 end)
 154 {
 155         u64 extent_start, extent_end, size;
 156         int ret;
 157
 158         mutex_lock(&info->pinned_mutex);
 159         while (start < end) {
 160                 ret = find_first_extent_bit(&info->pinned_extents, start,
 161                                             &extent_start, &extent_end,
 162                                             EXTENT_DIRTY);
 163                 if (ret)
 164                         break;
 165
 166                 if (extent_start == start) {
 167                         start = extent_end + 1;
 168                 } else if (extent_start > start && extent_start < end) {
 169                         size = extent_start - start;
 170                         ret = btrfs_add_free_space_lock(block_group, start,
 171                                                         size);
 172                         BUG_ON(ret);
 173                         start = extent_end + 1;
 174                 } else {
 175                         break;
 176                 }
 177         }
 178
 179         if (start < end) {
 180                 size = end - start;
 181                 ret = btrfs_add_free_space_lock(block_group, start, size);
 182                 BUG_ON(ret);
 183         }
 184         mutex_unlock(&info->pinned_mutex);
 185
 186         return 0;
 187 }
 188
 189 static int cache_block_group(struct btrfs_root *root,
 190                              struct btrfs_block_group_cache *block_group)
 191 {
 192         struct btrfs_path *path;
 193         int ret = 0;
 194         struct btrfs_key key;
 195         struct extent_buffer *leaf;
 196         int slot;
 197         u64 last = 0;
 198         u64 first_free;
 199         int found = 0;
 200
 201         if (!block_group)
 202                 return 0;
 203
 204         root = root->fs_info->extent_root;
 205
 206         if (block_group->cached)
 207                 return 0;
 208
 209         path = btrfs_alloc_path();
 210         if (!path)
 211                 return -ENOMEM;
 212
 213         path->reada = 2;
 214         /*
 215          * we get into deadlocks with paths held by callers of this function.
 216          * since the alloc_mutex is protecting things right now, just
 217          * skip the locking here
 218          */
 219         path->skip_locking = 1;
 220         first_free = max_t(u64, block_group->key.objectid,
 221                            BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 222         key.objectid = block_group->key.objectid;
 223         key.offset = 0;
 224         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 225         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 226         if (ret < 0)
 227                 goto err;
 228         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
 229         if (ret < 0)
 230                 goto err;
 231         if (ret == 0) {
 232                 leaf = path->nodes[0];
 233                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 234                 if (key.objectid + key.offset > first_free)
 235                         first_free = key.objectid + key.offset;
 236         }
 237         while(1) {
 238                 leaf = path->nodes[0];
 239                 slot = path->slots[0];
 240                 if (slot >= btrfs_header_nritems(leaf)) {
 241                         ret = btrfs_next_leaf(root, path);
 242                         if (ret < 0)
 243                                 goto err;
 244                         if (ret == 0)
 245                                 continue;
 246                         else
 247                                 break;
 248                 }
 249                 btrfs_item_key_to_cpu(leaf, &key, slot);
 250                 if (key.objectid < block_group->key.objectid)
 251                         goto next;
 252
 253                 if (key.objectid >= block_group->key.objectid +
 254                     block_group->key.offset)
 255                         break;
 256
 257                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 258                         if (!found) {
 259                                 last = first_free;
 260                                 found = 1;
 261                         }
 262
 263                         add_new_free_space(block_group, root->fs_info, last,
 264                                            key.objectid);
 265
 266                         last = key.objectid + key.offset;
 267                 }
 268 next:
 269                 path->slots[0]++;
 270         }
 271
 272         if (!found)
 273                 last = first_free;
 274
 275         add_new_free_space(block_group, root->fs_info, last,
 276                            block_group->key.objectid +
 277                            block_group->key.offset);
 278
 279         block_group->cached = 1;
 280         ret = 0;
 281 err:
 282         btrfs_free_path(path);
 283         return ret;
 284 }
 285
 286 /*
 287  * return the block group that starts at or after bytenr
 288  */
 289 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 290                                                        btrfs_fs_info *info,
 291                                                          u64 bytenr)
 292 {
 293         struct btrfs_block_group_cache *cache;
 294
 295         cache = block_group_cache_tree_search(info, bytenr, 0);
 296
 297         return cache;
 298 }
 299
 300 /*
 301  * return the block group that contains teh given bytenr
 302  */
 303 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 304                                                          btrfs_fs_info *info,
 305                                                          u64 bytenr)
 306 {
 307         struct btrfs_block_group_cache *cache;
 308
 309         cache = block_group_cache_tree_search(info, bytenr, 1);
 310
 311         return cache;
 312 }
 313
 314 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 315                                                   u64 flags)
 316 {
 317         struct list_head *head = &info->space_info;
 318         struct list_head *cur;
 319         struct btrfs_space_info *found;
 320         list_for_each(cur, head) {
 321                 found = list_entry(cur, struct btrfs_space_info, list);
 322                 if (found->flags == flags)
 323                         return found;
 324         }
 325         return NULL;
 326 }
 327
 328 static u64 div_factor(u64 num, int factor)
 329 {
 330         if (factor == 10)
 331                 return num;
 332         num *= factor;
 333         do_div(num, 10);
 334         return num;
 335 }
 336
 337 static struct btrfs_block_group_cache *
 338 __btrfs_find_block_group(struct btrfs_root *root,
 339                          struct btrfs_block_group_cache *hint,
 340                          u64 search_start, int data, int owner)
 341 {
 342         struct btrfs_block_group_cache *cache;
 343         struct btrfs_block_group_cache *found_group = NULL;
 344         struct btrfs_fs_info *info = root->fs_info;
 345         u64 used;
 346         u64 last = 0;
 347         u64 free_check;
 348         int full_search = 0;
 349         int factor = 10;
 350         int wrapped = 0;
 351
 352         if (data & BTRFS_BLOCK_GROUP_METADATA)
 353                 factor = 9;
 354
 355         if (search_start) {
 356                 struct btrfs_block_group_cache *shint;
 357                 shint = btrfs_lookup_first_block_group(info, search_start);
 358                 if (shint && block_group_bits(shint, data)) {
 359                         spin_lock(&shint->lock);
 360                         used = btrfs_block_group_used(&shint->item);
 361                         if (used + shint->pinned + shint->reserved <
 362                             div_factor(shint->key.offset, factor)) {
 363                                 spin_unlock(&shint->lock);
 364                                 return shint;
 365                         }
 366                         spin_unlock(&shint->lock);
 367                 }
 368         }
 369         if (hint && block_group_bits(hint, data)) {
 370                 spin_lock(&hint->lock);
 371                 used = btrfs_block_group_used(&hint->item);
 372                 if (used + hint->pinned + hint->reserved <
 373                     div_factor(hint->key.offset, factor)) {
 374                         spin_unlock(&hint->lock);
 375                         return hint;
 376                 }
 377                 spin_unlock(&hint->lock);
 378                 last = hint->key.objectid + hint->key.offset;
 379         } else {
 380                 if (hint)
 381                         last = max(hint->key.objectid, search_start);
 382                 else
 383                         last = search_start;
 384         }
 385 again:
 386         while (1) {
 387                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 388                 if (!cache)
 389                         break;
 390
 391                 spin_lock(&cache->lock);
 392                 last = cache->key.objectid + cache->key.offset;
 393                 used = btrfs_block_group_used(&cache->item);
 394
 395                 if (block_group_bits(cache, data)) {
 396                         free_check = div_factor(cache->key.offset, factor);
 397                         if (used + cache->pinned + cache->reserved <
 398                             free_check) {
 399                                 found_group = cache;
 400                                 spin_unlock(&cache->lock);
 401                                 goto found;
 402                         }
 403                 }
 404                 spin_unlock(&cache->lock);
 405                 cond_resched();
 406         }
 407         if (!wrapped) {
 408                 last = search_start;
 409                 wrapped = 1;
 410                 goto again;
 411         }
 412         if (!full_search && factor < 10) {
 413                 last = search_start;
 414                 full_search = 1;
 415                 factor = 10;
 416                 goto again;
 417         }
 418 found:
 419         return found_group;
 420 }
 421
 422 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 423                                                  struct btrfs_block_group_cache
 424                                                  *hint, u64 search_start,
 425                                                  int data, int owner)
 426 {
 427
 428         struct btrfs_block_group_cache *ret;
 429         ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
 430         return ret;
 431 }
 432
 433 /* simple helper to search for an existing extent at a given offset */
 434 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 435 {
 436         int ret;
 437         struct btrfs_key key;
 438         struct btrfs_path *path;
 439
 440         path = btrfs_alloc_path();
 441         BUG_ON(!path);
 442         key.objectid = start;
 443         key.offset = len;
 444         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 445         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 446                                 0, 0);
 447         btrfs_free_path(path);
 448         return ret;
 449 }
 450
 451 /*
 452  * Back reference rules.  Back refs have three main goals:
 453  *
 454  * 1) differentiate between all holders of references to an extent so that
 455  *    when a reference is dropped we can make sure it was a valid reference
 456  *    before freeing the extent.
 457  *
 458  * 2) Provide enough information to quickly find the holders of an extent
 459  *    if we notice a given block is corrupted or bad.
 460  *
 461  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 462  *    maintenance.  This is actually the same as #2, but with a slightly
 463  *    different use case.
 464  *
 465  * File extents can be referenced by:
 466  *
 467  * - multiple snapshots, subvolumes, or different generations in one subvol
 468  * - different files inside a single subvolume
 469  * - different offsets inside a file (bookend extents in file.c)
 470  *
 471  * The extent ref structure has fields for:
 472  *
 473  * - Objectid of the subvolume root
 474  * - Generation number of the tree holding the reference
 475  * - objectid of the file holding the reference
 476  * - number of references holding by parent node (alway 1 for tree blocks)
 477  *
 478  * Btree leaf may hold multiple references to a file extent. In most cases,
 479  * these references are from same file and the corresponding offsets inside
 480  * the file are close together.
 481  *
 482  * When a file extent is allocated the fields are filled in:
 483  *     (root_key.objectid, trans->transid, inode objectid, 1)
 484  *
 485  * When a leaf is cow'd new references are added for every file extent found
 486  * in the leaf.  It looks similar to the create case, but trans->transid will
 487  * be different when the block is cow'd.
 488  *
 489  *     (root_key.objectid, trans->transid, inode objectid,
 490  *      number of references in the leaf)
 491  *
 492  * When a file extent is removed either during snapshot deletion or
 493  * file truncation, we find the corresponding back reference and check
 494  * the following fields:
 495  *
 496  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 497  *      inode objectid)
 498  *
 499  * Btree extents can be referenced by:
 500  *
 501  * - Different subvolumes
 502  * - Different generations of the same subvolume
 503  *
 504  * When a tree block is created, back references are inserted:
 505  *
 506  * (root->root_key.objectid, trans->transid, level, 1)
 507  *
 508  * When a tree block is cow'd, new back references are added for all the
 509  * blocks it points to. If the tree block isn't in reference counted root,
 510  * the old back references are removed. These new back references are of
 511  * the form (trans->transid will have increased since creation):
 512  *
 513  * (root->root_key.objectid, trans->transid, level, 1)
 514  *
 515  * When a backref is in deleting, the following fields are checked:
 516  *
 517  * if backref was for a tree root:
 518  *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
 519  * else
 520  *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
 521  *
 522  * Back Reference Key composing:
 523  *
 524  * The key objectid corresponds to the first byte in the extent, the key
 525  * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
 526  * byte of parent extent. If a extent is tree root, the key offset is set
 527  * to the key objectid.
 528  */
 529
 530 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 531                                           struct btrfs_root *root,
 532                                           struct btrfs_path *path,
 533                                           u64 bytenr, u64 parent,
 534                                           u64 ref_root, u64 ref_generation,
 535                                           u64 owner_objectid, int del)
 536 {
 537         struct btrfs_key key;
 538         struct btrfs_extent_ref *ref;
 539         struct extent_buffer *leaf;
 540         u64 ref_objectid;
 541         int ret;
 542
 543         key.objectid = bytenr;
 544         key.type = BTRFS_EXTENT_REF_KEY;
 545         key.offset = parent;
 546
 547         ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
 548         if (ret < 0)
 549                 goto out;
 550         if (ret > 0) {
 551                 ret = -ENOENT;
 552                 goto out;
 553         }
 554
 555         leaf = path->nodes[0];
 556         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 557         ref_objectid = btrfs_ref_objectid(leaf, ref);
 558         if (btrfs_ref_root(leaf, ref) != ref_root ||
 559             btrfs_ref_generation(leaf, ref) != ref_generation ||
 560             (ref_objectid != owner_objectid &&
 561              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 562                 ret = -EIO;
 563                 WARN_ON(1);
 564                 goto out;
 565         }
 566         ret = 0;
 567 out:
 568         return ret;
 569 }
 570
 571 /*
 572  * updates all the backrefs that are pending on update_list for the
 573  * extent_root
 574  */
 575 static int noinline update_backrefs(struct btrfs_trans_handle *trans,
 576                                     struct btrfs_root *extent_root,
 577                                     struct btrfs_path *path,
 578                                     struct list_head *update_list)
 579 {
 580         struct btrfs_key key;
 581         struct btrfs_extent_ref *ref;
 582         struct btrfs_fs_info *info = extent_root->fs_info;
 583         struct pending_extent_op *op;
 584         struct extent_buffer *leaf;
 585         int ret = 0;
 586         struct list_head *cur = update_list->next;
 587         u64 ref_objectid;
 588         u64 ref_root = extent_root->root_key.objectid;
 589
 590         op = list_entry(cur, struct pending_extent_op, list);
 591
 592 search:
 593         key.objectid = op->bytenr;
 594         key.type = BTRFS_EXTENT_REF_KEY;
 595         key.offset = op->orig_parent;
 596
 597         ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
 598         BUG_ON(ret);
 599
 600         leaf = path->nodes[0];
 601
 602 loop:
 603         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 604
 605         ref_objectid = btrfs_ref_objectid(leaf, ref);
 606
 607         if (btrfs_ref_root(leaf, ref) != ref_root ||
 608             btrfs_ref_generation(leaf, ref) != op->orig_generation ||
 609             (ref_objectid != op->level &&
 610              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 611                 printk(KERN_ERR "couldn't find %Lu, parent %Lu, root %Lu, "
 612                        "owner %u\n", op->bytenr, op->orig_parent,
 613                        ref_root, op->level);
 614                 btrfs_print_leaf(extent_root, leaf);
 615                 BUG();
 616         }
 617
 618         key.objectid = op->bytenr;
 619         key.offset = op->parent;
 620         key.type = BTRFS_EXTENT_REF_KEY;
 621         ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
 622         BUG_ON(ret);
 623         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 624         btrfs_set_ref_generation(leaf, ref, op->generation);
 625
 626         cur = cur->next;
 627
 628         list_del_init(&op->list);
 629         unlock_extent(&info->extent_ins, op->bytenr,
 630                       op->bytenr + op->num_bytes - 1, GFP_NOFS);
 631         kfree(op);
 632
 633         if (cur == update_list) {
 634                 btrfs_mark_buffer_dirty(path->nodes[0]);
 635                 btrfs_release_path(extent_root, path);
 636                 goto out;
 637         }
 638
 639         op = list_entry(cur, struct pending_extent_op, list);
 640
 641         path->slots[0]++;
 642         while (path->slots[0] < btrfs_header_nritems(leaf)) {
 643                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 644                 if (key.objectid == op->bytenr &&
 645                     key.type == BTRFS_EXTENT_REF_KEY)
 646                         goto loop;
 647                 path->slots[0]++;
 648         }
 649
 650         btrfs_mark_buffer_dirty(path->nodes[0]);
 651         btrfs_release_path(extent_root, path);
 652         goto search;
 653
 654 out:
 655         return 0;
 656 }
 657
 658 static int noinline insert_extents(struct btrfs_trans_handle *trans,
 659                                    struct btrfs_root *extent_root,
 660                                    struct btrfs_path *path,
 661                                    struct list_head *insert_list, int nr)
 662 {
 663         struct btrfs_key *keys;
 664         u32 *data_size;
 665         struct pending_extent_op *op;
 666         struct extent_buffer *leaf;
 667         struct list_head *cur = insert_list->next;
 668         struct btrfs_fs_info *info = extent_root->fs_info;
 669         u64 ref_root = extent_root->root_key.objectid;
 670         int i = 0, last = 0, ret;
 671         int total = nr * 2;
 672
 673         if (!nr)
 674                 return 0;
 675
 676         keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
 677         if (!keys)
 678                 return -ENOMEM;
 679
 680         data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
 681         if (!data_size) {
 682                 kfree(keys);
 683                 return -ENOMEM;
 684         }
 685
 686         list_for_each_entry(op, insert_list, list) {
 687                 keys[i].objectid = op->bytenr;
 688                 keys[i].offset = op->num_bytes;
 689                 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
 690                 data_size[i] = sizeof(struct btrfs_extent_item);
 691                 i++;
 692
 693                 keys[i].objectid = op->bytenr;
 694                 keys[i].offset = op->parent;
 695                 keys[i].type = BTRFS_EXTENT_REF_KEY;
 696                 data_size[i] = sizeof(struct btrfs_extent_ref);
 697                 i++;
 698         }
 699
 700         op = list_entry(cur, struct pending_extent_op, list);
 701         i = 0;
 702         while (i < total) {
 703                 int c;
 704                 ret = btrfs_insert_some_items(trans, extent_root, path,
 705                                               keys+i, data_size+i, total-i);
 706                 BUG_ON(ret < 0);
 707
 708                 if (last && ret > 1)
 709                         BUG();
 710
 711                 leaf = path->nodes[0];
 712                 for (c = 0; c < ret; c++) {
 713                         int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
 714
 715                         /*
 716                          * if the first item we inserted was a backref, then
 717                          * the EXTENT_ITEM will be the odd c's, else it will
 718                          * be the even c's
 719                          */
 720                         if ((ref_first && (c % 2)) ||
 721                             (!ref_first && !(c % 2))) {
 722                                 struct btrfs_extent_item *itm;
 723
 724                                 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
 725                                                      struct btrfs_extent_item);
 726                                 btrfs_set_extent_refs(path->nodes[0], itm, 1);
 727                                 op->del++;
 728                         } else {
 729                                 struct btrfs_extent_ref *ref;
 730
 731                                 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
 732                                                      struct btrfs_extent_ref);
 733                                 btrfs_set_ref_root(leaf, ref, ref_root);
 734                                 btrfs_set_ref_generation(leaf, ref,
 735                                                          op->generation);
 736                                 btrfs_set_ref_objectid(leaf, ref, op->level);
 737                                 btrfs_set_ref_num_refs(leaf, ref, 1);
 738                                 op->del++;
 739                         }
 740
 741                         /*
 742                          * using del to see when its ok to free up the
 743                          * pending_extent_op.  In the case where we insert the
 744                          * last item on the list in order to help do batching
 745                          * we need to not free the extent op until we actually
 746                          * insert the extent_item
 747                          */
 748                         if (op->del == 2) {
 749                                 unlock_extent(&info->extent_ins, op->bytenr,
 750                                               op->bytenr + op->num_bytes - 1,
 751                                               GFP_NOFS);
 752                                 cur = cur->next;
 753                                 list_del_init(&op->list);
 754                                 kfree(op);
 755                                 if (cur != insert_list)
 756                                         op = list_entry(cur,
 757                                                 struct pending_extent_op,
 758                                                 list);
 759                         }
 760                 }
 761                 btrfs_mark_buffer_dirty(leaf);
 762                 btrfs_release_path(extent_root, path);
 763
 764                 /*
 765                  * Ok backref's and items usually go right next to eachother,
 766                  * but if we could only insert 1 item that means that we
 767                  * inserted on the end of a leaf, and we have no idea what may
 768                  * be on the next leaf so we just play it safe.  In order to
 769                  * try and help this case we insert the last thing on our
 770                  * insert list so hopefully it will end up being the last
 771                  * thing on the leaf and everything else will be before it,
 772                  * which will let us insert a whole bunch of items at the same
 773                  * time.
 774                  */
 775                 if (ret == 1 && !last && (i + ret < total)) {
 776                         /*
 777                          * last: where we will pick up the next time around
 778                          * i: our current key to insert, will be total - 1
 779                          * cur: the current op we are screwing with
 780                          * op: duh
 781                          */
 782                         last = i + ret;
 783                         i = total - 1;
 784                         cur = insert_list->prev;
 785                         op = list_entry(cur, struct pending_extent_op, list);
 786                 } else if (last) {
 787                         /*
 788                          * ok we successfully inserted the last item on the
 789                          * list, lets reset everything
 790                          *
 791                          * i: our current key to insert, so where we left off
 792                          *    last time
 793                          * last: done with this
 794                          * cur: the op we are messing with
 795                          * op: duh
 796                          * total: since we inserted the last key, we need to
 797                          *        decrement total so we dont overflow
 798                          */
 799                         i = last;
 800                         last = 0;
 801                         cur = insert_list->next;
 802                         op = list_entry(cur, struct pending_extent_op, list);
 803                         total--;
 804                 } else {
 805                         i += ret;
 806                 }
 807
 808                 cond_resched();
 809         }
 810         ret = 0;
 811         kfree(keys);
 812         kfree(data_size);
 813         return ret;
 814 }
 815
 816 static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
 817                                           struct btrfs_root *root,
 818                                           struct btrfs_path *path,
 819                                           u64 bytenr, u64 parent,
 820                                           u64 ref_root, u64 ref_generation,
 821                                           u64 owner_objectid)
 822 {
 823         struct btrfs_key key;
 824         struct extent_buffer *leaf;
 825         struct btrfs_extent_ref *ref;
 826         u32 num_refs;
 827         int ret;
 828
 829         key.objectid = bytenr;
 830         key.type = BTRFS_EXTENT_REF_KEY;
 831         key.offset = parent;
 832
 833         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
 834         if (ret == 0) {
 835                 leaf = path->nodes[0];
 836                 ref = btrfs_item_ptr(leaf, path->slots[0],
 837                                      struct btrfs_extent_ref);
 838                 btrfs_set_ref_root(leaf, ref, ref_root);
 839                 btrfs_set_ref_generation(leaf, ref, ref_generation);
 840                 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
 841                 btrfs_set_ref_num_refs(leaf, ref, 1);
 842         } else if (ret == -EEXIST) {
 843                 u64 existing_owner;
 844                 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
 845                 leaf = path->nodes[0];
 846                 ref = btrfs_item_ptr(leaf, path->slots[0],
 847                                      struct btrfs_extent_ref);
 848                 if (btrfs_ref_root(leaf, ref) != ref_root ||
 849                     btrfs_ref_generation(leaf, ref) != ref_generation) {
 850                         ret = -EIO;
 851                         WARN_ON(1);
 852                         goto out;
 853                 }
 854
 855                 num_refs = btrfs_ref_num_refs(leaf, ref);
 856                 BUG_ON(num_refs == 0);
 857                 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
 858
 859                 existing_owner = btrfs_ref_objectid(leaf, ref);
 860                 if (existing_owner != owner_objectid &&
 861                     existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
 862                         btrfs_set_ref_objectid(leaf, ref,
 863                                         BTRFS_MULTIPLE_OBJECTIDS);
 864                 }
 865                 ret = 0;
 866         } else {
 867                 goto out;
 868         }
 869         btrfs_mark_buffer_dirty(path->nodes[0]);
 870 out:
 871         btrfs_release_path(root, path);
 872         return ret;
 873 }
 874
 875 static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
 876                                           struct btrfs_root *root,
 877                                           struct btrfs_path *path)
 878 {
 879         struct extent_buffer *leaf;
 880         struct btrfs_extent_ref *ref;
 881         u32 num_refs;
 882         int ret = 0;
 883
 884         leaf = path->nodes[0];
 885         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 886         num_refs = btrfs_ref_num_refs(leaf, ref);
 887         BUG_ON(num_refs == 0);
 888         num_refs -= 1;
 889         if (num_refs == 0) {
 890                 ret = btrfs_del_item(trans, root, path);
 891         } else {
 892                 btrfs_set_ref_num_refs(leaf, ref, num_refs);
 893                 btrfs_mark_buffer_dirty(leaf);
 894         }
 895         btrfs_release_path(root, path);
 896         return ret;
 897 }
 898
 899 static int noinline free_extents(struct btrfs_trans_handle *trans,
 900                                  struct btrfs_root *extent_root,
 901                                  struct list_head *del_list)
 902 {
 903         struct btrfs_fs_info *info = extent_root->fs_info;
 904         struct btrfs_path *path;
 905         struct btrfs_key key, found_key;
 906         struct extent_buffer *leaf;
 907         struct list_head *cur;
 908         struct pending_extent_op *op;
 909         struct btrfs_extent_item *ei;
 910         int ret, num_to_del, extent_slot = 0, found_extent = 0;
 911         u32 refs;
 912         u64 bytes_freed = 0;
 913
 914         path = btrfs_alloc_path();
 915         if (!path)
 916                 return -ENOMEM;
 917         path->reada = 1;
 918
 919 search:
 920         /* search for the backref for the current ref we want to delete */
 921         cur = del_list->next;
 922         op = list_entry(cur, struct pending_extent_op, list);
 923         ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
 924                                     op->orig_parent,
 925                                     extent_root->root_key.objectid,
 926                                     op->orig_generation, op->level, 1);
 927         if (ret) {
 928                 printk("Unable to find backref byte nr %Lu root %Lu gen %Lu "
 929                        "owner %u\n", op->bytenr,
 930                        extent_root->root_key.objectid, op->orig_generation,
 931                        op->level);
 932                 btrfs_print_leaf(extent_root, path->nodes[0]);
 933                 WARN_ON(1);
 934                 goto out;
 935         }
 936
 937         extent_slot = path->slots[0];
 938         num_to_del = 1;
 939         found_extent = 0;
 940
 941         /*
 942          * if we aren't the first item on the leaf we can move back one and see
 943          * if our ref is right next to our extent item
 944          */
 945         if (likely(extent_slot)) {
 946                 extent_slot--;
 947                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 948                                       extent_slot);
 949                 if (found_key.objectid == op->bytenr &&
 950                     found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 951                     found_key.offset == op->num_bytes) {
 952                         num_to_del++;
 953                         found_extent = 1;
 954                 }
 955         }
 956
 957         /*
 958          * if we didn't find the extent we need to delete the backref and then
 959          * search for the extent item key so we can update its ref count
 960          */
 961         if (!found_extent) {
 962                 key.objectid = op->bytenr;
 963                 key.type = BTRFS_EXTENT_ITEM_KEY;
 964                 key.offset = op->num_bytes;
 965
 966                 ret = remove_extent_backref(trans, extent_root, path);
 967                 BUG_ON(ret);
 968                 btrfs_release_path(extent_root, path);
 969                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
 970                 BUG_ON(ret);
 971                 extent_slot = path->slots[0];
 972         }
 973
 974         /* this is where we update the ref count for the extent */
 975         leaf = path->nodes[0];
 976         ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
 977         refs = btrfs_extent_refs(leaf, ei);
 978         BUG_ON(refs == 0);
 979         refs--;
 980         btrfs_set_extent_refs(leaf, ei, refs);
 981
 982         btrfs_mark_buffer_dirty(leaf);
 983
 984         /*
 985          * This extent needs deleting.  The reason cur_slot is extent_slot +
 986          * num_to_del is because extent_slot points to the slot where the extent
 987          * is, and if the backref was not right next to the extent we will be
 988          * deleting at least 1 item, and will want to start searching at the
 989          * slot directly next to extent_slot.  However if we did find the
 990          * backref next to the extent item them we will be deleting at least 2
 991          * items and will want to start searching directly after the ref slot
 992          */
 993         if (!refs) {
 994                 struct list_head *pos, *n, *end;
 995                 int cur_slot = extent_slot+num_to_del;
 996                 u64 super_used;
 997                 u64 root_used;
 998
 999                 path->slots[0] = extent_slot;
1000                 bytes_freed = op->num_bytes;
1001
1002                 /*
1003                  * we need to see if we can delete multiple things at once, so
1004                  * start looping through the list of extents we are wanting to
1005                  * delete and see if their extent/backref's are right next to
1006                  * eachother and the extents only have 1 ref
1007                  */
1008                 for (pos = cur->next; pos != del_list; pos = pos->next) {
1009                         struct pending_extent_op *tmp;
1010
1011                         tmp = list_entry(pos, struct pending_extent_op, list);
1012
1013                         /* we only want to delete extent+ref at this stage */
1014                         if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1015                                 break;
1016
1017                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1018                         if (found_key.objectid != tmp->bytenr ||
1019                             found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1020                             found_key.offset != tmp->num_bytes)
1021                                 break;
1022
1023                         /* check to make sure this extent only has one ref */
1024                         ei = btrfs_item_ptr(leaf, cur_slot,
1025                                             struct btrfs_extent_item);
1026                         if (btrfs_extent_refs(leaf, ei) != 1)
1027                                 break;
1028
1029                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1030                         if (found_key.objectid != tmp->bytenr ||
1031                             found_key.type != BTRFS_EXTENT_REF_KEY ||
1032                             found_key.offset != tmp->orig_parent)
1033                                 break;
1034
1035                         /*
1036                          * the ref is right next to the extent, we can set the
1037                          * ref count to 0 since we will delete them both now
1038                          */
1039                         btrfs_set_extent_refs(leaf, ei, 0);
1040
1041                         /* pin down the bytes for this extent */
1042                         mutex_lock(&info->pinned_mutex);
1043                         ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1044                                              tmp->num_bytes, tmp->level >=
1045                                              BTRFS_FIRST_FREE_OBJECTID);
1046                         mutex_unlock(&info->pinned_mutex);
1047                         BUG_ON(ret < 0);
1048
1049                         /*
1050                          * use the del field to tell if we need to go ahead and
1051                          * free up the extent when we delete the item or not.
1052                          */
1053                         tmp->del = ret;
1054                         bytes_freed += tmp->num_bytes;
1055
1056                         num_to_del += 2;
1057                         cur_slot += 2;
1058                 }
1059                 end = pos;
1060
1061                 /* update the free space counters */
1062                 spin_lock_irq(&info->delalloc_lock);
1063                 super_used = btrfs_super_bytes_used(&info->super_copy);
1064                 btrfs_set_super_bytes_used(&info->super_copy,
1065                                            super_used - bytes_freed);
1066                 spin_unlock_irq(&info->delalloc_lock);
1067
1068                 root_used = btrfs_root_used(&extent_root->root_item);
1069                 btrfs_set_root_used(&extent_root->root_item,
1070                                     root_used - bytes_freed);
1071
1072                 /* delete the items */
1073                 ret = btrfs_del_items(trans, extent_root, path,
1074                                       path->slots[0], num_to_del);
1075                 BUG_ON(ret);
1076
1077                 /*
1078                  * loop through the extents we deleted and do the cleanup work
1079                  * on them
1080                  */
1081                 for (pos = cur, n = pos->next; pos != end;
1082                      pos = n, n = pos->next) {
1083                         struct pending_extent_op *tmp;
1084 #ifdef BIO_RW_DISCARD
1085                         u64 map_length;
1086                         struct btrfs_multi_bio *multi = NULL;
1087 #endif
1088                         tmp = list_entry(pos, struct pending_extent_op, list);
1089
1090                         /*
1091                          * remember tmp->del tells us wether or not we pinned
1092                          * down the extent
1093                          */
1094                         ret = update_block_group(trans, extent_root,
1095                                                  tmp->bytenr, tmp->num_bytes, 0,
1096                                                  tmp->del);
1097                         BUG_ON(ret);
1098
1099 #ifdef BIO_RW_DISCARD
1100                         ret = btrfs_map_block(&info->mapping_tree, READ,
1101                                               tmp->bytenr, &map_length, &multi,
1102                                               0);
1103                         if (!ret) {
1104                                 struct btrfs_bio_stripe *stripe;
1105                                 int i;
1106
1107                                 stripe = multi->stripe;
1108
1109                                 if (map_length > tmp->num_bytes)
1110                                         map_length = tmp->num_bytes;
1111
1112                                 for (i = 0; i < multi->num_stripes;
1113                                      i++, stripe++)
1114                                         blkdev_issue_discard(stripe->dev->bdev,
1115                                                         stripe->physical >> 9,
1116                                                         map_length >> 9);
1117                                 kfree(multi);
1118                         }
1119 #endif
1120                         list_del_init(&tmp->list);
1121                         unlock_extent(&info->extent_ins, tmp->bytenr,
1122                                       tmp->bytenr + tmp->num_bytes - 1,
1123                                       GFP_NOFS);
1124                         kfree(tmp);
1125                 }
1126         } else if (refs && found_extent) {
1127                 /*
1128                  * the ref and extent were right next to eachother, but the
1129                  * extent still has a ref, so just free the backref and keep
1130                  * going
1131                  */
1132                 ret = remove_extent_backref(trans, extent_root, path);
1133                 BUG_ON(ret);
1134
1135                 list_del_init(&op->list);
1136                 unlock_extent(&info->extent_ins, op->bytenr,
1137                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1138                 kfree(op);
1139         } else {
1140                 /*
1141                  * the extent has multiple refs and the backref we were looking
1142                  * for was not right next to it, so just unlock and go next,
1143                  * we're good to go
1144                  */
1145                 list_del_init(&op->list);
1146                 unlock_extent(&info->extent_ins, op->bytenr,
1147                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1148                 kfree(op);
1149         }
1150
1151         btrfs_release_path(extent_root, path);
1152         if (!list_empty(del_list))
1153                 goto search;
1154
1155 out:
1156         btrfs_free_path(path);
1157         return ret;
1158 }
1159
1160 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1161                                      struct btrfs_root *root, u64 bytenr,
1162                                      u64 orig_parent, u64 parent,
1163                                      u64 orig_root, u64 ref_root,
1164                                      u64 orig_generation, u64 ref_generation,
1165                                      u64 owner_objectid)
1166 {
1167         int ret;
1168         struct btrfs_root *extent_root = root->fs_info->extent_root;
1169         struct btrfs_path *path;
1170
1171         if (root == root->fs_info->extent_root) {
1172                 struct pending_extent_op *extent_op;
1173                 u64 num_bytes;
1174
1175                 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1176                 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1177                 mutex_lock(&root->fs_info->extent_ins_mutex);
1178                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1179                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1180                         u64 priv;
1181                         ret = get_state_private(&root->fs_info->extent_ins,
1182                                                 bytenr, &priv);
1183                         BUG_ON(ret);
1184                         extent_op = (struct pending_extent_op *)
1185                                                         (unsigned long)priv;
1186                         BUG_ON(extent_op->parent != orig_parent);
1187                         BUG_ON(extent_op->generation != orig_generation);
1188
1189                         extent_op->parent = parent;
1190                         extent_op->generation = ref_generation;
1191                 } else {
1192                         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1193                         BUG_ON(!extent_op);
1194
1195                         extent_op->type = PENDING_BACKREF_UPDATE;
1196                         extent_op->bytenr = bytenr;
1197                         extent_op->num_bytes = num_bytes;
1198                         extent_op->parent = parent;
1199                         extent_op->orig_parent = orig_parent;
1200                         extent_op->generation = ref_generation;
1201                         extent_op->orig_generation = orig_generation;
1202                         extent_op->level = (int)owner_objectid;
1203                         INIT_LIST_HEAD(&extent_op->list);
1204                         extent_op->del = 0;
1205
1206                         set_extent_bits(&root->fs_info->extent_ins,
1207                                         bytenr, bytenr + num_bytes - 1,
1208                                         EXTENT_WRITEBACK, GFP_NOFS);
1209                         set_state_private(&root->fs_info->extent_ins,
1210                                           bytenr, (unsigned long)extent_op);
1211                 }
1212                 mutex_unlock(&root->fs_info->extent_ins_mutex);
1213                 return 0;
1214         }
1215
1216         path = btrfs_alloc_path();
1217         if (!path)
1218                 return -ENOMEM;
1219         ret = lookup_extent_backref(trans, extent_root, path,
1220                                     bytenr, orig_parent, orig_root,
1221                                     orig_generation, owner_objectid, 1);
1222         if (ret)
1223                 goto out;
1224         ret = remove_extent_backref(trans, extent_root, path);
1225         if (ret)
1226                 goto out;
1227         ret = insert_extent_backref(trans, extent_root, path, bytenr,
1228                                     parent, ref_root, ref_generation,
1229                                     owner_objectid);
1230         BUG_ON(ret);
1231         finish_current_insert(trans, extent_root, 0);
1232         del_pending_extents(trans, extent_root, 0);
1233 out:
1234         btrfs_free_path(path);
1235         return ret;
1236 }
1237
1238 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1239                             struct btrfs_root *root, u64 bytenr,
1240                             u64 orig_parent, u64 parent,
1241                             u64 ref_root, u64 ref_generation,
1242                             u64 owner_objectid)
1243 {
1244         int ret;
1245         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1246             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1247                 return 0;
1248         ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1249                                         parent, ref_root, ref_root,
1250                                         ref_generation, ref_generation,
1251                                         owner_objectid);
1252         return ret;
1253 }
1254
1255 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1256                                   struct btrfs_root *root, u64 bytenr,
1257                                   u64 orig_parent, u64 parent,
1258                                   u64 orig_root, u64 ref_root,
1259                                   u64 orig_generation, u64 ref_generation,
1260                                   u64 owner_objectid)
1261 {
1262         struct btrfs_path *path;
1263         int ret;
1264         struct btrfs_key key;
1265         struct extent_buffer *l;
1266         struct btrfs_extent_item *item;
1267         u32 refs;
1268
1269         path = btrfs_alloc_path();
1270         if (!path)
1271                 return -ENOMEM;
1272
1273         path->reada = 1;
1274         key.objectid = bytenr;
1275         key.type = BTRFS_EXTENT_ITEM_KEY;
1276         key.offset = (u64)-1;
1277
1278         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1279                                 0, 1);
1280         if (ret < 0)
1281                 return ret;
1282         BUG_ON(ret == 0 || path->slots[0] == 0);
1283
1284         path->slots[0]--;
1285         l = path->nodes[0];
1286
1287         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1288         if (key.objectid != bytenr) {
1289                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1290                 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
1291                 BUG();
1292         }
1293         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1294
1295         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1296         refs = btrfs_extent_refs(l, item);
1297         btrfs_set_extent_refs(l, item, refs + 1);
1298         btrfs_mark_buffer_dirty(path->nodes[0]);
1299
1300         btrfs_release_path(root->fs_info->extent_root, path);
1301
1302         path->reada = 1;
1303         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1304                                     path, bytenr, parent,
1305                                     ref_root, ref_generation,
1306                                     owner_objectid);
1307         BUG_ON(ret);
1308         finish_current_insert(trans, root->fs_info->extent_root, 0);
1309         del_pending_extents(trans, root->fs_info->extent_root, 0);
1310
1311         btrfs_free_path(path);
1312         return 0;
1313 }
1314
1315 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1316                          struct btrfs_root *root,
1317                          u64 bytenr, u64 num_bytes, u64 parent,
1318                          u64 ref_root, u64 ref_generation,
1319                          u64 owner_objectid)
1320 {
1321         int ret;
1322         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1323             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1324                 return 0;
1325         ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1326                                      0, ref_root, 0, ref_generation,
1327                                      owner_objectid);
1328         return ret;
1329 }
1330
1331 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1332                          struct btrfs_root *root)
1333 {
1334         finish_current_insert(trans, root->fs_info->extent_root, 1);
1335         del_pending_extents(trans, root->fs_info->extent_root, 1);
1336         return 0;
1337 }
1338
1339 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1340                             struct btrfs_root *root, u64 bytenr,
1341                             u64 num_bytes, u32 *refs)
1342 {
1343         struct btrfs_path *path;
1344         int ret;
1345         struct btrfs_key key;
1346         struct extent_buffer *l;
1347         struct btrfs_extent_item *item;
1348
1349         WARN_ON(num_bytes < root->sectorsize);
1350         path = btrfs_alloc_path();
1351         path->reada = 1;
1352         key.objectid = bytenr;
1353         key.offset = num_bytes;
1354         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1355         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1356                                 0, 0);
1357         if (ret < 0)
1358                 goto out;
1359         if (ret != 0) {
1360                 btrfs_print_leaf(root, path->nodes[0]);
1361                 printk("failed to find block number %Lu\n", bytenr);
1362                 BUG();
1363         }
1364         l = path->nodes[0];
1365         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1366         *refs = btrfs_extent_refs(l, item);
1367 out:
1368         btrfs_free_path(path);
1369         return 0;
1370 }
1371
1372 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1373                           struct btrfs_root *root, u64 bytenr)
1374 {
1375         struct btrfs_root *extent_root = root->fs_info->extent_root;
1376         struct btrfs_path *path;
1377         struct extent_buffer *leaf;
1378         struct btrfs_extent_ref *ref_item;
1379         struct btrfs_key key;
1380         struct btrfs_key found_key;
1381         u64 ref_root;
1382         u64 last_snapshot;
1383         u32 nritems;
1384         int ret;
1385
1386         key.objectid = bytenr;
1387         key.offset = (u64)-1;
1388         key.type = BTRFS_EXTENT_ITEM_KEY;
1389
1390         path = btrfs_alloc_path();
1391         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1392         if (ret < 0)
1393                 goto out;
1394         BUG_ON(ret == 0);
1395
1396         ret = -ENOENT;
1397         if (path->slots[0] == 0)
1398                 goto out;
1399
1400         path->slots[0]--;
1401         leaf = path->nodes[0];
1402         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1403
1404         if (found_key.objectid != bytenr ||
1405             found_key.type != BTRFS_EXTENT_ITEM_KEY)
1406                 goto out;
1407
1408         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1409         while (1) {
1410                 leaf = path->nodes[0];
1411                 nritems = btrfs_header_nritems(leaf);
1412                 if (path->slots[0] >= nritems) {
1413                         ret = btrfs_next_leaf(extent_root, path);
1414                         if (ret < 0)
1415                                 goto out;
1416                         if (ret == 0)
1417                                 continue;
1418                         break;
1419                 }
1420                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1421                 if (found_key.objectid != bytenr)
1422                         break;
1423
1424                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1425                         path->slots[0]++;
1426                         continue;
1427                 }
1428
1429                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1430                                           struct btrfs_extent_ref);
1431                 ref_root = btrfs_ref_root(leaf, ref_item);
1432                 if (ref_root != root->root_key.objectid &&
1433                     ref_root != BTRFS_TREE_LOG_OBJECTID) {
1434                         ret = 1;
1435                         goto out;
1436                 }
1437                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1438                         ret = 1;
1439                         goto out;
1440                 }
1441
1442                 path->slots[0]++;
1443         }
1444         ret = 0;
1445 out:
1446         btrfs_free_path(path);
1447         return ret;
1448 }
1449
1450 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1451                     struct extent_buffer *buf, u32 nr_extents)
1452 {
1453         struct btrfs_key key;
1454         struct btrfs_file_extent_item *fi;
1455         u64 root_gen;
1456         u32 nritems;
1457         int i;
1458         int level;
1459         int ret = 0;
1460         int shared = 0;
1461
1462         if (!root->ref_cows)
1463                 return 0;
1464
1465         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1466                 shared = 0;
1467                 root_gen = root->root_key.offset;
1468         } else {
1469                 shared = 1;
1470                 root_gen = trans->transid - 1;
1471         }
1472
1473         level = btrfs_header_level(buf);
1474         nritems = btrfs_header_nritems(buf);
1475
1476         if (level == 0) {
1477                 struct btrfs_leaf_ref *ref;
1478                 struct btrfs_extent_info *info;
1479
1480                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1481                 if (!ref) {
1482                         ret = -ENOMEM;
1483                         goto out;
1484                 }
1485
1486                 ref->root_gen = root_gen;
1487                 ref->bytenr = buf->start;
1488                 ref->owner = btrfs_header_owner(buf);
1489                 ref->generation = btrfs_header_generation(buf);
1490                 ref->nritems = nr_extents;
1491                 info = ref->extents;
1492
1493                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1494                         u64 disk_bytenr;
1495                         btrfs_item_key_to_cpu(buf, &key, i);
1496                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1497                                 continue;
1498                         fi = btrfs_item_ptr(buf, i,
1499                                             struct btrfs_file_extent_item);
1500                         if (btrfs_file_extent_type(buf, fi) ==
1501                             BTRFS_FILE_EXTENT_INLINE)
1502                                 continue;
1503                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1504                         if (disk_bytenr == 0)
1505                                 continue;
1506
1507                         info->bytenr = disk_bytenr;
1508                         info->num_bytes =
1509                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1510                         info->objectid = key.objectid;
1511                         info->offset = key.offset;
1512                         info++;
1513                 }
1514
1515                 ret = btrfs_add_leaf_ref(root, ref, shared);
1516                 if (ret == -EEXIST && shared) {
1517                         struct btrfs_leaf_ref *old;
1518                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1519                         BUG_ON(!old);
1520                         btrfs_remove_leaf_ref(root, old);
1521                         btrfs_free_leaf_ref(root, old);
1522                         ret = btrfs_add_leaf_ref(root, ref, shared);
1523                 }
1524                 WARN_ON(ret);
1525                 btrfs_free_leaf_ref(root, ref);
1526         }
1527 out:
1528         return ret;
1529 }
1530
1531 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1532                   struct extent_buffer *orig_buf, struct extent_buffer *buf,
1533                   u32 *nr_extents)
1534 {
1535         u64 bytenr;
1536         u64 ref_root;
1537         u64 orig_root;
1538         u64 ref_generation;
1539         u64 orig_generation;
1540         u32 nritems;
1541         u32 nr_file_extents = 0;
1542         struct btrfs_key key;
1543         struct btrfs_file_extent_item *fi;
1544         int i;
1545         int level;
1546         int ret = 0;
1547         int faili = 0;
1548         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1549                             u64, u64, u64, u64, u64, u64, u64, u64);
1550
1551         ref_root = btrfs_header_owner(buf);
1552         ref_generation = btrfs_header_generation(buf);
1553         orig_root = btrfs_header_owner(orig_buf);
1554         orig_generation = btrfs_header_generation(orig_buf);
1555
1556         nritems = btrfs_header_nritems(buf);
1557         level = btrfs_header_level(buf);
1558
1559         if (root->ref_cows) {
1560                 process_func = __btrfs_inc_extent_ref;
1561         } else {
1562                 if (level == 0 &&
1563                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1564                         goto out;
1565                 if (level != 0 &&
1566                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1567                         goto out;
1568                 process_func = __btrfs_update_extent_ref;
1569         }
1570
1571         for (i = 0; i < nritems; i++) {
1572                 cond_resched();
1573                 if (level == 0) {
1574                         btrfs_item_key_to_cpu(buf, &key, i);
1575                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1576                                 continue;
1577                         fi = btrfs_item_ptr(buf, i,
1578                                             struct btrfs_file_extent_item);
1579                         if (btrfs_file_extent_type(buf, fi) ==
1580                             BTRFS_FILE_EXTENT_INLINE)
1581                                 continue;
1582                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1583                         if (bytenr == 0)
1584                                 continue;
1585
1586                         nr_file_extents++;
1587
1588                         ret = process_func(trans, root, bytenr,
1589                                            orig_buf->start, buf->start,
1590                                            orig_root, ref_root,
1591                                            orig_generation, ref_generation,
1592                                            key.objectid);
1593
1594                         if (ret) {
1595                                 faili = i;
1596                                 WARN_ON(1);
1597                                 goto fail;
1598                         }
1599                 } else {
1600                         bytenr = btrfs_node_blockptr(buf, i);
1601                         ret = process_func(trans, root, bytenr,
1602                                            orig_buf->start, buf->start,
1603                                            orig_root, ref_root,
1604                                            orig_generation, ref_generation,
1605                                            level - 1);
1606                         if (ret) {
1607                                 faili = i;
1608                                 WARN_ON(1);
1609                                 goto fail;
1610                         }
1611                 }
1612         }
1613 out:
1614         if (nr_extents) {
1615                 if (level == 0)
1616                         *nr_extents = nr_file_extents;
1617                 else
1618                         *nr_extents = nritems;
1619         }
1620         return 0;
1621 fail:
1622         WARN_ON(1);
1623         return ret;
1624 }
1625
1626 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1627                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1628                      struct extent_buffer *buf, int start_slot, int nr)
1629
1630 {
1631         u64 bytenr;
1632         u64 ref_root;
1633         u64 orig_root;
1634         u64 ref_generation;
1635         u64 orig_generation;
1636         struct btrfs_key key;
1637         struct btrfs_file_extent_item *fi;
1638         int i;
1639         int ret;
1640         int slot;
1641         int level;
1642
1643         BUG_ON(start_slot < 0);
1644         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1645
1646         ref_root = btrfs_header_owner(buf);
1647         ref_generation = btrfs_header_generation(buf);
1648         orig_root = btrfs_header_owner(orig_buf);
1649         orig_generation = btrfs_header_generation(orig_buf);
1650         level = btrfs_header_level(buf);
1651
1652         if (!root->ref_cows) {
1653                 if (level == 0 &&
1654                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1655                         return 0;
1656                 if (level != 0 &&
1657                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1658                         return 0;
1659         }
1660
1661         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1662                 cond_resched();
1663                 if (level == 0) {
1664                         btrfs_item_key_to_cpu(buf, &key, slot);
1665                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1666                                 continue;
1667                         fi = btrfs_item_ptr(buf, slot,
1668                                             struct btrfs_file_extent_item);
1669                         if (btrfs_file_extent_type(buf, fi) ==
1670                             BTRFS_FILE_EXTENT_INLINE)
1671                                 continue;
1672                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1673                         if (bytenr == 0)
1674                                 continue;
1675                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1676                                             orig_buf->start, buf->start,
1677                                             orig_root, ref_root,
1678                                             orig_generation, ref_generation,
1679                                             key.objectid);
1680                         if (ret)
1681                                 goto fail;
1682                 } else {
1683                         bytenr = btrfs_node_blockptr(buf, slot);
1684                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1685                                             orig_buf->start, buf->start,
1686                                             orig_root, ref_root,
1687                                             orig_generation, ref_generation,
1688                                             level - 1);
1689                         if (ret)
1690                                 goto fail;
1691                 }
1692         }
1693         return 0;
1694 fail:
1695         WARN_ON(1);
1696         return -1;
1697 }
1698
1699 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1700                                  struct btrfs_root *root,
1701                                  struct btrfs_path *path,
1702                                  struct btrfs_block_group_cache *cache)
1703 {
1704         int ret;
1705         int pending_ret;
1706         struct btrfs_root *extent_root = root->fs_info->extent_root;
1707         unsigned long bi;
1708         struct extent_buffer *leaf;
1709
1710         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1711         if (ret < 0)
1712                 goto fail;
1713         BUG_ON(ret);
1714
1715         leaf = path->nodes[0];
1716         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1717         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1718         btrfs_mark_buffer_dirty(leaf);
1719         btrfs_release_path(extent_root, path);
1720 fail:
1721         finish_current_insert(trans, extent_root, 0);
1722         pending_ret = del_pending_extents(trans, extent_root, 0);
1723         if (ret)
1724                 return ret;
1725         if (pending_ret)
1726                 return pending_ret;
1727         return 0;
1728
1729 }
1730
1731 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1732                                    struct btrfs_root *root)
1733 {
1734         struct btrfs_block_group_cache *cache, *entry;
1735         struct rb_node *n;
1736         int err = 0;
1737         int werr = 0;
1738         struct btrfs_path *path;
1739         u64 last = 0;
1740
1741         path = btrfs_alloc_path();
1742         if (!path)
1743                 return -ENOMEM;
1744
1745         while(1) {
1746                 cache = NULL;
1747                 spin_lock(&root->fs_info->block_group_cache_lock);
1748                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1749                      n; n = rb_next(n)) {
1750                         entry = rb_entry(n, struct btrfs_block_group_cache,
1751                                          cache_node);
1752                         if (entry->dirty) {
1753                                 cache = entry;
1754                                 break;
1755                         }
1756                 }
1757                 spin_unlock(&root->fs_info->block_group_cache_lock);
1758
1759                 if (!cache)
1760                         break;
1761
1762                 cache->dirty = 0;
1763                 last += cache->key.offset;
1764
1765                 err = write_one_cache_group(trans, root,
1766                                             path, cache);
1767                 /*
1768                  * if we fail to write the cache group, we want
1769                  * to keep it marked dirty in hopes that a later
1770                  * write will work
1771                  */
1772                 if (err) {
1773                         werr = err;
1774                         continue;
1775                 }
1776         }
1777         btrfs_free_path(path);
1778         return werr;
1779 }
1780
1781 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1782                              u64 total_bytes, u64 bytes_used,
1783                              struct btrfs_space_info **space_info)
1784 {
1785         struct btrfs_space_info *found;
1786
1787         found = __find_space_info(info, flags);
1788         if (found) {
1789                 spin_lock(&found->lock);
1790                 found->total_bytes += total_bytes;
1791                 found->bytes_used += bytes_used;
1792                 found->full = 0;
1793                 spin_unlock(&found->lock);
1794                 *space_info = found;
1795                 return 0;
1796         }
1797         found = kzalloc(sizeof(*found), GFP_NOFS);
1798         if (!found)
1799                 return -ENOMEM;
1800
1801         list_add(&found->list, &info->space_info);
1802         INIT_LIST_HEAD(&found->block_groups);
1803         init_rwsem(&found->groups_sem);
1804         spin_lock_init(&found->lock);
1805         found->flags = flags;
1806         found->total_bytes = total_bytes;
1807         found->bytes_used = bytes_used;
1808         found->bytes_pinned = 0;
1809         found->bytes_reserved = 0;
1810         found->bytes_readonly = 0;
1811         found->full = 0;
1812         found->force_alloc = 0;
1813         *space_info = found;
1814         return 0;
1815 }
1816
1817 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1818 {
1819         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1820                                    BTRFS_BLOCK_GROUP_RAID1 |
1821                                    BTRFS_BLOCK_GROUP_RAID10 |
1822                                    BTRFS_BLOCK_GROUP_DUP);
1823         if (extra_flags) {
1824                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1825                         fs_info->avail_data_alloc_bits |= extra_flags;
1826                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1827                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1828                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1829                         fs_info->avail_system_alloc_bits |= extra_flags;
1830         }
1831 }
1832
1833 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1834 {
1835         spin_lock(&cache->space_info->lock);
1836         spin_lock(&cache->lock);
1837         if (!cache->ro) {
1838                 cache->space_info->bytes_readonly += cache->key.offset -
1839                                         btrfs_block_group_used(&cache->item);
1840                 cache->ro = 1;
1841         }
1842         spin_unlock(&cache->lock);
1843         spin_unlock(&cache->space_info->lock);
1844 }
1845
1846 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1847 {
1848         u64 num_devices = root->fs_info->fs_devices->rw_devices;
1849
1850         if (num_devices == 1)
1851                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1852         if (num_devices < 4)
1853                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1854
1855         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1856             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1857                       BTRFS_BLOCK_GROUP_RAID10))) {
1858                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1859         }
1860
1861         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1862             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1863                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1864         }
1865
1866         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1867             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1868              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1869              (flags & BTRFS_BLOCK_GROUP_DUP)))
1870                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1871         return flags;
1872 }
1873
1874 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1875                           struct btrfs_root *extent_root, u64 alloc_bytes,
1876                           u64 flags, int force)
1877 {
1878         struct btrfs_space_info *space_info;
1879         u64 thresh;
1880         int ret = 0;
1881
1882         mutex_lock(&extent_root->fs_info->chunk_mutex);
1883
1884         flags = btrfs_reduce_alloc_profile(extent_root, flags);
1885
1886         space_info = __find_space_info(extent_root->fs_info, flags);
1887         if (!space_info) {
1888                 ret = update_space_info(extent_root->fs_info, flags,
1889                                         0, 0, &space_info);
1890                 BUG_ON(ret);
1891         }
1892         BUG_ON(!space_info);
1893
1894         spin_lock(&space_info->lock);
1895         if (space_info->force_alloc) {
1896                 force = 1;
1897                 space_info->force_alloc = 0;
1898         }
1899         if (space_info->full) {
1900                 spin_unlock(&space_info->lock);
1901                 goto out;
1902         }
1903
1904         thresh = space_info->total_bytes - space_info->bytes_readonly;
1905         thresh = div_factor(thresh, 6);
1906         if (!force &&
1907            (space_info->bytes_used + space_info->bytes_pinned +
1908             space_info->bytes_reserved + alloc_bytes) < thresh) {
1909                 spin_unlock(&space_info->lock);
1910                 goto out;
1911         }
1912         spin_unlock(&space_info->lock);
1913
1914         ret = btrfs_alloc_chunk(trans, extent_root, flags);
1915         if (ret) {
1916 printk("space info full %Lu\n", flags);
1917                 space_info->full = 1;
1918         }
1919 out:
1920         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1921         return ret;
1922 }
1923
1924 static int update_block_group(struct btrfs_trans_handle *trans,
1925                               struct btrfs_root *root,
1926                               u64 bytenr, u64 num_bytes, int alloc,
1927                               int mark_free)
1928 {
1929         struct btrfs_block_group_cache *cache;
1930         struct btrfs_fs_info *info = root->fs_info;
1931         u64 total = num_bytes;
1932         u64 old_val;
1933         u64 byte_in_group;
1934
1935         while(total) {
1936                 cache = btrfs_lookup_block_group(info, bytenr);
1937                 if (!cache)
1938                         return -1;
1939                 byte_in_group = bytenr - cache->key.objectid;
1940                 WARN_ON(byte_in_group > cache->key.offset);
1941
1942                 spin_lock(&cache->space_info->lock);
1943                 spin_lock(&cache->lock);
1944                 cache->dirty = 1;
1945                 old_val = btrfs_block_group_used(&cache->item);
1946                 num_bytes = min(total, cache->key.offset - byte_in_group);
1947                 if (alloc) {
1948                         old_val += num_bytes;
1949                         cache->space_info->bytes_used += num_bytes;
1950                         if (cache->ro) {
1951                                 cache->space_info->bytes_readonly -= num_bytes;
1952                                 WARN_ON(1);
1953                         }
1954                         btrfs_set_block_group_used(&cache->item, old_val);
1955                         spin_unlock(&cache->lock);
1956                         spin_unlock(&cache->space_info->lock);
1957                 } else {
1958                         old_val -= num_bytes;
1959                         cache->space_info->bytes_used -= num_bytes;
1960                         if (cache->ro)
1961                                 cache->space_info->bytes_readonly += num_bytes;
1962                         btrfs_set_block_group_used(&cache->item, old_val);
1963                         spin_unlock(&cache->lock);
1964                         spin_unlock(&cache->space_info->lock);
1965                         if (mark_free) {
1966                                 int ret;
1967                                 ret = btrfs_add_free_space(cache, bytenr,
1968                                                            num_bytes);
1969                                 if (ret)
1970                                         return -1;
1971                         }
1972                 }
1973                 total -= num_bytes;
1974                 bytenr += num_bytes;
1975         }
1976         return 0;
1977 }
1978
1979 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1980 {
1981         struct btrfs_block_group_cache *cache;
1982
1983         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1984         if (!cache)
1985                 return 0;
1986
1987         return cache->key.objectid;
1988 }
1989
1990 int btrfs_update_pinned_extents(struct btrfs_root *root,
1991                                 u64 bytenr, u64 num, int pin)
1992 {
1993         u64 len;
1994         struct btrfs_block_group_cache *cache;
1995         struct btrfs_fs_info *fs_info = root->fs_info;
1996
1997         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
1998         if (pin) {
1999                 set_extent_dirty(&fs_info->pinned_extents,
2000                                 bytenr, bytenr + num - 1, GFP_NOFS);
2001         } else {
2002                 clear_extent_dirty(&fs_info->pinned_extents,
2003                                 bytenr, bytenr + num - 1, GFP_NOFS);
2004         }
2005         while (num > 0) {
2006                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2007                 BUG_ON(!cache);
2008                 len = min(num, cache->key.offset -
2009                           (bytenr - cache->key.objectid));
2010                 if (pin) {
2011                         spin_lock(&cache->space_info->lock);
2012                         spin_lock(&cache->lock);
2013                         cache->pinned += len;
2014                         cache->space_info->bytes_pinned += len;
2015                         spin_unlock(&cache->lock);
2016                         spin_unlock(&cache->space_info->lock);
2017                         fs_info->total_pinned += len;
2018                 } else {
2019                         spin_lock(&cache->space_info->lock);
2020                         spin_lock(&cache->lock);
2021                         cache->pinned -= len;
2022                         cache->space_info->bytes_pinned -= len;
2023                         spin_unlock(&cache->lock);
2024                         spin_unlock(&cache->space_info->lock);
2025                         fs_info->total_pinned -= len;
2026                 }
2027                 bytenr += len;
2028                 num -= len;
2029         }
2030         return 0;
2031 }
2032
2033 static int update_reserved_extents(struct btrfs_root *root,
2034                                    u64 bytenr, u64 num, int reserve)
2035 {
2036         u64 len;
2037         struct btrfs_block_group_cache *cache;
2038         struct btrfs_fs_info *fs_info = root->fs_info;
2039
2040         while (num > 0) {
2041                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2042                 BUG_ON(!cache);
2043                 len = min(num, cache->key.offset -
2044                           (bytenr - cache->key.objectid));
2045
2046                 spin_lock(&cache->space_info->lock);
2047                 spin_lock(&cache->lock);
2048                 if (reserve) {
2049                         cache->reserved += len;
2050                         cache->space_info->bytes_reserved += len;
2051                 } else {
2052                         cache->reserved -= len;
2053                         cache->space_info->bytes_reserved -= len;
2054                 }
2055                 spin_unlock(&cache->lock);
2056                 spin_unlock(&cache->space_info->lock);
2057                 bytenr += len;
2058                 num -= len;
2059         }
2060         return 0;
2061 }
2062
2063 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2064 {
2065         u64 last = 0;
2066         u64 start;
2067         u64 end;
2068         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2069         int ret;
2070
2071         mutex_lock(&root->fs_info->pinned_mutex);
2072         while(1) {
2073                 ret = find_first_extent_bit(pinned_extents, last,
2074                                             &start, &end, EXTENT_DIRTY);
2075                 if (ret)
2076                         break;
2077                 set_extent_dirty(copy, start, end, GFP_NOFS);
2078                 last = end + 1;
2079         }
2080         mutex_unlock(&root->fs_info->pinned_mutex);
2081         return 0;
2082 }
2083
2084 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2085                                struct btrfs_root *root,
2086                                struct extent_io_tree *unpin)
2087 {
2088         u64 start;
2089         u64 end;
2090         int ret;
2091         struct btrfs_block_group_cache *cache;
2092
2093         mutex_lock(&root->fs_info->pinned_mutex);
2094         while(1) {
2095                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2096                                             EXTENT_DIRTY);
2097                 if (ret)
2098                         break;
2099                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2100                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2101                 cache = btrfs_lookup_block_group(root->fs_info, start);
2102                 if (cache->cached)
2103                         btrfs_add_free_space(cache, start, end - start + 1);
2104                 if (need_resched()) {
2105                         mutex_unlock(&root->fs_info->pinned_mutex);
2106                         cond_resched();
2107                         mutex_lock(&root->fs_info->pinned_mutex);
2108                 }
2109         }
2110         mutex_unlock(&root->fs_info->pinned_mutex);
2111         return 0;
2112 }
2113
2114 static int finish_current_insert(struct btrfs_trans_handle *trans,
2115                                  struct btrfs_root *extent_root, int all)
2116 {
2117         u64 start;
2118         u64 end;
2119         u64 priv;
2120         u64 search = 0;
2121         u64 skipped = 0;
2122         struct btrfs_fs_info *info = extent_root->fs_info;
2123         struct btrfs_path *path;
2124         struct pending_extent_op *extent_op, *tmp;
2125         struct list_head insert_list, update_list;
2126         int ret;
2127         int num_inserts = 0, max_inserts;
2128
2129         path = btrfs_alloc_path();
2130         INIT_LIST_HEAD(&insert_list);
2131         INIT_LIST_HEAD(&update_list);
2132
2133         max_inserts = extent_root->leafsize /
2134                 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2135                  sizeof(struct btrfs_extent_ref) +
2136                  sizeof(struct btrfs_extent_item));
2137 again:
2138         mutex_lock(&info->extent_ins_mutex);
2139         while (1) {
2140                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2141                                             &end, EXTENT_WRITEBACK);
2142                 if (ret) {
2143                         if (skipped && all && !num_inserts) {
2144                                 skipped = 0;
2145                                 continue;
2146                         }
2147                         mutex_unlock(&info->extent_ins_mutex);
2148                         break;
2149                 }
2150
2151                 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2152                 if (!ret) {
2153                         skipped = 1;
2154                         search = end + 1;
2155                         if (need_resched()) {
2156                                 mutex_unlock(&info->extent_ins_mutex);
2157                                 cond_resched();
2158                                 mutex_lock(&info->extent_ins_mutex);
2159                         }
2160                         continue;
2161                 }
2162
2163                 ret = get_state_private(&info->extent_ins, start, &priv);
2164                 BUG_ON(ret);
2165                 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2166
2167                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2168                         num_inserts++;
2169                         list_add_tail(&extent_op->list, &insert_list);
2170                         search = end + 1;
2171                         if (num_inserts == max_inserts) {
2172                                 mutex_unlock(&info->extent_ins_mutex);
2173                                 break;
2174                         }
2175                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2176                         list_add_tail(&extent_op->list, &update_list);
2177                         search = end + 1;
2178                 } else {
2179                         BUG();
2180                 }
2181         }
2182
2183         /*
2184          * process teh update list, clear the writeback bit for it, and if
2185          * somebody marked this thing for deletion then just unlock it and be
2186          * done, the free_extents will handle it
2187          */
2188         mutex_lock(&info->extent_ins_mutex);
2189         list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2190                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2191                                   extent_op->bytenr + extent_op->num_bytes - 1,
2192                                   EXTENT_WRITEBACK, GFP_NOFS);
2193                 if (extent_op->del) {
2194                         list_del_init(&extent_op->list);
2195                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2196                                       extent_op->bytenr + extent_op->num_bytes
2197                                       - 1, GFP_NOFS);
2198                         kfree(extent_op);
2199                 }
2200         }
2201         mutex_unlock(&info->extent_ins_mutex);
2202
2203         /*
2204          * still have things left on the update list, go ahead an update
2205          * everything
2206          */
2207         if (!list_empty(&update_list)) {
2208                 ret = update_backrefs(trans, extent_root, path, &update_list);
2209                 BUG_ON(ret);
2210         }
2211
2212         /*
2213          * if no inserts need to be done, but we skipped some extents and we
2214          * need to make sure everything is cleaned then reset everything and
2215          * go back to the beginning
2216          */
2217         if (!num_inserts && all && skipped) {
2218                 search = 0;
2219                 skipped = 0;
2220                 INIT_LIST_HEAD(&update_list);
2221                 INIT_LIST_HEAD(&insert_list);
2222                 goto again;
2223         } else if (!num_inserts) {
2224                 goto out;
2225         }
2226
2227         /*
2228          * process the insert extents list.  Again if we are deleting this
2229          * extent, then just unlock it, pin down the bytes if need be, and be
2230          * done with it.  Saves us from having to actually insert the extent
2231          * into the tree and then subsequently come along and delete it
2232          */
2233         mutex_lock(&info->extent_ins_mutex);
2234         list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2235                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2236                                   extent_op->bytenr + extent_op->num_bytes - 1,
2237                                   EXTENT_WRITEBACK, GFP_NOFS);
2238                 if (extent_op->del) {
2239                         list_del_init(&extent_op->list);
2240                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2241                                       extent_op->bytenr + extent_op->num_bytes
2242                                       - 1, GFP_NOFS);
2243
2244                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2245                         ret = pin_down_bytes(trans, extent_root,
2246                                              extent_op->bytenr,
2247                                              extent_op->num_bytes, 0);
2248                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2249
2250                         ret = update_block_group(trans, extent_root,
2251                                                  extent_op->bytenr,
2252                                                  extent_op->num_bytes,
2253                                                  0, ret > 0);
2254                         BUG_ON(ret);
2255                         kfree(extent_op);
2256                         num_inserts--;
2257                 }
2258         }
2259         mutex_unlock(&info->extent_ins_mutex);
2260
2261         ret = insert_extents(trans, extent_root, path, &insert_list,
2262                              num_inserts);
2263         BUG_ON(ret);
2264
2265         /*
2266          * if we broke out of the loop in order to insert stuff because we hit
2267          * the maximum number of inserts at a time we can handle, then loop
2268          * back and pick up where we left off
2269          */
2270         if (num_inserts == max_inserts) {
2271                 INIT_LIST_HEAD(&insert_list);
2272                 INIT_LIST_HEAD(&update_list);
2273                 num_inserts = 0;
2274                 goto again;
2275         }
2276
2277         /*
2278          * again, if we need to make absolutely sure there are no more pending
2279          * extent operations left and we know that we skipped some, go back to
2280          * the beginning and do it all again
2281          */
2282         if (all && skipped) {
2283                 INIT_LIST_HEAD(&insert_list);
2284                 INIT_LIST_HEAD(&update_list);
2285                 search = 0;
2286                 skipped = 0;
2287                 num_inserts = 0;
2288                 goto again;
2289         }
2290 out:
2291         btrfs_free_path(path);
2292         return 0;
2293 }
2294
2295 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2296                           struct btrfs_root *root,
2297                           u64 bytenr, u64 num_bytes, int is_data)
2298 {
2299         int err = 0;
2300         struct extent_buffer *buf;
2301
2302         if (is_data)
2303                 goto pinit;
2304
2305         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2306         if (!buf)
2307                 goto pinit;
2308
2309         /* we can reuse a block if it hasn't been written
2310          * and it is from this transaction.  We can't
2311          * reuse anything from the tree log root because
2312          * it has tiny sub-transactions.
2313          */
2314         if (btrfs_buffer_uptodate(buf, 0) &&
2315             btrfs_try_tree_lock(buf)) {
2316                 u64 header_owner = btrfs_header_owner(buf);
2317                 u64 header_transid = btrfs_header_generation(buf);
2318                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2319                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2320                     header_transid == trans->transid &&
2321                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2322                         clean_tree_block(NULL, root, buf);
2323                         btrfs_tree_unlock(buf);
2324                         free_extent_buffer(buf);
2325                         return 1;
2326                 }
2327                 btrfs_tree_unlock(buf);
2328         }
2329         free_extent_buffer(buf);
2330 pinit:
2331         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2332
2333         BUG_ON(err < 0);
2334         return 0;
2335 }
2336
2337 /*
2338  * remove an extent from the root, returns 0 on success
2339  */
2340 static int __free_extent(struct btrfs_trans_handle *trans,
2341                          struct btrfs_root *root,
2342                          u64 bytenr, u64 num_bytes, u64 parent,
2343                          u64 root_objectid, u64 ref_generation,
2344                          u64 owner_objectid, int pin, int mark_free)
2345 {
2346         struct btrfs_path *path;
2347         struct btrfs_key key;
2348         struct btrfs_fs_info *info = root->fs_info;
2349         struct btrfs_root *extent_root = info->extent_root;
2350         struct extent_buffer *leaf;
2351         int ret;
2352         int extent_slot = 0;
2353         int found_extent = 0;
2354         int num_to_del = 1;
2355         struct btrfs_extent_item *ei;
2356         u32 refs;
2357
2358         key.objectid = bytenr;
2359         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2360         key.offset = num_bytes;
2361         path = btrfs_alloc_path();
2362         if (!path)
2363                 return -ENOMEM;
2364
2365         path->reada = 1;
2366         ret = lookup_extent_backref(trans, extent_root, path,
2367                                     bytenr, parent, root_objectid,
2368                                     ref_generation, owner_objectid, 1);
2369         if (ret == 0) {
2370                 struct btrfs_key found_key;
2371                 extent_slot = path->slots[0];
2372                 while(extent_slot > 0) {
2373                         extent_slot--;
2374                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2375                                               extent_slot);
2376                         if (found_key.objectid != bytenr)
2377                                 break;
2378                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2379                             found_key.offset == num_bytes) {
2380                                 found_extent = 1;
2381                                 break;
2382                         }
2383                         if (path->slots[0] - extent_slot > 5)
2384                                 break;
2385                 }
2386                 if (!found_extent) {
2387                         ret = remove_extent_backref(trans, extent_root, path);
2388                         BUG_ON(ret);
2389                         btrfs_release_path(extent_root, path);
2390                         ret = btrfs_search_slot(trans, extent_root,
2391                                                 &key, path, -1, 1);
2392                         if (ret) {
2393                                 printk(KERN_ERR "umm, got %d back from search"
2394                                        ", was looking for %Lu\n", ret,
2395                                        bytenr);
2396                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2397                         }
2398                         BUG_ON(ret);
2399                         extent_slot = path->slots[0];
2400                 }
2401         } else {
2402                 btrfs_print_leaf(extent_root, path->nodes[0]);
2403                 WARN_ON(1);
2404                 printk("Unable to find ref byte nr %Lu root %Lu "
2405                        "gen %Lu owner %Lu\n", bytenr,
2406                        root_objectid, ref_generation, owner_objectid);
2407         }
2408
2409         leaf = path->nodes[0];
2410         ei = btrfs_item_ptr(leaf, extent_slot,
2411                             struct btrfs_extent_item);
2412         refs = btrfs_extent_refs(leaf, ei);
2413         BUG_ON(refs == 0);
2414         refs -= 1;
2415         btrfs_set_extent_refs(leaf, ei, refs);
2416
2417         btrfs_mark_buffer_dirty(leaf);
2418
2419         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2420                 struct btrfs_extent_ref *ref;
2421                 ref = btrfs_item_ptr(leaf, path->slots[0],
2422                                      struct btrfs_extent_ref);
2423                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2424                 /* if the back ref and the extent are next to each other
2425                  * they get deleted below in one shot
2426                  */
2427                 path->slots[0] = extent_slot;
2428                 num_to_del = 2;
2429         } else if (found_extent) {
2430                 /* otherwise delete the extent back ref */
2431                 ret = remove_extent_backref(trans, extent_root, path);
2432                 BUG_ON(ret);
2433                 /* if refs are 0, we need to setup the path for deletion */
2434                 if (refs == 0) {
2435                         btrfs_release_path(extent_root, path);
2436                         ret = btrfs_search_slot(trans, extent_root, &key, path,
2437                                                 -1, 1);
2438                         BUG_ON(ret);
2439                 }
2440         }
2441
2442         if (refs == 0) {
2443                 u64 super_used;
2444                 u64 root_used;
2445 #ifdef BIO_RW_DISCARD
2446                 u64 map_length = num_bytes;
2447                 struct btrfs_multi_bio *multi = NULL;
2448 #endif
2449
2450                 if (pin) {
2451                         mutex_lock(&root->fs_info->pinned_mutex);
2452                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2453                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2454                         mutex_unlock(&root->fs_info->pinned_mutex);
2455                         if (ret > 0)
2456                                 mark_free = 1;
2457                         BUG_ON(ret < 0);
2458                 }
2459
2460                 /* block accounting for super block */
2461                 spin_lock_irq(&info->delalloc_lock);
2462                 super_used = btrfs_super_bytes_used(&info->super_copy);
2463                 btrfs_set_super_bytes_used(&info->super_copy,
2464                                            super_used - num_bytes);
2465                 spin_unlock_irq(&info->delalloc_lock);
2466
2467                 /* block accounting for root item */
2468                 root_used = btrfs_root_used(&root->root_item);
2469                 btrfs_set_root_used(&root->root_item,
2470                                            root_used - num_bytes);
2471                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2472                                       num_to_del);
2473                 BUG_ON(ret);
2474                 btrfs_release_path(extent_root, path);
2475                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2476                                          mark_free);
2477                 BUG_ON(ret);
2478
2479 #ifdef BIO_RW_DISCARD
2480                 /* Tell the block device(s) that the sectors can be discarded */
2481                 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2482                                       bytenr, &map_length, &multi, 0);
2483                 if (!ret) {
2484                         struct btrfs_bio_stripe *stripe = multi->stripes;
2485                         int i;
2486
2487                         if (map_length > num_bytes)
2488                                 map_length = num_bytes;
2489
2490                         for (i = 0; i < multi->num_stripes; i++, stripe++) {
2491                                 blkdev_issue_discard(stripe->dev->bdev,
2492                                                      stripe->physical >> 9,
2493                                                      map_length >> 9);
2494                         }
2495                         kfree(multi);
2496                 }
2497 #endif
2498         }
2499         btrfs_free_path(path);
2500         finish_current_insert(trans, extent_root, 0);
2501         return ret;
2502 }
2503
2504 /*
2505  * find all the blocks marked as pending in the radix tree and remove
2506  * them from the extent map
2507  */
2508 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2509                                btrfs_root *extent_root, int all)
2510 {
2511         int ret;
2512         int err = 0;
2513         u64 start;
2514         u64 end;
2515         u64 priv;
2516         u64 search = 0;
2517         int nr = 0, skipped = 0;
2518         struct extent_io_tree *pending_del;
2519         struct extent_io_tree *extent_ins;
2520         struct pending_extent_op *extent_op;
2521         struct btrfs_fs_info *info = extent_root->fs_info;
2522         struct list_head delete_list;
2523
2524         INIT_LIST_HEAD(&delete_list);
2525         extent_ins = &extent_root->fs_info->extent_ins;
2526         pending_del = &extent_root->fs_info->pending_del;
2527
2528 again:
2529         mutex_lock(&info->extent_ins_mutex);
2530         while(1) {
2531                 ret = find_first_extent_bit(pending_del, search, &start, &end,
2532                                             EXTENT_WRITEBACK);
2533                 if (ret) {
2534                         if (all && skipped && !nr) {
2535                                 search = 0;
2536                                 continue;
2537                         }
2538                         mutex_unlock(&info->extent_ins_mutex);
2539                         break;
2540                 }
2541
2542                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2543                 if (!ret) {
2544                         search = end+1;
2545                         skipped = 1;
2546
2547                         if (need_resched()) {
2548                                 mutex_unlock(&info->extent_ins_mutex);
2549                                 cond_resched();
2550                                 mutex_lock(&info->extent_ins_mutex);
2551                         }
2552
2553                         continue;
2554                 }
2555                 BUG_ON(ret < 0);
2556
2557                 ret = get_state_private(pending_del, start, &priv);
2558                 BUG_ON(ret);
2559                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2560
2561                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2562                                   GFP_NOFS);
2563                 if (!test_range_bit(extent_ins, start, end,
2564                                     EXTENT_WRITEBACK, 0)) {
2565                         list_add_tail(&extent_op->list, &delete_list);
2566                         nr++;
2567                 } else {
2568                         kfree(extent_op);
2569
2570                         ret = get_state_private(&info->extent_ins, start,
2571                                                 &priv);
2572                         BUG_ON(ret);
2573                         extent_op = (struct pending_extent_op *)
2574                                                 (unsigned long)priv;
2575
2576                         clear_extent_bits(&info->extent_ins, start, end,
2577                                           EXTENT_WRITEBACK, GFP_NOFS);
2578
2579                         if (extent_op->type == PENDING_BACKREF_UPDATE) {
2580                                 list_add_tail(&extent_op->list, &delete_list);
2581                                 search = end + 1;
2582                                 nr++;
2583                                 continue;
2584                         }
2585
2586                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2587                         ret = pin_down_bytes(trans, extent_root, start,
2588                                              end + 1 - start, 0);
2589                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2590
2591                         ret = update_block_group(trans, extent_root, start,
2592                                                 end + 1 - start, 0, ret > 0);
2593
2594                         unlock_extent(extent_ins, start, end, GFP_NOFS);
2595                         BUG_ON(ret);
2596                         kfree(extent_op);
2597                 }
2598                 if (ret)
2599                         err = ret;
2600
2601                 search = end + 1;
2602
2603                 if (need_resched()) {
2604                         mutex_unlock(&info->extent_ins_mutex);
2605                         cond_resched();
2606                         mutex_lock(&info->extent_ins_mutex);
2607                 }
2608         }
2609
2610         if (nr) {
2611                 ret = free_extents(trans, extent_root, &delete_list);
2612                 BUG_ON(ret);
2613         }
2614
2615         if (all && skipped) {
2616                 INIT_LIST_HEAD(&delete_list);
2617                 search = 0;
2618                 nr = 0;
2619                 goto again;
2620         }
2621
2622         return err;
2623 }
2624
2625 /*
2626  * remove an extent from the root, returns 0 on success
2627  */
2628 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2629                                struct btrfs_root *root,
2630                                u64 bytenr, u64 num_bytes, u64 parent,
2631                                u64 root_objectid, u64 ref_generation,
2632                                u64 owner_objectid, int pin)
2633 {
2634         struct btrfs_root *extent_root = root->fs_info->extent_root;
2635         int pending_ret;
2636         int ret;
2637
2638         WARN_ON(num_bytes < root->sectorsize);
2639         if (root == extent_root) {
2640                 struct pending_extent_op *extent_op = NULL;
2641
2642                 mutex_lock(&root->fs_info->extent_ins_mutex);
2643                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2644                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2645                         u64 priv;
2646                         ret = get_state_private(&root->fs_info->extent_ins,
2647                                                 bytenr, &priv);
2648                         BUG_ON(ret);
2649                         extent_op = (struct pending_extent_op *)
2650                                                 (unsigned long)priv;
2651
2652                         extent_op->del = 1;
2653                         if (extent_op->type == PENDING_EXTENT_INSERT) {
2654                                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2655                                 return 0;
2656                         }
2657                 }
2658
2659                 if (extent_op) {
2660                         ref_generation = extent_op->orig_generation;
2661                         parent = extent_op->orig_parent;
2662                 }
2663
2664                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2665                 BUG_ON(!extent_op);
2666
2667                 extent_op->type = PENDING_EXTENT_DELETE;
2668                 extent_op->bytenr = bytenr;
2669                 extent_op->num_bytes = num_bytes;
2670                 extent_op->parent = parent;
2671                 extent_op->orig_parent = parent;
2672                 extent_op->generation = ref_generation;
2673                 extent_op->orig_generation = ref_generation;
2674                 extent_op->level = (int)owner_objectid;
2675                 INIT_LIST_HEAD(&extent_op->list);
2676                 extent_op->del = 0;
2677
2678                 set_extent_bits(&root->fs_info->pending_del,
2679                                 bytenr, bytenr + num_bytes - 1,
2680                                 EXTENT_WRITEBACK, GFP_NOFS);
2681                 set_state_private(&root->fs_info->pending_del,
2682                                   bytenr, (unsigned long)extent_op);
2683                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2684                 return 0;
2685         }
2686         /* if metadata always pin */
2687         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2688                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2689                         struct btrfs_block_group_cache *cache;
2690
2691                         /* btrfs_free_reserved_extent */
2692                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2693                         BUG_ON(!cache);
2694                         btrfs_add_free_space(cache, bytenr, num_bytes);
2695                         update_reserved_extents(root, bytenr, num_bytes, 0);
2696                         return 0;
2697                 }
2698                 pin = 1;
2699         }
2700
2701         /* if data pin when any transaction has committed this */
2702         if (ref_generation != trans->transid)
2703                 pin = 1;
2704
2705         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2706                             root_objectid, ref_generation,
2707                             owner_objectid, pin, pin == 0);
2708
2709         finish_current_insert(trans, root->fs_info->extent_root, 0);
2710         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2711         return ret ? ret : pending_ret;
2712 }
2713
2714 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2715                       struct btrfs_root *root,
2716                       u64 bytenr, u64 num_bytes, u64 parent,
2717                       u64 root_objectid, u64 ref_generation,
2718                       u64 owner_objectid, int pin)
2719 {
2720         int ret;
2721
2722         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2723                                   root_objectid, ref_generation,
2724                                   owner_objectid, pin);
2725         return ret;
2726 }
2727
2728 static u64 stripe_align(struct btrfs_root *root, u64 val)
2729 {
2730         u64 mask = ((u64)root->stripesize - 1);
2731         u64 ret = (val + mask) & ~mask;
2732         return ret;
2733 }
2734
2735 /*
2736  * walks the btree of allocated extents and find a hole of a given size.
2737  * The key ins is changed to record the hole:
2738  * ins->objectid == block start
2739  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2740  * ins->offset == number of blocks
2741  * Any available blocks before search_start are skipped.
2742  */
2743 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2744                                      struct btrfs_root *orig_root,
2745                                      u64 num_bytes, u64 empty_size,
2746                                      u64 search_start, u64 search_end,
2747                                      u64 hint_byte, struct btrfs_key *ins,
2748                                      u64 exclude_start, u64 exclude_nr,
2749                                      int data)
2750 {
2751         int ret = 0;
2752         struct btrfs_root * root = orig_root->fs_info->extent_root;
2753         u64 total_needed = num_bytes;
2754         u64 *last_ptr = NULL;
2755         u64 last_wanted = 0;
2756         struct btrfs_block_group_cache *block_group = NULL;
2757         int chunk_alloc_done = 0;
2758         int empty_cluster = 2 * 1024 * 1024;
2759         int allowed_chunk_alloc = 0;
2760         struct list_head *head = NULL, *cur = NULL;
2761         int loop = 0;
2762         int extra_loop = 0;
2763         struct btrfs_space_info *space_info;
2764
2765         WARN_ON(num_bytes < root->sectorsize);
2766         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2767         ins->objectid = 0;
2768         ins->offset = 0;
2769
2770         if (orig_root->ref_cows || empty_size)
2771                 allowed_chunk_alloc = 1;
2772
2773         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2774                 last_ptr = &root->fs_info->last_alloc;
2775                 empty_cluster = 64 * 1024;
2776         }
2777
2778         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2779                 last_ptr = &root->fs_info->last_data_alloc;
2780
2781         if (last_ptr) {
2782                 if (*last_ptr) {
2783                         hint_byte = *last_ptr;
2784                         last_wanted = *last_ptr;
2785                 } else
2786                         empty_size += empty_cluster;
2787         } else {
2788                 empty_cluster = 0;
2789         }
2790         search_start = max(search_start, first_logical_byte(root, 0));
2791         search_start = max(search_start, hint_byte);
2792
2793         if (last_wanted && search_start != last_wanted) {
2794                 last_wanted = 0;
2795                 empty_size += empty_cluster;
2796         }
2797
2798         total_needed += empty_size;
2799         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2800         if (!block_group)
2801                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2802                                                              search_start);
2803         space_info = __find_space_info(root->fs_info, data);
2804
2805         down_read(&space_info->groups_sem);
2806         while (1) {
2807                 struct btrfs_free_space *free_space;
2808                 /*
2809                  * the only way this happens if our hint points to a block
2810                  * group thats not of the proper type, while looping this
2811                  * should never happen
2812                  */
2813                 if (empty_size)
2814                         extra_loop = 1;
2815
2816                 if (!block_group)
2817                         goto new_group_no_lock;
2818
2819                 mutex_lock(&block_group->alloc_mutex);
2820                 if (unlikely(!block_group_bits(block_group, data)))
2821                         goto new_group;
2822
2823                 ret = cache_block_group(root, block_group);
2824                 if (ret) {
2825                         mutex_unlock(&block_group->alloc_mutex);
2826                         break;
2827                 }
2828
2829                 if (block_group->ro)
2830                         goto new_group;
2831
2832                 free_space = btrfs_find_free_space(block_group, search_start,
2833                                                    total_needed);
2834                 if (free_space) {
2835                         u64 start = block_group->key.objectid;
2836                         u64 end = block_group->key.objectid +
2837                                 block_group->key.offset;
2838
2839                         search_start = stripe_align(root, free_space->offset);
2840
2841                         /* move on to the next group */
2842                         if (search_start + num_bytes >= search_end)
2843                                 goto new_group;
2844
2845                         /* move on to the next group */
2846                         if (search_start + num_bytes > end)
2847                                 goto new_group;
2848
2849                         if (last_wanted && search_start != last_wanted) {
2850                                 total_needed += empty_cluster;
2851                                 empty_size += empty_cluster;
2852                                 last_wanted = 0;
2853                                 /*
2854                                  * if search_start is still in this block group
2855                                  * then we just re-search this block group
2856                                  */
2857                                 if (search_start >= start &&
2858                                     search_start < end) {
2859                                         mutex_unlock(&block_group->alloc_mutex);
2860                                         continue;
2861                                 }
2862
2863                                 /* else we go to the next block group */
2864                                 goto new_group;
2865                         }
2866
2867                         if (exclude_nr > 0 &&
2868                             (search_start + num_bytes > exclude_start &&
2869                              search_start < exclude_start + exclude_nr)) {
2870                                 search_start = exclude_start + exclude_nr;
2871                                 /*
2872                                  * if search_start is still in this block group
2873                                  * then we just re-search this block group
2874                                  */
2875                                 if (search_start >= start &&
2876                                     search_start < end) {
2877                                         mutex_unlock(&block_group->alloc_mutex);
2878                                         last_wanted = 0;
2879                                         continue;
2880                                 }
2881
2882                                 /* else we go to the next block group */
2883                                 goto new_group;
2884                         }
2885
2886                         ins->objectid = search_start;
2887                         ins->offset = num_bytes;
2888
2889                         btrfs_remove_free_space_lock(block_group, search_start,
2890                                                      num_bytes);
2891                         /* we are all good, lets return */
2892                         mutex_unlock(&block_group->alloc_mutex);
2893                         break;
2894                 }
2895 new_group:
2896                 mutex_unlock(&block_group->alloc_mutex);
2897 new_group_no_lock:
2898                 /* don't try to compare new allocations against the
2899                  * last allocation any more
2900                  */
2901                 last_wanted = 0;
2902
2903                 /*
2904                  * Here's how this works.
2905                  * loop == 0: we were searching a block group via a hint
2906                  *              and didn't find anything, so we start at
2907                  *              the head of the block groups and keep searching
2908                  * loop == 1: we're searching through all of the block groups
2909                  *              if we hit the head again we have searched
2910                  *              all of the block groups for this space and we
2911                  *              need to try and allocate, if we cant error out.
2912                  * loop == 2: we allocated more space and are looping through
2913                  *              all of the block groups again.
2914                  */
2915                 if (loop == 0) {
2916                         head = &space_info->block_groups;
2917                         cur = head->next;
2918                         loop++;
2919                 } else if (loop == 1 && cur == head) {
2920                         int keep_going;
2921
2922                         /* at this point we give up on the empty_size
2923                          * allocations and just try to allocate the min
2924                          * space.
2925                          *
2926                          * The extra_loop field was set if an empty_size
2927                          * allocation was attempted above, and if this
2928                          * is try we need to try the loop again without
2929                          * the additional empty_size.
2930                          */
2931                         total_needed -= empty_size;
2932                         empty_size = 0;
2933                         keep_going = extra_loop;
2934                         loop++;
2935
2936                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2937                                 up_read(&space_info->groups_sem);
2938                                 ret = do_chunk_alloc(trans, root, num_bytes +
2939                                                      2 * 1024 * 1024, data, 1);
2940                                 down_read(&space_info->groups_sem);
2941                                 if (ret < 0)
2942                                         goto loop_check;
2943                                 head = &space_info->block_groups;
2944                                 /*
2945                                  * we've allocated a new chunk, keep
2946                                  * trying
2947                                  */
2948                                 keep_going = 1;
2949                                 chunk_alloc_done = 1;
2950                         } else if (!allowed_chunk_alloc) {
2951                                 space_info->force_alloc = 1;
2952                         }
2953 loop_check:
2954                         if (keep_going) {
2955                                 cur = head->next;
2956                                 extra_loop = 0;
2957                         } else {
2958                                 break;
2959                         }
2960                 } else if (cur == head) {
2961                         break;
2962                 }
2963
2964                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2965                                          list);
2966                 search_start = block_group->key.objectid;
2967                 cur = cur->next;
2968         }
2969
2970         /* we found what we needed */
2971         if (ins->objectid) {
2972                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2973                         trans->block_group = block_group;
2974
2975                 if (last_ptr)
2976                         *last_ptr = ins->objectid + ins->offset;
2977                 ret = 0;
2978         } else if (!ret) {
2979                 ret = -ENOSPC;
2980         }
2981
2982         up_read(&space_info->groups_sem);
2983         return ret;
2984 }
2985
2986 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2987 {
2988         struct btrfs_block_group_cache *cache;
2989         struct list_head *l;
2990
2991         printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
2992                info->total_bytes - info->bytes_used - info->bytes_pinned -
2993                info->bytes_reserved, (info->full) ? "" : "not ");
2994
2995         down_read(&info->groups_sem);
2996         list_for_each(l, &info->block_groups) {
2997                 cache = list_entry(l, struct btrfs_block_group_cache, list);
2998                 spin_lock(&cache->lock);
2999                 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
3000                        "%Lu pinned %Lu reserved\n",
3001                        cache->key.objectid, cache->key.offset,
3002                        btrfs_block_group_used(&cache->item),
3003                        cache->pinned, cache->reserved);
3004                 btrfs_dump_free_space(cache, bytes);
3005                 spin_unlock(&cache->lock);
3006         }
3007         up_read(&info->groups_sem);
3008 }
3009
3010 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3011                                   struct btrfs_root *root,
3012                                   u64 num_bytes, u64 min_alloc_size,
3013                                   u64 empty_size, u64 hint_byte,
3014                                   u64 search_end, struct btrfs_key *ins,
3015                                   u64 data)
3016 {
3017         int ret;
3018         u64 search_start = 0;
3019         u64 alloc_profile;
3020         struct btrfs_fs_info *info = root->fs_info;
3021
3022         if (data) {
3023                 alloc_profile = info->avail_data_alloc_bits &
3024                                 info->data_alloc_profile;
3025                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3026         } else if (root == root->fs_info->chunk_root) {
3027                 alloc_profile = info->avail_system_alloc_bits &
3028                                 info->system_alloc_profile;
3029                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3030         } else {
3031                 alloc_profile = info->avail_metadata_alloc_bits &
3032                                 info->metadata_alloc_profile;
3033                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3034         }
3035 again:
3036         data = btrfs_reduce_alloc_profile(root, data);
3037         /*
3038          * the only place that sets empty_size is btrfs_realloc_node, which
3039          * is not called recursively on allocations
3040          */
3041         if (empty_size || root->ref_cows) {
3042                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3043                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3044                                      2 * 1024 * 1024,
3045                                      BTRFS_BLOCK_GROUP_METADATA |
3046                                      (info->metadata_alloc_profile &
3047                                       info->avail_metadata_alloc_bits), 0);
3048                 }
3049                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3050                                      num_bytes + 2 * 1024 * 1024, data, 0);
3051         }
3052
3053         WARN_ON(num_bytes < root->sectorsize);
3054         ret = find_free_extent(trans, root, num_bytes, empty_size,
3055                                search_start, search_end, hint_byte, ins,
3056                                trans->alloc_exclude_start,
3057                                trans->alloc_exclude_nr, data);
3058
3059         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3060                 num_bytes = num_bytes >> 1;
3061                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3062                 num_bytes = max(num_bytes, min_alloc_size);
3063                 do_chunk_alloc(trans, root->fs_info->extent_root,
3064                                num_bytes, data, 1);
3065                 goto again;
3066         }
3067         if (ret) {
3068                 struct btrfs_space_info *sinfo;
3069
3070                 sinfo = __find_space_info(root->fs_info, data);
3071                 printk("allocation failed flags %Lu, wanted %Lu\n",
3072                        data, num_bytes);
3073                 dump_space_info(sinfo, num_bytes);
3074                 BUG();
3075         }
3076
3077         return ret;
3078 }
3079
3080 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3081 {
3082         struct btrfs_block_group_cache *cache;
3083
3084         cache = btrfs_lookup_block_group(root->fs_info, start);
3085         if (!cache) {
3086                 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
3087                 return -ENOSPC;
3088         }
3089         btrfs_add_free_space(cache, start, len);
3090         update_reserved_extents(root, start, len, 0);
3091         return 0;
3092 }
3093
3094 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3095                                   struct btrfs_root *root,
3096                                   u64 num_bytes, u64 min_alloc_size,
3097                                   u64 empty_size, u64 hint_byte,
3098                                   u64 search_end, struct btrfs_key *ins,
3099                                   u64 data)
3100 {
3101         int ret;
3102         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3103                                      empty_size, hint_byte, search_end, ins,
3104                                      data);
3105         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3106         return ret;
3107 }
3108
3109 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3110                                          struct btrfs_root *root, u64 parent,
3111                                          u64 root_objectid, u64 ref_generation,
3112                                          u64 owner, struct btrfs_key *ins)
3113 {
3114         int ret;
3115         int pending_ret;
3116         u64 super_used;
3117         u64 root_used;
3118         u64 num_bytes = ins->offset;
3119         u32 sizes[2];
3120         struct btrfs_fs_info *info = root->fs_info;
3121         struct btrfs_root *extent_root = info->extent_root;
3122         struct btrfs_extent_item *extent_item;
3123         struct btrfs_extent_ref *ref;
3124         struct btrfs_path *path;
3125         struct btrfs_key keys[2];
3126
3127         if (parent == 0)
3128                 parent = ins->objectid;
3129
3130         /* block accounting for super block */
3131         spin_lock_irq(&info->delalloc_lock);
3132         super_used = btrfs_super_bytes_used(&info->super_copy);
3133         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3134         spin_unlock_irq(&info->delalloc_lock);
3135
3136         /* block accounting for root item */
3137         root_used = btrfs_root_used(&root->root_item);
3138         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3139
3140         if (root == extent_root) {
3141                 struct pending_extent_op *extent_op;
3142
3143                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3144                 BUG_ON(!extent_op);
3145
3146                 extent_op->type = PENDING_EXTENT_INSERT;
3147                 extent_op->bytenr = ins->objectid;
3148                 extent_op->num_bytes = ins->offset;
3149                 extent_op->parent = parent;
3150                 extent_op->orig_parent = 0;
3151                 extent_op->generation = ref_generation;
3152                 extent_op->orig_generation = 0;
3153                 extent_op->level = (int)owner;
3154                 INIT_LIST_HEAD(&extent_op->list);
3155                 extent_op->del = 0;
3156
3157                 mutex_lock(&root->fs_info->extent_ins_mutex);
3158                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3159                                 ins->objectid + ins->offset - 1,
3160                                 EXTENT_WRITEBACK, GFP_NOFS);
3161                 set_state_private(&root->fs_info->extent_ins,
3162                                   ins->objectid, (unsigned long)extent_op);
3163                 mutex_unlock(&root->fs_info->extent_ins_mutex);
3164                 goto update_block;
3165         }
3166
3167         memcpy(&keys[0], ins, sizeof(*ins));
3168         keys[1].objectid = ins->objectid;
3169         keys[1].type = BTRFS_EXTENT_REF_KEY;
3170         keys[1].offset = parent;
3171         sizes[0] = sizeof(*extent_item);
3172         sizes[1] = sizeof(*ref);
3173
3174         path = btrfs_alloc_path();
3175         BUG_ON(!path);
3176
3177         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3178                                        sizes, 2);
3179         BUG_ON(ret);
3180
3181         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3182                                      struct btrfs_extent_item);
3183         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3184         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3185                              struct btrfs_extent_ref);
3186
3187         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3188         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3189         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3190         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3191
3192         btrfs_mark_buffer_dirty(path->nodes[0]);
3193
3194         trans->alloc_exclude_start = 0;
3195         trans->alloc_exclude_nr = 0;
3196         btrfs_free_path(path);
3197         finish_current_insert(trans, extent_root, 0);
3198         pending_ret = del_pending_extents(trans, extent_root, 0);
3199
3200         if (ret)
3201                 goto out;
3202         if (pending_ret) {
3203                 ret = pending_ret;
3204                 goto out;
3205         }
3206
3207 update_block:
3208         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
3209         if (ret) {
3210                 printk("update block group failed for %Lu %Lu\n",
3211                        ins->objectid, ins->offset);
3212                 BUG();
3213         }
3214 out:
3215         return ret;
3216 }
3217
3218 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3219                                 struct btrfs_root *root, u64 parent,
3220                                 u64 root_objectid, u64 ref_generation,
3221                                 u64 owner, struct btrfs_key *ins)
3222 {
3223         int ret;
3224
3225         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3226                 return 0;
3227         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3228                                             ref_generation, owner, ins);
3229         update_reserved_extents(root, ins->objectid, ins->offset, 0);
3230         return ret;
3231 }
3232
3233 /*
3234  * this is used by the tree logging recovery code.  It records that
3235  * an extent has been allocated and makes sure to clear the free
3236  * space cache bits as well
3237  */
3238 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3239                                 struct btrfs_root *root, u64 parent,
3240                                 u64 root_objectid, u64 ref_generation,
3241                                 u64 owner, struct btrfs_key *ins)
3242 {
3243         int ret;
3244         struct btrfs_block_group_cache *block_group;
3245
3246         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3247         mutex_lock(&block_group->alloc_mutex);
3248         cache_block_group(root, block_group);
3249
3250         ret = btrfs_remove_free_space_lock(block_group, ins->objectid,
3251                                            ins->offset);
3252         mutex_unlock(&block_group->alloc_mutex);
3253         BUG_ON(ret);
3254         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3255                                             ref_generation, owner, ins);
3256         return ret;
3257 }
3258
3259 /*
3260  * finds a free extent and does all the dirty work required for allocation
3261  * returns the key for the extent through ins, and a tree buffer for
3262  * the first block of the extent through buf.
3263  *
3264  * returns 0 if everything worked, non-zero otherwise.
3265  */
3266 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3267                        struct btrfs_root *root,
3268                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3269                        u64 root_objectid, u64 ref_generation,
3270                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3271                        u64 search_end, struct btrfs_key *ins, u64 data)
3272 {
3273         int ret;
3274
3275         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3276                                      min_alloc_size, empty_size, hint_byte,
3277                                      search_end, ins, data);
3278         BUG_ON(ret);
3279         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3280                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3281                                         root_objectid, ref_generation,
3282                                         owner_objectid, ins);
3283                 BUG_ON(ret);
3284
3285         } else {
3286                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3287         }
3288         return ret;
3289 }
3290
3291 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3292                                             struct btrfs_root *root,
3293                                             u64 bytenr, u32 blocksize)
3294 {
3295         struct extent_buffer *buf;
3296
3297         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3298         if (!buf)
3299                 return ERR_PTR(-ENOMEM);
3300         btrfs_set_header_generation(buf, trans->transid);
3301         btrfs_tree_lock(buf);
3302         clean_tree_block(trans, root, buf);
3303         btrfs_set_buffer_uptodate(buf);
3304         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3305                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3306                          buf->start + buf->len - 1, GFP_NOFS);
3307         } else {
3308                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3309                          buf->start + buf->len - 1, GFP_NOFS);
3310         }
3311         trans->blocks_used++;
3312         return buf;
3313 }
3314
3315 /*
3316  * helper function to allocate a block for a given tree
3317  * returns the tree buffer or NULL.
3318  */
3319 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3320                                              struct btrfs_root *root,
3321                                              u32 blocksize, u64 parent,
3322                                              u64 root_objectid,
3323                                              u64 ref_generation,
3324                                              int level,
3325                                              u64 hint,
3326                                              u64 empty_size)
3327 {
3328         struct btrfs_key ins;
3329         int ret;
3330         struct extent_buffer *buf;
3331
3332         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3333                                  root_objectid, ref_generation, level,
3334                                  empty_size, hint, (u64)-1, &ins, 0);
3335         if (ret) {
3336                 BUG_ON(ret > 0);
3337                 return ERR_PTR(ret);
3338         }
3339
3340         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3341         return buf;
3342 }
3343
3344 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3345                         struct btrfs_root *root, struct extent_buffer *leaf)
3346 {
3347         u64 leaf_owner;
3348         u64 leaf_generation;
3349         struct btrfs_key key;
3350         struct btrfs_file_extent_item *fi;
3351         int i;
3352         int nritems;
3353         int ret;
3354
3355         BUG_ON(!btrfs_is_leaf(leaf));
3356         nritems = btrfs_header_nritems(leaf);
3357         leaf_owner = btrfs_header_owner(leaf);
3358         leaf_generation = btrfs_header_generation(leaf);
3359
3360         for (i = 0; i < nritems; i++) {
3361                 u64 disk_bytenr;
3362                 cond_resched();
3363
3364                 btrfs_item_key_to_cpu(leaf, &key, i);
3365                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3366                         continue;
3367                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3368                 if (btrfs_file_extent_type(leaf, fi) ==
3369                     BTRFS_FILE_EXTENT_INLINE)
3370                         continue;
3371                 /*
3372                  * FIXME make sure to insert a trans record that
3373                  * repeats the snapshot del on crash
3374                  */
3375                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3376                 if (disk_bytenr == 0)
3377                         continue;
3378
3379                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3380                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3381                                 leaf->start, leaf_owner, leaf_generation,
3382                                 key.objectid, 0);
3383                 BUG_ON(ret);
3384
3385                 atomic_inc(&root->fs_info->throttle_gen);
3386                 wake_up(&root->fs_info->transaction_throttle);
3387                 cond_resched();
3388         }
3389         return 0;
3390 }
3391
3392 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3393                                         struct btrfs_root *root,
3394                                         struct btrfs_leaf_ref *ref)
3395 {
3396         int i;
3397         int ret;
3398         struct btrfs_extent_info *info = ref->extents;
3399
3400         for (i = 0; i < ref->nritems; i++) {
3401                 ret = __btrfs_free_extent(trans, root, info->bytenr,
3402                                           info->num_bytes, ref->bytenr,
3403                                           ref->owner, ref->generation,
3404                                           info->objectid, 0);
3405
3406                 atomic_inc(&root->fs_info->throttle_gen);
3407                 wake_up(&root->fs_info->transaction_throttle);
3408                 cond_resched();
3409
3410                 BUG_ON(ret);
3411                 info++;
3412         }
3413
3414         return 0;
3415 }
3416
3417 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
3418                               u32 *refs)
3419 {
3420         int ret;
3421
3422         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3423         BUG_ON(ret);
3424
3425 #if 0 // some debugging code in case we see problems here
3426         /* if the refs count is one, it won't get increased again.  But
3427          * if the ref count is > 1, someone may be decreasing it at
3428          * the same time we are.
3429          */
3430         if (*refs != 1) {
3431                 struct extent_buffer *eb = NULL;
3432                 eb = btrfs_find_create_tree_block(root, start, len);
3433                 if (eb)
3434                         btrfs_tree_lock(eb);
3435
3436                 mutex_lock(&root->fs_info->alloc_mutex);
3437                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3438                 BUG_ON(ret);
3439                 mutex_unlock(&root->fs_info->alloc_mutex);
3440
3441                 if (eb) {
3442                         btrfs_tree_unlock(eb);
3443                         free_extent_buffer(eb);
3444                 }
3445                 if (*refs == 1) {
3446                         printk("block %llu went down to one during drop_snap\n",
3447                                (unsigned long long)start);
3448                 }
3449
3450         }
3451 #endif
3452
3453         cond_resched();
3454         return ret;
3455 }
3456
3457 /*
3458  * helper function for drop_snapshot, this walks down the tree dropping ref
3459  * counts as it goes.
3460  */
3461 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
3462                                    struct btrfs_root *root,
3463                                    struct btrfs_path *path, int *level)
3464 {
3465         u64 root_owner;
3466         u64 root_gen;
3467         u64 bytenr;
3468         u64 ptr_gen;
3469         struct extent_buffer *next;
3470         struct extent_buffer *cur;
3471         struct extent_buffer *parent;
3472         struct btrfs_leaf_ref *ref;
3473         u32 blocksize;
3474         int ret;
3475         u32 refs;
3476
3477         WARN_ON(*level < 0);
3478         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3479         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3480                                 path->nodes[*level]->len, &refs);
3481         BUG_ON(ret);
3482         if (refs > 1)
3483                 goto out;
3484
3485         /*
3486          * walk down to the last node level and free all the leaves
3487          */
3488         while(*level >= 0) {
3489                 WARN_ON(*level < 0);
3490                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3491                 cur = path->nodes[*level];
3492
3493                 if (btrfs_header_level(cur) != *level)
3494                         WARN_ON(1);
3495
3496                 if (path->slots[*level] >=
3497                     btrfs_header_nritems(cur))
3498                         break;
3499                 if (*level == 0) {
3500                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3501                         BUG_ON(ret);
3502                         break;
3503                 }
3504                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3505                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3506                 blocksize = btrfs_level_size(root, *level - 1);
3507
3508                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3509                 BUG_ON(ret);
3510                 if (refs != 1) {
3511                         parent = path->nodes[*level];
3512                         root_owner = btrfs_header_owner(parent);
3513                         root_gen = btrfs_header_generation(parent);
3514                         path->slots[*level]++;
3515
3516                         ret = __btrfs_free_extent(trans, root, bytenr,
3517                                                 blocksize, parent->start,
3518                                                 root_owner, root_gen,
3519                                                 *level - 1, 1);
3520                         BUG_ON(ret);
3521
3522                         atomic_inc(&root->fs_info->throttle_gen);
3523                         wake_up(&root->fs_info->transaction_throttle);
3524                         cond_resched();
3525
3526                         continue;
3527                 }
3528                 /*
3529                  * at this point, we have a single ref, and since the
3530                  * only place referencing this extent is a dead root
3531                  * the reference count should never go higher.
3532                  * So, we don't need to check it again
3533                  */
3534                 if (*level == 1) {
3535                         ref = btrfs_lookup_leaf_ref(root, bytenr);
3536                         if (ref && ref->generation != ptr_gen) {
3537                                 btrfs_free_leaf_ref(root, ref);
3538                                 ref = NULL;
3539                         }
3540                         if (ref) {
3541                                 ret = cache_drop_leaf_ref(trans, root, ref);
3542                                 BUG_ON(ret);
3543                                 btrfs_remove_leaf_ref(root, ref);
3544                                 btrfs_free_leaf_ref(root, ref);
3545                                 *level = 0;
3546                                 break;
3547                         }
3548                         if (printk_ratelimit()) {
3549                                 printk("leaf ref miss for bytenr %llu\n",
3550                                        (unsigned long long)bytenr);
3551                         }
3552                 }
3553                 next = btrfs_find_tree_block(root, bytenr, blocksize);
3554                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3555                         free_extent_buffer(next);
3556
3557                         next = read_tree_block(root, bytenr, blocksize,
3558                                                ptr_gen);
3559                         cond_resched();
3560 #if 0
3561                         /*
3562                          * this is a debugging check and can go away
3563                          * the ref should never go all the way down to 1
3564                          * at this point
3565                          */
3566                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3567                                                 &refs);
3568                         BUG_ON(ret);
3569                         WARN_ON(refs != 1);
3570 #endif
3571                 }
3572                 WARN_ON(*level <= 0);
3573                 if (path->nodes[*level-1])
3574                         free_extent_buffer(path->nodes[*level-1]);
3575                 path->nodes[*level-1] = next;
3576                 *level = btrfs_header_level(next);
3577                 path->slots[*level] = 0;
3578                 cond_resched();
3579         }
3580 out:
3581         WARN_ON(*level < 0);
3582         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3583
3584         if (path->nodes[*level] == root->node) {
3585                 parent = path->nodes[*level];
3586                 bytenr = path->nodes[*level]->start;
3587         } else {
3588                 parent = path->nodes[*level + 1];
3589                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3590         }
3591
3592         blocksize = btrfs_level_size(root, *level);
3593         root_owner = btrfs_header_owner(parent);
3594         root_gen = btrfs_header_generation(parent);
3595
3596         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3597                                   parent->start, root_owner, root_gen,
3598                                   *level, 1);
3599         free_extent_buffer(path->nodes[*level]);
3600         path->nodes[*level] = NULL;
3601         *level += 1;
3602         BUG_ON(ret);
3603
3604         cond_resched();
3605         return 0;
3606 }
3607
3608 /*
3609  * helper function for drop_subtree, this function is similar to
3610  * walk_down_tree. The main difference is that it checks reference
3611  * counts while tree blocks are locked.
3612  */
3613 static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
3614                                       struct btrfs_root *root,
3615                                       struct btrfs_path *path, int *level)
3616 {
3617         struct extent_buffer *next;
3618         struct extent_buffer *cur;
3619         struct extent_buffer *parent;
3620         u64 bytenr;
3621         u64 ptr_gen;
3622         u32 blocksize;
3623         u32 refs;
3624         int ret;
3625
3626         cur = path->nodes[*level];
3627         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3628                                       &refs);
3629         BUG_ON(ret);
3630         if (refs > 1)
3631                 goto out;
3632
3633         while (*level >= 0) {
3634                 cur = path->nodes[*level];
3635                 if (*level == 0) {
3636                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3637                         BUG_ON(ret);
3638                         clean_tree_block(trans, root, cur);
3639                         break;
3640                 }
3641                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3642                         clean_tree_block(trans, root, cur);
3643                         break;
3644                 }
3645
3646                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3647                 blocksize = btrfs_level_size(root, *level - 1);
3648                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3649
3650                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3651                 btrfs_tree_lock(next);
3652
3653                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3654                                               &refs);
3655                 BUG_ON(ret);
3656                 if (refs > 1) {
3657                         parent = path->nodes[*level];
3658                         ret = btrfs_free_extent(trans, root, bytenr,
3659                                         blocksize, parent->start,
3660                                         btrfs_header_owner(parent),
3661                                         btrfs_header_generation(parent),
3662                                         *level - 1, 1);
3663                         BUG_ON(ret);
3664                         path->slots[*level]++;
3665                         btrfs_tree_unlock(next);
3666                         free_extent_buffer(next);
3667                         continue;
3668                 }
3669
3670                 *level = btrfs_header_level(next);
3671                 path->nodes[*level] = next;
3672                 path->slots[*level] = 0;
3673                 path->locks[*level] = 1;
3674                 cond_resched();
3675         }
3676 out:
3677         parent = path->nodes[*level + 1];
3678         bytenr = path->nodes[*level]->start;
3679         blocksize = path->nodes[*level]->len;
3680
3681         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3682                         parent->start, btrfs_header_owner(parent),
3683                         btrfs_header_generation(parent), *level, 1);
3684         BUG_ON(ret);
3685
3686         if (path->locks[*level]) {
3687                 btrfs_tree_unlock(path->nodes[*level]);
3688                 path->locks[*level] = 0;
3689         }
3690         free_extent_buffer(path->nodes[*level]);
3691         path->nodes[*level] = NULL;
3692         *level += 1;
3693         cond_resched();
3694         return 0;
3695 }
3696
3697 /*
3698  * helper for dropping snapshots.  This walks back up the tree in the path
3699  * to find the first node higher up where we haven't yet gone through
3700  * all the slots
3701  */
3702 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3703                                  struct btrfs_root *root,
3704                                  struct btrfs_path *path,
3705                                  int *level, int max_level)
3706 {
3707         u64 root_owner;
3708         u64 root_gen;
3709         struct btrfs_root_item *root_item = &root->root_item;
3710         int i;
3711         int slot;
3712         int ret;
3713
3714         for (i = *level; i < max_level && path->nodes[i]; i++) {
3715                 slot = path->slots[i];
3716                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3717                         struct extent_buffer *node;
3718                         struct btrfs_disk_key disk_key;
3719                         node = path->nodes[i];
3720                         path->slots[i]++;
3721                         *level = i;
3722                         WARN_ON(*level == 0);
3723                         btrfs_node_key(node, &disk_key, path->slots[i]);
3724                         memcpy(&root_item->drop_progress,
3725                                &disk_key, sizeof(disk_key));
3726                         root_item->drop_level = i;
3727                         return 0;
3728                 } else {
3729                         struct extent_buffer *parent;
3730                         if (path->nodes[*level] == root->node)
3731                                 parent = path->nodes[*level];
3732                         else
3733                                 parent = path->nodes[*level + 1];
3734
3735                         root_owner = btrfs_header_owner(parent);
3736                         root_gen = btrfs_header_generation(parent);
3737
3738                         clean_tree_block(trans, root, path->nodes[*level]);
3739                         ret = btrfs_free_extent(trans, root,
3740                                                 path->nodes[*level]->start,
3741                                                 path->nodes[*level]->len,
3742                                                 parent->start, root_owner,
3743                                                 root_gen, *level, 1);
3744                         BUG_ON(ret);
3745                         if (path->locks[*level]) {
3746                                 btrfs_tree_unlock(path->nodes[*level]);
3747                                 path->locks[*level] = 0;
3748                         }
3749                         free_extent_buffer(path->nodes[*level]);
3750                         path->nodes[*level] = NULL;
3751                         *level = i + 1;
3752                 }
3753         }
3754         return 1;
3755 }
3756
3757 /*
3758  * drop the reference count on the tree rooted at 'snap'.  This traverses
3759  * the tree freeing any blocks that have a ref count of zero after being
3760  * decremented.
3761  */
3762 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3763                         *root)
3764 {
3765         int ret = 0;
3766         int wret;
3767         int level;
3768         struct btrfs_path *path;
3769         int i;
3770         int orig_level;
3771         struct btrfs_root_item *root_item = &root->root_item;
3772
3773         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3774         path = btrfs_alloc_path();
3775         BUG_ON(!path);
3776
3777         level = btrfs_header_level(root->node);
3778         orig_level = level;
3779         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3780                 path->nodes[level] = root->node;
3781                 extent_buffer_get(root->node);
3782                 path->slots[level] = 0;
3783         } else {
3784                 struct btrfs_key key;
3785                 struct btrfs_disk_key found_key;
3786                 struct extent_buffer *node;
3787
3788                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3789                 level = root_item->drop_level;
3790                 path->lowest_level = level;
3791                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3792                 if (wret < 0) {
3793                         ret = wret;
3794                         goto out;
3795                 }
3796                 node = path->nodes[level];
3797                 btrfs_node_key(node, &found_key, path->slots[level]);
3798                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3799                                sizeof(found_key)));
3800                 /*
3801                  * unlock our path, this is safe because only this
3802                  * function is allowed to delete this snapshot
3803                  */
3804                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3805                         if (path->nodes[i] && path->locks[i]) {
3806                                 path->locks[i] = 0;
3807                                 btrfs_tree_unlock(path->nodes[i]);
3808                         }
3809                 }
3810         }
3811         while(1) {
3812                 wret = walk_down_tree(trans, root, path, &level);
3813                 if (wret > 0)
3814                         break;
3815                 if (wret < 0)
3816                         ret = wret;
3817
3818                 wret = walk_up_tree(trans, root, path, &level,
3819                                     BTRFS_MAX_LEVEL);
3820                 if (wret > 0)
3821                         break;
3822                 if (wret < 0)
3823                         ret = wret;
3824                 if (trans->transaction->in_commit) {
3825                         ret = -EAGAIN;
3826                         break;
3827                 }
3828                 atomic_inc(&root->fs_info->throttle_gen);
3829                 wake_up(&root->fs_info->transaction_throttle);
3830         }
3831         for (i = 0; i <= orig_level; i++) {
3832                 if (path->nodes[i]) {
3833                         free_extent_buffer(path->nodes[i]);
3834                         path->nodes[i] = NULL;
3835                 }
3836         }
3837 out:
3838         btrfs_free_path(path);
3839         return ret;
3840 }
3841
3842 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3843                         struct btrfs_root *root,
3844                         struct extent_buffer *node,
3845                         struct extent_buffer *parent)
3846 {
3847         struct btrfs_path *path;
3848         int level;
3849         int parent_level;
3850         int ret = 0;
3851         int wret;
3852
3853         path = btrfs_alloc_path();
3854         BUG_ON(!path);
3855
3856         BUG_ON(!btrfs_tree_locked(parent));
3857         parent_level = btrfs_header_level(parent);
3858         extent_buffer_get(parent);
3859         path->nodes[parent_level] = parent;
3860         path->slots[parent_level] = btrfs_header_nritems(parent);
3861
3862         BUG_ON(!btrfs_tree_locked(node));
3863         level = btrfs_header_level(node);
3864         extent_buffer_get(node);
3865         path->nodes[level] = node;
3866         path->slots[level] = 0;
3867
3868         while (1) {
3869                 wret = walk_down_subtree(trans, root, path, &level);
3870                 if (wret < 0)
3871                         ret = wret;
3872                 if (wret != 0)
3873                         break;
3874
3875                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3876                 if (wret < 0)
3877                         ret = wret;
3878                 if (wret != 0)
3879                         break;
3880         }
3881
3882         btrfs_free_path(path);
3883         return ret;
3884 }
3885
3886 static unsigned long calc_ra(unsigned long start, unsigned long last,
3887                              unsigned long nr)
3888 {
3889         return min(last, start + nr - 1);
3890 }
3891
3892 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3893                                          u64 len)
3894 {
3895         u64 page_start;
3896         u64 page_end;
3897         unsigned long first_index;
3898         unsigned long last_index;
3899         unsigned long i;
3900         struct page *page;
3901         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3902         struct file_ra_state *ra;
3903         struct btrfs_ordered_extent *ordered;
3904         unsigned int total_read = 0;
3905         unsigned int total_dirty = 0;
3906         int ret = 0;
3907
3908         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3909
3910         mutex_lock(&inode->i_mutex);
3911         first_index = start >> PAGE_CACHE_SHIFT;
3912         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3913
3914         /* make sure the dirty trick played by the caller work */
3915         ret = invalidate_inode_pages2_range(inode->i_mapping,
3916                                             first_index, last_index);
3917         if (ret)
3918                 goto out_unlock;
3919
3920         file_ra_state_init(ra, inode->i_mapping);
3921
3922         for (i = first_index ; i <= last_index; i++) {
3923                 if (total_read % ra->ra_pages == 0) {
3924                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3925                                        calc_ra(i, last_index, ra->ra_pages));
3926                 }
3927                 total_read++;
3928 again:
3929                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3930                         BUG_ON(1);
3931                 page = grab_cache_page(inode->i_mapping, i);
3932                 if (!page) {
3933                         ret = -ENOMEM;
3934                         goto out_unlock;
3935                 }
3936                 if (!PageUptodate(page)) {
3937                         btrfs_readpage(NULL, page);
3938                         lock_page(page);
3939                         if (!PageUptodate(page)) {
3940                                 unlock_page(page);
3941                                 page_cache_release(page);
3942                                 ret = -EIO;
3943                                 goto out_unlock;
3944                         }
3945                 }
3946                 wait_on_page_writeback(page);
3947
3948                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3949                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3950                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3951
3952                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3953                 if (ordered) {
3954                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3955                         unlock_page(page);
3956                         page_cache_release(page);
3957                         btrfs_start_ordered_extent(inode, ordered, 1);
3958                         btrfs_put_ordered_extent(ordered);
3959                         goto again;
3960                 }
3961                 set_page_extent_mapped(page);
3962
3963                 btrfs_set_extent_delalloc(inode, page_start, page_end);
3964                 if (i == first_index)
3965                         set_extent_bits(io_tree, page_start, page_end,
3966                                         EXTENT_BOUNDARY, GFP_NOFS);
3967
3968                 set_page_dirty(page);
3969                 total_dirty++;
3970
3971                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3972                 unlock_page(page);
3973                 page_cache_release(page);
3974         }
3975
3976 out_unlock:
3977         kfree(ra);
3978         mutex_unlock(&inode->i_mutex);
3979         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3980         return ret;
3981 }
3982
3983 static int noinline relocate_data_extent(struct inode *reloc_inode,
3984                                          struct btrfs_key *extent_key,
3985                                          u64 offset)
3986 {
3987         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3988         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3989         struct extent_map *em;
3990         u64 start = extent_key->objectid - offset;
3991         u64 end = start + extent_key->offset - 1;
3992
3993         em = alloc_extent_map(GFP_NOFS);
3994         BUG_ON(!em || IS_ERR(em));
3995
3996         em->start = start;
3997         em->len = extent_key->offset;
3998         em->block_len = extent_key->offset;
3999         em->block_start = extent_key->objectid;
4000         em->bdev = root->fs_info->fs_devices->latest_bdev;
4001         set_bit(EXTENT_FLAG_PINNED, &em->flags);
4002
4003         /* setup extent map to cheat btrfs_readpage */
4004         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4005         while (1) {
4006                 int ret;
4007                 spin_lock(&em_tree->lock);
4008                 ret = add_extent_mapping(em_tree, em);
4009                 spin_unlock(&em_tree->lock);
4010                 if (ret != -EEXIST) {
4011                         free_extent_map(em);
4012                         break;
4013                 }
4014                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4015         }
4016         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4017
4018         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4019 }
4020
4021 struct btrfs_ref_path {
4022         u64 extent_start;
4023         u64 nodes[BTRFS_MAX_LEVEL];
4024         u64 root_objectid;
4025         u64 root_generation;
4026         u64 owner_objectid;
4027         u32 num_refs;
4028         int lowest_level;
4029         int current_level;
4030         int shared_level;
4031
4032         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4033         u64 new_nodes[BTRFS_MAX_LEVEL];
4034 };
4035
4036 struct disk_extent {
4037         u64 ram_bytes;
4038         u64 disk_bytenr;
4039         u64 disk_num_bytes;
4040         u64 offset;
4041         u64 num_bytes;
4042         u8 compression;
4043         u8 encryption;
4044         u16 other_encoding;
4045 };
4046
4047 static int is_cowonly_root(u64 root_objectid)
4048 {
4049         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4050             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4051             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4052             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4053             root_objectid == BTRFS_TREE_LOG_OBJECTID)
4054                 return 1;
4055         return 0;
4056 }
4057
4058 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
4059                                     struct btrfs_root *extent_root,
4060                                     struct btrfs_ref_path *ref_path,
4061                                     int first_time)
4062 {
4063         struct extent_buffer *leaf;
4064         struct btrfs_path *path;
4065         struct btrfs_extent_ref *ref;
4066         struct btrfs_key key;
4067         struct btrfs_key found_key;
4068         u64 bytenr;
4069         u32 nritems;
4070         int level;
4071         int ret = 1;
4072
4073         path = btrfs_alloc_path();
4074         if (!path)
4075                 return -ENOMEM;
4076
4077         if (first_time) {
4078                 ref_path->lowest_level = -1;
4079                 ref_path->current_level = -1;
4080                 ref_path->shared_level = -1;
4081                 goto walk_up;
4082         }
4083 walk_down:
4084         level = ref_path->current_level - 1;
4085         while (level >= -1) {
4086                 u64 parent;
4087                 if (level < ref_path->lowest_level)
4088                         break;
4089
4090                 if (level >= 0) {
4091                         bytenr = ref_path->nodes[level];
4092                 } else {
4093                         bytenr = ref_path->extent_start;
4094                 }
4095                 BUG_ON(bytenr == 0);
4096
4097                 parent = ref_path->nodes[level + 1];
4098                 ref_path->nodes[level + 1] = 0;
4099                 ref_path->current_level = level;
4100                 BUG_ON(parent == 0);
4101
4102                 key.objectid = bytenr;
4103                 key.offset = parent + 1;
4104                 key.type = BTRFS_EXTENT_REF_KEY;
4105
4106                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4107                 if (ret < 0)
4108                         goto out;
4109                 BUG_ON(ret == 0);
4110
4111                 leaf = path->nodes[0];
4112                 nritems = btrfs_header_nritems(leaf);
4113                 if (path->slots[0] >= nritems) {
4114                         ret = btrfs_next_leaf(extent_root, path);
4115                         if (ret < 0)
4116                                 goto out;
4117                         if (ret > 0)
4118                                 goto next;
4119                         leaf = path->nodes[0];
4120                 }
4121
4122                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4123                 if (found_key.objectid == bytenr &&
4124                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4125                         if (level < ref_path->shared_level)
4126                                 ref_path->shared_level = level;
4127                         goto found;
4128                 }
4129 next:
4130                 level--;
4131                 btrfs_release_path(extent_root, path);
4132                 cond_resched();
4133         }
4134         /* reached lowest level */
4135         ret = 1;
4136         goto out;
4137 walk_up:
4138         level = ref_path->current_level;
4139         while (level < BTRFS_MAX_LEVEL - 1) {
4140                 u64 ref_objectid;
4141                 if (level >= 0) {
4142                         bytenr = ref_path->nodes[level];
4143                 } else {
4144                         bytenr = ref_path->extent_start;
4145                 }
4146                 BUG_ON(bytenr == 0);
4147
4148                 key.objectid = bytenr;
4149                 key.offset = 0;
4150                 key.type = BTRFS_EXTENT_REF_KEY;
4151
4152                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4153                 if (ret < 0)
4154                         goto out;
4155
4156                 leaf = path->nodes[0];
4157                 nritems = btrfs_header_nritems(leaf);
4158                 if (path->slots[0] >= nritems) {
4159                         ret = btrfs_next_leaf(extent_root, path);
4160                         if (ret < 0)
4161                                 goto out;
4162                         if (ret > 0) {
4163                                 /* the extent was freed by someone */
4164                                 if (ref_path->lowest_level == level)
4165                                         goto out;
4166                                 btrfs_release_path(extent_root, path);
4167                                 goto walk_down;
4168                         }
4169                         leaf = path->nodes[0];
4170                 }
4171
4172                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4173                 if (found_key.objectid != bytenr ||
4174                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4175                         /* the extent was freed by someone */
4176                         if (ref_path->lowest_level == level) {
4177                                 ret = 1;
4178                                 goto out;
4179                         }
4180                         btrfs_release_path(extent_root, path);
4181                         goto walk_down;
4182                 }
4183 found:
4184                 ref = btrfs_item_ptr(leaf, path->slots[0],
4185                                 struct btrfs_extent_ref);
4186                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4187                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4188                         if (first_time) {
4189                                 level = (int)ref_objectid;
4190                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4191                                 ref_path->lowest_level = level;
4192                                 ref_path->current_level = level;
4193                                 ref_path->nodes[level] = bytenr;
4194                         } else {
4195                                 WARN_ON(ref_objectid != level);
4196                         }
4197                 } else {
4198                         WARN_ON(level != -1);
4199                 }
4200                 first_time = 0;
4201
4202                 if (ref_path->lowest_level == level) {
4203                         ref_path->owner_objectid = ref_objectid;
4204                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4205                 }
4206
4207                 /*
4208                  * the block is tree root or the block isn't in reference
4209                  * counted tree.
4210                  */
4211                 if (found_key.objectid == found_key.offset ||
4212                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4213                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4214                         ref_path->root_generation =
4215                                 btrfs_ref_generation(leaf, ref);
4216                         if (level < 0) {
4217                                 /* special reference from the tree log */
4218                                 ref_path->nodes[0] = found_key.offset;
4219                                 ref_path->current_level = 0;
4220                         }
4221                         ret = 0;
4222                         goto out;
4223                 }
4224
4225                 level++;
4226                 BUG_ON(ref_path->nodes[level] != 0);
4227                 ref_path->nodes[level] = found_key.offset;
4228                 ref_path->current_level = level;
4229
4230                 /*
4231                  * the reference was created in the running transaction,
4232                  * no need to continue walking up.
4233                  */
4234                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4235                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4236                         ref_path->root_generation =
4237                                 btrfs_ref_generation(leaf, ref);
4238                         ret = 0;
4239                         goto out;
4240                 }
4241
4242                 btrfs_release_path(extent_root, path);
4243                 cond_resched();
4244         }
4245         /* reached max tree level, but no tree root found. */
4246         BUG();
4247 out:
4248         btrfs_free_path(path);
4249         return ret;
4250 }
4251
4252 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4253                                 struct btrfs_root *extent_root,
4254                                 struct btrfs_ref_path *ref_path,
4255                                 u64 extent_start)
4256 {
4257         memset(ref_path, 0, sizeof(*ref_path));
4258         ref_path->extent_start = extent_start;
4259
4260         return __next_ref_path(trans, extent_root, ref_path, 1);
4261 }
4262
4263 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4264                                struct btrfs_root *extent_root,
4265                                struct btrfs_ref_path *ref_path)
4266 {
4267         return __next_ref_path(trans, extent_root, ref_path, 0);
4268 }
4269
4270 static int noinline get_new_locations(struct inode *reloc_inode,
4271                                       struct btrfs_key *extent_key,
4272                                       u64 offset, int no_fragment,
4273                                       struct disk_extent **extents,
4274                                       int *nr_extents)
4275 {
4276         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4277         struct btrfs_path *path;
4278         struct btrfs_file_extent_item *fi;
4279         struct extent_buffer *leaf;
4280         struct disk_extent *exts = *extents;
4281         struct btrfs_key found_key;
4282         u64 cur_pos;
4283         u64 last_byte;
4284         u32 nritems;
4285         int nr = 0;
4286         int max = *nr_extents;
4287         int ret;
4288
4289         WARN_ON(!no_fragment && *extents);
4290         if (!exts) {
4291                 max = 1;
4292                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4293                 if (!exts)
4294                         return -ENOMEM;
4295         }
4296
4297         path = btrfs_alloc_path();
4298         BUG_ON(!path);
4299
4300         cur_pos = extent_key->objectid - offset;
4301         last_byte = extent_key->objectid + extent_key->offset;
4302         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4303                                        cur_pos, 0);
4304         if (ret < 0)
4305                 goto out;
4306         if (ret > 0) {
4307                 ret = -ENOENT;
4308                 goto out;
4309         }
4310
4311         while (1) {
4312                 leaf = path->nodes[0];
4313                 nritems = btrfs_header_nritems(leaf);
4314                 if (path->slots[0] >= nritems) {
4315                         ret = btrfs_next_leaf(root, path);
4316                         if (ret < 0)
4317                                 goto out;
4318                         if (ret > 0)
4319                                 break;
4320                         leaf = path->nodes[0];
4321                 }
4322
4323                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4324                 if (found_key.offset != cur_pos ||
4325                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4326                     found_key.objectid != reloc_inode->i_ino)
4327                         break;
4328
4329                 fi = btrfs_item_ptr(leaf, path->slots[0],
4330                                     struct btrfs_file_extent_item);
4331                 if (btrfs_file_extent_type(leaf, fi) !=
4332                     BTRFS_FILE_EXTENT_REG ||
4333                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4334                         break;
4335
4336                 if (nr == max) {
4337                         struct disk_extent *old = exts;
4338                         max *= 2;
4339                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4340                         memcpy(exts, old, sizeof(*exts) * nr);
4341                         if (old != *extents)
4342                                 kfree(old);
4343                 }
4344
4345                 exts[nr].disk_bytenr =
4346                         btrfs_file_extent_disk_bytenr(leaf, fi);
4347                 exts[nr].disk_num_bytes =
4348                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4349                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4350                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4351                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4352                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4353                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4354                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4355                                                                            fi);
4356                 BUG_ON(exts[nr].offset > 0);
4357                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4358                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4359
4360                 cur_pos += exts[nr].num_bytes;
4361                 nr++;
4362
4363                 if (cur_pos + offset >= last_byte)
4364                         break;
4365
4366                 if (no_fragment) {
4367                         ret = 1;
4368                         goto out;
4369                 }
4370                 path->slots[0]++;
4371         }
4372
4373         WARN_ON(cur_pos + offset > last_byte);
4374         if (cur_pos + offset < last_byte) {
4375                 ret = -ENOENT;
4376                 goto out;
4377         }
4378         ret = 0;
4379 out:
4380         btrfs_free_path(path);
4381         if (ret) {
4382                 if (exts != *extents)
4383                         kfree(exts);
4384         } else {
4385                 *extents = exts;
4386                 *nr_extents = nr;
4387         }
4388         return ret;
4389 }
4390
4391 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
4392                                         struct btrfs_root *root,
4393                                         struct btrfs_path *path,
4394                                         struct btrfs_key *extent_key,
4395                                         struct btrfs_key *leaf_key,
4396                                         struct btrfs_ref_path *ref_path,
4397                                         struct disk_extent *new_extents,
4398                                         int nr_extents)
4399 {
4400         struct extent_buffer *leaf;
4401         struct btrfs_file_extent_item *fi;
4402         struct inode *inode = NULL;
4403         struct btrfs_key key;
4404         u64 lock_start = 0;
4405         u64 lock_end = 0;
4406         u64 num_bytes;
4407         u64 ext_offset;
4408         u64 first_pos;
4409         u32 nritems;
4410         int nr_scaned = 0;
4411         int extent_locked = 0;
4412         int extent_type;
4413         int ret;
4414
4415         memcpy(&key, leaf_key, sizeof(key));
4416         first_pos = INT_LIMIT(loff_t) - extent_key->offset;
4417         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4418                 if (key.objectid < ref_path->owner_objectid ||
4419                     (key.objectid == ref_path->owner_objectid &&
4420                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4421                         key.objectid = ref_path->owner_objectid;
4422                         key.type = BTRFS_EXTENT_DATA_KEY;
4423                         key.offset = 0;
4424                 }
4425         }
4426
4427         while (1) {
4428                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4429                 if (ret < 0)
4430                         goto out;
4431
4432                 leaf = path->nodes[0];
4433                 nritems = btrfs_header_nritems(leaf);
4434 next:
4435                 if (extent_locked && ret > 0) {
4436                         /*
4437                          * the file extent item was modified by someone
4438                          * before the extent got locked.
4439                          */
4440                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4441                                       lock_end, GFP_NOFS);
4442                         extent_locked = 0;
4443                 }
4444
4445                 if (path->slots[0] >= nritems) {
4446                         if (++nr_scaned > 2)
4447                                 break;
4448
4449                         BUG_ON(extent_locked);
4450                         ret = btrfs_next_leaf(root, path);
4451                         if (ret < 0)
4452                                 goto out;
4453                         if (ret > 0)
4454                                 break;
4455                         leaf = path->nodes[0];
4456                         nritems = btrfs_header_nritems(leaf);
4457                 }
4458
4459                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4460
4461                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4462                         if ((key.objectid > ref_path->owner_objectid) ||
4463                             (key.objectid == ref_path->owner_objectid &&
4464                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4465                             (key.offset >= first_pos + extent_key->offset))
4466                                 break;
4467                 }
4468
4469                 if (inode && key.objectid != inode->i_ino) {
4470                         BUG_ON(extent_locked);
4471                         btrfs_release_path(root, path);
4472                         mutex_unlock(&inode->i_mutex);
4473                         iput(inode);
4474                         inode = NULL;
4475                         continue;
4476                 }
4477
4478                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4479                         path->slots[0]++;
4480                         ret = 1;
4481                         goto next;
4482                 }
4483                 fi = btrfs_item_ptr(leaf, path->slots[0],
4484                                     struct btrfs_file_extent_item);
4485                 extent_type = btrfs_file_extent_type(leaf, fi);
4486                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4487                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4488                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4489                      extent_key->objectid)) {
4490                         path->slots[0]++;
4491                         ret = 1;
4492                         goto next;
4493                 }
4494
4495                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4496                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4497
4498                 if (first_pos > key.offset - ext_offset)
4499                         first_pos = key.offset - ext_offset;
4500
4501                 if (!extent_locked) {
4502                         lock_start = key.offset;
4503                         lock_end = lock_start + num_bytes - 1;
4504                 } else {
4505                         if (lock_start > key.offset ||
4506                             lock_end + 1 < key.offset + num_bytes) {
4507                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4508                                               lock_start, lock_end, GFP_NOFS);
4509                                 extent_locked = 0;
4510                         }
4511                 }
4512
4513                 if (!inode) {
4514                         btrfs_release_path(root, path);
4515
4516                         inode = btrfs_iget_locked(root->fs_info->sb,
4517                                                   key.objectid, root);
4518                         if (inode->i_state & I_NEW) {
4519                                 BTRFS_I(inode)->root = root;
4520                                 BTRFS_I(inode)->location.objectid =
4521                                         key.objectid;
4522                                 BTRFS_I(inode)->location.type =
4523                                         BTRFS_INODE_ITEM_KEY;
4524                                 BTRFS_I(inode)->location.offset = 0;
4525                                 btrfs_read_locked_inode(inode);
4526                                 unlock_new_inode(inode);
4527                         }
4528                         /*
4529                          * some code call btrfs_commit_transaction while
4530                          * holding the i_mutex, so we can't use mutex_lock
4531                          * here.
4532                          */
4533                         if (is_bad_inode(inode) ||
4534                             !mutex_trylock(&inode->i_mutex)) {
4535                                 iput(inode);
4536                                 inode = NULL;
4537                                 key.offset = (u64)-1;
4538                                 goto skip;
4539                         }
4540                 }
4541
4542                 if (!extent_locked) {
4543                         struct btrfs_ordered_extent *ordered;
4544
4545                         btrfs_release_path(root, path);
4546
4547                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4548                                     lock_end, GFP_NOFS);
4549                         ordered = btrfs_lookup_first_ordered_extent(inode,
4550                                                                     lock_end);
4551                         if (ordered &&
4552                             ordered->file_offset <= lock_end &&
4553                             ordered->file_offset + ordered->len > lock_start) {
4554                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4555                                               lock_start, lock_end, GFP_NOFS);
4556                                 btrfs_start_ordered_extent(inode, ordered, 1);
4557                                 btrfs_put_ordered_extent(ordered);
4558                                 key.offset += num_bytes;
4559                                 goto skip;
4560                         }
4561                         if (ordered)
4562                                 btrfs_put_ordered_extent(ordered);
4563
4564                         extent_locked = 1;
4565                         continue;
4566                 }
4567
4568                 if (nr_extents == 1) {
4569                         /* update extent pointer in place */
4570                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4571                                                 new_extents[0].disk_bytenr);
4572                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4573                                                 new_extents[0].disk_num_bytes);
4574                         btrfs_mark_buffer_dirty(leaf);
4575
4576                         btrfs_drop_extent_cache(inode, key.offset,
4577                                                 key.offset + num_bytes - 1, 0);
4578
4579                         ret = btrfs_inc_extent_ref(trans, root,
4580                                                 new_extents[0].disk_bytenr,
4581                                                 new_extents[0].disk_num_bytes,
4582                                                 leaf->start,
4583                                                 root->root_key.objectid,
4584                                                 trans->transid,
4585                                                 key.objectid);
4586                         BUG_ON(ret);
4587
4588                         ret = btrfs_free_extent(trans, root,
4589                                                 extent_key->objectid,
4590                                                 extent_key->offset,
4591                                                 leaf->start,
4592                                                 btrfs_header_owner(leaf),
4593                                                 btrfs_header_generation(leaf),
4594                                                 key.objectid, 0);
4595                         BUG_ON(ret);
4596
4597                         btrfs_release_path(root, path);
4598                         key.offset += num_bytes;
4599                 } else {
4600                         BUG_ON(1);
4601 #if 0
4602                         u64 alloc_hint;
4603                         u64 extent_len;
4604                         int i;
4605                         /*
4606                          * drop old extent pointer at first, then insert the
4607                          * new pointers one bye one
4608                          */
4609                         btrfs_release_path(root, path);
4610                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4611                                                  key.offset + num_bytes,
4612                                                  key.offset, &alloc_hint);
4613                         BUG_ON(ret);
4614
4615                         for (i = 0; i < nr_extents; i++) {
4616                                 if (ext_offset >= new_extents[i].num_bytes) {
4617                                         ext_offset -= new_extents[i].num_bytes;
4618                                         continue;
4619                                 }
4620                                 extent_len = min(new_extents[i].num_bytes -
4621                                                  ext_offset, num_bytes);
4622
4623                                 ret = btrfs_insert_empty_item(trans, root,
4624                                                               path, &key,
4625                                                               sizeof(*fi));
4626                                 BUG_ON(ret);
4627
4628                                 leaf = path->nodes[0];
4629                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4630                                                 struct btrfs_file_extent_item);
4631                                 btrfs_set_file_extent_generation(leaf, fi,
4632                                                         trans->transid);
4633                                 btrfs_set_file_extent_type(leaf, fi,
4634                                                         BTRFS_FILE_EXTENT_REG);
4635                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4636                                                 new_extents[i].disk_bytenr);
4637                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4638                                                 new_extents[i].disk_num_bytes);
4639                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4640                                                 new_extents[i].ram_bytes);
4641
4642                                 btrfs_set_file_extent_compression(leaf, fi,
4643                                                 new_extents[i].compression);
4644                                 btrfs_set_file_extent_encryption(leaf, fi,
4645                                                 new_extents[i].encryption);
4646                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4647                                                 new_extents[i].other_encoding);
4648
4649                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4650                                                         extent_len);
4651                                 ext_offset += new_extents[i].offset;
4652                                 btrfs_set_file_extent_offset(leaf, fi,
4653                                                         ext_offset);
4654                                 btrfs_mark_buffer_dirty(leaf);
4655
4656                                 btrfs_drop_extent_cache(inode, key.offset,
4657                                                 key.offset + extent_len - 1, 0);
4658
4659                                 ret = btrfs_inc_extent_ref(trans, root,
4660                                                 new_extents[i].disk_bytenr,
4661                                                 new_extents[i].disk_num_bytes,
4662                                                 leaf->start,
4663                                                 root->root_key.objectid,
4664                                                 trans->transid, key.objectid);
4665                                 BUG_ON(ret);
4666                                 btrfs_release_path(root, path);
4667
4668                                 inode_add_bytes(inode, extent_len);
4669
4670                                 ext_offset = 0;
4671                                 num_bytes -= extent_len;
4672                                 key.offset += extent_len;
4673
4674                                 if (num_bytes == 0)
4675                                         break;
4676                         }
4677                         BUG_ON(i >= nr_extents);
4678 #endif
4679                 }
4680
4681                 if (extent_locked) {
4682                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4683                                       lock_end, GFP_NOFS);
4684                         extent_locked = 0;
4685                 }
4686 skip:
4687                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4688                     key.offset >= first_pos + extent_key->offset)
4689                         break;
4690
4691                 cond_resched();
4692         }
4693         ret = 0;
4694 out:
4695         btrfs_release_path(root, path);
4696         if (inode) {
4697                 mutex_unlock(&inode->i_mutex);
4698                 if (extent_locked) {
4699                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4700                                       lock_end, GFP_NOFS);
4701                 }
4702                 iput(inode);
4703         }
4704         return ret;
4705 }
4706
4707 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4708                                struct btrfs_root *root,
4709                                struct extent_buffer *buf, u64 orig_start)
4710 {
4711         int level;
4712         int ret;
4713
4714         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4715         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4716
4717         level = btrfs_header_level(buf);
4718         if (level == 0) {
4719                 struct btrfs_leaf_ref *ref;
4720                 struct btrfs_leaf_ref *orig_ref;
4721
4722                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4723                 if (!orig_ref)
4724                         return -ENOENT;
4725
4726                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4727                 if (!ref) {
4728                         btrfs_free_leaf_ref(root, orig_ref);
4729                         return -ENOMEM;
4730                 }
4731
4732                 ref->nritems = orig_ref->nritems;
4733                 memcpy(ref->extents, orig_ref->extents,
4734                         sizeof(ref->extents[0]) * ref->nritems);
4735
4736                 btrfs_free_leaf_ref(root, orig_ref);
4737
4738                 ref->root_gen = trans->transid;
4739                 ref->bytenr = buf->start;
4740                 ref->owner = btrfs_header_owner(buf);
4741                 ref->generation = btrfs_header_generation(buf);
4742                 ret = btrfs_add_leaf_ref(root, ref, 0);
4743                 WARN_ON(ret);
4744                 btrfs_free_leaf_ref(root, ref);
4745         }
4746         return 0;
4747 }
4748
4749 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4750                                         struct extent_buffer *leaf,
4751                                         struct btrfs_block_group_cache *group,
4752                                         struct btrfs_root *target_root)
4753 {
4754         struct btrfs_key key;
4755         struct inode *inode = NULL;
4756         struct btrfs_file_extent_item *fi;
4757         u64 num_bytes;
4758         u64 skip_objectid = 0;
4759         u32 nritems;
4760         u32 i;
4761
4762         nritems = btrfs_header_nritems(leaf);
4763         for (i = 0; i < nritems; i++) {
4764                 btrfs_item_key_to_cpu(leaf, &key, i);
4765                 if (key.objectid == skip_objectid ||
4766                     key.type != BTRFS_EXTENT_DATA_KEY)
4767                         continue;
4768                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4769                 if (btrfs_file_extent_type(leaf, fi) ==
4770                     BTRFS_FILE_EXTENT_INLINE)
4771                         continue;
4772                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4773                         continue;
4774                 if (!inode || inode->i_ino != key.objectid) {
4775                         iput(inode);
4776                         inode = btrfs_ilookup(target_root->fs_info->sb,
4777                                               key.objectid, target_root, 1);
4778                 }
4779                 if (!inode) {
4780                         skip_objectid = key.objectid;
4781                         continue;
4782                 }
4783                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4784
4785                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4786                             key.offset + num_bytes - 1, GFP_NOFS);
4787                 btrfs_drop_extent_cache(inode, key.offset,
4788                                         key.offset + num_bytes - 1, 1);
4789                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4790                               key.offset + num_bytes - 1, GFP_NOFS);
4791                 cond_resched();
4792         }
4793         iput(inode);
4794         return 0;
4795 }
4796
4797 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4798                                         struct btrfs_root *root,
4799                                         struct extent_buffer *leaf,
4800                                         struct btrfs_block_group_cache *group,
4801                                         struct inode *reloc_inode)
4802 {
4803         struct btrfs_key key;
4804         struct btrfs_key extent_key;
4805         struct btrfs_file_extent_item *fi;
4806         struct btrfs_leaf_ref *ref;
4807         struct disk_extent *new_extent;
4808         u64 bytenr;
4809         u64 num_bytes;
4810         u32 nritems;
4811         u32 i;
4812         int ext_index;
4813         int nr_extent;
4814         int ret;
4815
4816         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4817         BUG_ON(!new_extent);
4818
4819         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4820         BUG_ON(!ref);
4821
4822         ext_index = -1;
4823         nritems = btrfs_header_nritems(leaf);
4824         for (i = 0; i < nritems; i++) {
4825                 btrfs_item_key_to_cpu(leaf, &key, i);
4826                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4827                         continue;
4828                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4829                 if (btrfs_file_extent_type(leaf, fi) ==
4830                     BTRFS_FILE_EXTENT_INLINE)
4831                         continue;
4832                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4833                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4834                 if (bytenr == 0)
4835                         continue;
4836
4837                 ext_index++;
4838                 if (bytenr >= group->key.objectid + group->key.offset ||
4839                     bytenr + num_bytes <= group->key.objectid)
4840                         continue;
4841
4842                 extent_key.objectid = bytenr;
4843                 extent_key.offset = num_bytes;
4844                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4845                 nr_extent = 1;
4846                 ret = get_new_locations(reloc_inode, &extent_key,
4847                                         group->key.objectid, 1,
4848                                         &new_extent, &nr_extent);
4849                 if (ret > 0)
4850                         continue;
4851                 BUG_ON(ret < 0);
4852
4853                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4854                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4855                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4856                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4857
4858                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4859                                                 new_extent->disk_bytenr);
4860                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4861                                                 new_extent->disk_num_bytes);
4862                 btrfs_mark_buffer_dirty(leaf);
4863
4864                 ret = btrfs_inc_extent_ref(trans, root,
4865                                         new_extent->disk_bytenr,
4866                                         new_extent->disk_num_bytes,
4867                                         leaf->start,
4868                                         root->root_key.objectid,
4869                                         trans->transid, key.objectid);
4870                 BUG_ON(ret);
4871                 ret = btrfs_free_extent(trans, root,
4872                                         bytenr, num_bytes, leaf->start,
4873                                         btrfs_header_owner(leaf),
4874                                         btrfs_header_generation(leaf),
4875                                         key.objectid, 0);
4876                 BUG_ON(ret);
4877                 cond_resched();
4878         }
4879         kfree(new_extent);
4880         BUG_ON(ext_index + 1 != ref->nritems);
4881         btrfs_free_leaf_ref(root, ref);
4882         return 0;
4883 }
4884
4885 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4886                           struct btrfs_root *root)
4887 {
4888         struct btrfs_root *reloc_root;
4889         int ret;
4890
4891         if (root->reloc_root) {
4892                 reloc_root = root->reloc_root;
4893                 root->reloc_root = NULL;
4894                 list_add(&reloc_root->dead_list,
4895                          &root->fs_info->dead_reloc_roots);
4896
4897                 btrfs_set_root_bytenr(&reloc_root->root_item,
4898                                       reloc_root->node->start);
4899                 btrfs_set_root_level(&root->root_item,
4900                                      btrfs_header_level(reloc_root->node));
4901                 memset(&reloc_root->root_item.drop_progress, 0,
4902                         sizeof(struct btrfs_disk_key));
4903                 reloc_root->root_item.drop_level = 0;
4904
4905                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4906                                         &reloc_root->root_key,
4907                                         &reloc_root->root_item);
4908                 BUG_ON(ret);
4909         }
4910         return 0;
4911 }
4912
4913 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4914 {
4915         struct btrfs_trans_handle *trans;
4916         struct btrfs_root *reloc_root;
4917         struct btrfs_root *prev_root = NULL;
4918         struct list_head dead_roots;
4919         int ret;
4920         unsigned long nr;
4921
4922         INIT_LIST_HEAD(&dead_roots);
4923         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4924
4925         while (!list_empty(&dead_roots)) {
4926                 reloc_root = list_entry(dead_roots.prev,
4927                                         struct btrfs_root, dead_list);
4928                 list_del_init(&reloc_root->dead_list);
4929
4930                 BUG_ON(reloc_root->commit_root != NULL);
4931                 while (1) {
4932                         trans = btrfs_join_transaction(root, 1);
4933                         BUG_ON(!trans);
4934
4935                         mutex_lock(&root->fs_info->drop_mutex);
4936                         ret = btrfs_drop_snapshot(trans, reloc_root);
4937                         if (ret != -EAGAIN)
4938                                 break;
4939                         mutex_unlock(&root->fs_info->drop_mutex);
4940
4941                         nr = trans->blocks_used;
4942                         ret = btrfs_end_transaction(trans, root);
4943                         BUG_ON(ret);
4944                         btrfs_btree_balance_dirty(root, nr);
4945                 }
4946
4947                 free_extent_buffer(reloc_root->node);
4948
4949                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4950                                      &reloc_root->root_key);
4951                 BUG_ON(ret);
4952                 mutex_unlock(&root->fs_info->drop_mutex);
4953
4954                 nr = trans->blocks_used;
4955                 ret = btrfs_end_transaction(trans, root);
4956                 BUG_ON(ret);
4957                 btrfs_btree_balance_dirty(root, nr);
4958
4959                 kfree(prev_root);
4960                 prev_root = reloc_root;
4961         }
4962         if (prev_root) {
4963                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4964                 kfree(prev_root);
4965         }
4966         return 0;
4967 }
4968
4969 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4970 {
4971         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4972         return 0;
4973 }
4974
4975 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4976 {
4977         struct btrfs_root *reloc_root;
4978         struct btrfs_trans_handle *trans;
4979         struct btrfs_key location;
4980         int found;
4981         int ret;
4982
4983         mutex_lock(&root->fs_info->tree_reloc_mutex);
4984         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4985         BUG_ON(ret);
4986         found = !list_empty(&root->fs_info->dead_reloc_roots);
4987         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4988
4989         if (found) {
4990                 trans = btrfs_start_transaction(root, 1);
4991                 BUG_ON(!trans);
4992                 ret = btrfs_commit_transaction(trans, root);
4993                 BUG_ON(ret);
4994         }
4995
4996         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4997         location.offset = (u64)-1;
4998         location.type = BTRFS_ROOT_ITEM_KEY;
4999
5000         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5001         BUG_ON(!reloc_root);
5002         btrfs_orphan_cleanup(reloc_root);
5003         return 0;
5004 }
5005
5006 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
5007                                     struct btrfs_root *root)
5008 {
5009         struct btrfs_root *reloc_root;
5010         struct extent_buffer *eb;
5011         struct btrfs_root_item *root_item;
5012         struct btrfs_key root_key;
5013         int ret;
5014
5015         BUG_ON(!root->ref_cows);
5016         if (root->reloc_root)
5017                 return 0;
5018
5019         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5020         BUG_ON(!root_item);
5021
5022         ret = btrfs_copy_root(trans, root, root->commit_root,
5023                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5024         BUG_ON(ret);
5025
5026         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5027         root_key.offset = root->root_key.objectid;
5028         root_key.type = BTRFS_ROOT_ITEM_KEY;
5029
5030         memcpy(root_item, &root->root_item, sizeof(root_item));
5031         btrfs_set_root_refs(root_item, 0);
5032         btrfs_set_root_bytenr(root_item, eb->start);
5033         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5034         btrfs_set_root_generation(root_item, trans->transid);
5035
5036         btrfs_tree_unlock(eb);
5037         free_extent_buffer(eb);
5038
5039         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5040                                 &root_key, root_item);
5041         BUG_ON(ret);
5042         kfree(root_item);
5043
5044         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5045                                                  &root_key);
5046         BUG_ON(!reloc_root);
5047         reloc_root->last_trans = trans->transid;
5048         reloc_root->commit_root = NULL;
5049         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5050
5051         root->reloc_root = reloc_root;
5052         return 0;
5053 }
5054
5055 /*
5056  * Core function of space balance.
5057  *
5058  * The idea is using reloc trees to relocate tree blocks in reference
5059  * counted roots. There is one reloc tree for each subvol, and all
5060  * reloc trees share same root key objectid. Reloc trees are snapshots
5061  * of the latest committed roots of subvols (root->commit_root).
5062  *
5063  * To relocate a tree block referenced by a subvol, there are two steps.
5064  * COW the block through subvol's reloc tree, then update block pointer
5065  * in the subvol to point to the new block. Since all reloc trees share
5066  * same root key objectid, doing special handing for tree blocks owned
5067  * by them is easy. Once a tree block has been COWed in one reloc tree,
5068  * we can use the resulting new block directly when the same block is
5069  * required to COW again through other reloc trees. By this way, relocated
5070  * tree blocks are shared between reloc trees, so they are also shared
5071  * between subvols.
5072  */
5073 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
5074                                       struct btrfs_root *root,
5075                                       struct btrfs_path *path,
5076                                       struct btrfs_key *first_key,
5077                                       struct btrfs_ref_path *ref_path,
5078                                       struct btrfs_block_group_cache *group,
5079                                       struct inode *reloc_inode)
5080 {
5081         struct btrfs_root *reloc_root;
5082         struct extent_buffer *eb = NULL;
5083         struct btrfs_key *keys;
5084         u64 *nodes;
5085         int level;
5086         int shared_level;
5087         int lowest_level = 0;
5088         int ret;
5089
5090         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5091                 lowest_level = ref_path->owner_objectid;
5092
5093         if (!root->ref_cows) {
5094                 path->lowest_level = lowest_level;
5095                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5096                 BUG_ON(ret < 0);
5097                 path->lowest_level = 0;
5098                 btrfs_release_path(root, path);
5099                 return 0;
5100         }
5101
5102         mutex_lock(&root->fs_info->tree_reloc_mutex);
5103         ret = init_reloc_tree(trans, root);
5104         BUG_ON(ret);
5105         reloc_root = root->reloc_root;
5106
5107         shared_level = ref_path->shared_level;
5108         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5109
5110         keys = ref_path->node_keys;
5111         nodes = ref_path->new_nodes;
5112         memset(&keys[shared_level + 1], 0,
5113                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5114         memset(&nodes[shared_level + 1], 0,
5115                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5116
5117         if (nodes[lowest_level] == 0) {
5118                 path->lowest_level = lowest_level;
5119                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5120                                         0, 1);
5121                 BUG_ON(ret);
5122                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5123                         eb = path->nodes[level];
5124                         if (!eb || eb == reloc_root->node)
5125                                 break;
5126                         nodes[level] = eb->start;
5127                         if (level == 0)
5128                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5129                         else
5130                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5131                 }
5132                 if (nodes[0] &&
5133                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5134                         eb = path->nodes[0];
5135                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5136                                                       group, reloc_inode);
5137                         BUG_ON(ret);
5138                 }
5139                 btrfs_release_path(reloc_root, path);
5140         } else {
5141                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5142                                        lowest_level);
5143                 BUG_ON(ret);
5144         }
5145
5146         /*
5147          * replace tree blocks in the fs tree with tree blocks in
5148          * the reloc tree.
5149          */
5150         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5151         BUG_ON(ret < 0);
5152
5153         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5154                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5155                                         0, 0);
5156                 BUG_ON(ret);
5157                 extent_buffer_get(path->nodes[0]);
5158                 eb = path->nodes[0];
5159                 btrfs_release_path(reloc_root, path);
5160                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5161                 BUG_ON(ret);
5162                 free_extent_buffer(eb);
5163         }
5164
5165         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5166         path->lowest_level = 0;
5167         return 0;
5168 }
5169
5170 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
5171                                         struct btrfs_root *root,
5172                                         struct btrfs_path *path,
5173                                         struct btrfs_key *first_key,
5174                                         struct btrfs_ref_path *ref_path)
5175 {
5176         int ret;
5177
5178         ret = relocate_one_path(trans, root, path, first_key,
5179                                 ref_path, NULL, NULL);
5180         BUG_ON(ret);
5181
5182         if (root == root->fs_info->extent_root)
5183                 btrfs_extent_post_op(trans, root);
5184
5185         return 0;
5186 }
5187
5188 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
5189                                     struct btrfs_root *extent_root,
5190                                     struct btrfs_path *path,
5191                                     struct btrfs_key *extent_key)
5192 {
5193         int ret;
5194
5195         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5196         if (ret)
5197                 goto out;
5198         ret = btrfs_del_item(trans, extent_root, path);
5199 out:
5200         btrfs_release_path(extent_root, path);
5201         return ret;
5202 }
5203
5204 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
5205                                                 struct btrfs_ref_path *ref_path)
5206 {
5207         struct btrfs_key root_key;
5208
5209         root_key.objectid = ref_path->root_objectid;
5210         root_key.type = BTRFS_ROOT_ITEM_KEY;
5211         if (is_cowonly_root(ref_path->root_objectid))
5212                 root_key.offset = 0;
5213         else
5214                 root_key.offset = (u64)-1;
5215
5216         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5217 }
5218
5219 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
5220                                         struct btrfs_path *path,
5221                                         struct btrfs_key *extent_key,
5222                                         struct btrfs_block_group_cache *group,
5223                                         struct inode *reloc_inode, int pass)
5224 {
5225         struct btrfs_trans_handle *trans;
5226         struct btrfs_root *found_root;
5227         struct btrfs_ref_path *ref_path = NULL;
5228         struct disk_extent *new_extents = NULL;
5229         int nr_extents = 0;
5230         int loops;
5231         int ret;
5232         int level;
5233         struct btrfs_key first_key;
5234         u64 prev_block = 0;
5235
5236
5237         trans = btrfs_start_transaction(extent_root, 1);
5238         BUG_ON(!trans);
5239
5240         if (extent_key->objectid == 0) {
5241                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5242                 goto out;
5243         }
5244
5245         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5246         if (!ref_path) {
5247                ret = -ENOMEM;
5248                goto out;
5249         }
5250
5251         for (loops = 0; ; loops++) {
5252                 if (loops == 0) {
5253                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5254                                                    extent_key->objectid);
5255                 } else {
5256                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5257                 }
5258                 if (ret < 0)
5259                         goto out;
5260                 if (ret > 0)
5261                         break;
5262
5263                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5264                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5265                         continue;
5266
5267                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5268                 BUG_ON(!found_root);
5269                 /*
5270                  * for reference counted tree, only process reference paths
5271                  * rooted at the latest committed root.
5272                  */
5273                 if (found_root->ref_cows &&
5274                     ref_path->root_generation != found_root->root_key.offset)
5275                         continue;
5276
5277                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5278                         if (pass == 0) {
5279                                 /*
5280                                  * copy data extents to new locations
5281                                  */
5282                                 u64 group_start = group->key.objectid;
5283                                 ret = relocate_data_extent(reloc_inode,
5284                                                            extent_key,
5285                                                            group_start);
5286                                 if (ret < 0)
5287                                         goto out;
5288                                 break;
5289                         }
5290                         level = 0;
5291                 } else {
5292                         level = ref_path->owner_objectid;
5293                 }
5294
5295                 if (prev_block != ref_path->nodes[level]) {
5296                         struct extent_buffer *eb;
5297                         u64 block_start = ref_path->nodes[level];
5298                         u64 block_size = btrfs_level_size(found_root, level);
5299
5300                         eb = read_tree_block(found_root, block_start,
5301                                              block_size, 0);
5302                         btrfs_tree_lock(eb);
5303                         BUG_ON(level != btrfs_header_level(eb));
5304
5305                         if (level == 0)
5306                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5307                         else
5308                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5309
5310                         btrfs_tree_unlock(eb);
5311                         free_extent_buffer(eb);
5312                         prev_block = block_start;
5313                 }
5314
5315                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
5316                     pass >= 2) {
5317                         /*
5318                          * use fallback method to process the remaining
5319                          * references.
5320                          */
5321                         if (!new_extents) {
5322                                 u64 group_start = group->key.objectid;
5323                                 new_extents = kmalloc(sizeof(*new_extents),
5324                                                       GFP_NOFS);
5325                                 nr_extents = 1;
5326                                 ret = get_new_locations(reloc_inode,
5327                                                         extent_key,
5328                                                         group_start, 1,
5329                                                         &new_extents,
5330                                                         &nr_extents);
5331                                 if (ret)
5332                                         goto out;
5333                         }
5334                         btrfs_record_root_in_trans(found_root);
5335                         ret = replace_one_extent(trans, found_root,
5336                                                 path, extent_key,
5337                                                 &first_key, ref_path,
5338                                                 new_extents, nr_extents);
5339                         if (ret < 0)
5340                                 goto out;
5341                         continue;
5342                 }
5343
5344                 btrfs_record_root_in_trans(found_root);
5345                 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5346                         ret = relocate_tree_block(trans, found_root, path,
5347                                                   &first_key, ref_path);
5348                 } else {
5349                         /*
5350                          * try to update data extent references while
5351                          * keeping metadata shared between snapshots.
5352                          */
5353                         ret = relocate_one_path(trans, found_root, path,
5354                                                 &first_key, ref_path,
5355                                                 group, reloc_inode);
5356                 }
5357                 if (ret < 0)
5358                         goto out;
5359         }
5360         ret = 0;
5361 out:
5362         btrfs_end_transaction(trans, extent_root);
5363         kfree(new_extents);
5364         kfree(ref_path);
5365         return ret;
5366 }
5367
5368 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5369 {
5370         u64 num_devices;
5371         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5372                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5373
5374         num_devices = root->fs_info->fs_devices->rw_devices;
5375         if (num_devices == 1) {
5376                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5377                 stripped = flags & ~stripped;
5378
5379                 /* turn raid0 into single device chunks */
5380                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5381                         return stripped;
5382
5383                 /* turn mirroring into duplication */
5384                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5385                              BTRFS_BLOCK_GROUP_RAID10))
5386                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5387                 return flags;
5388         } else {
5389                 /* they already had raid on here, just return */
5390                 if (flags & stripped)
5391                         return flags;
5392
5393                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5394                 stripped = flags & ~stripped;
5395
5396                 /* switch duplicated blocks with raid1 */
5397                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5398                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5399
5400                 /* turn single device chunks into raid0 */
5401                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5402         }
5403         return flags;
5404 }
5405
5406 int __alloc_chunk_for_shrink(struct btrfs_root *root,
5407                      struct btrfs_block_group_cache *shrink_block_group,
5408                      int force)
5409 {
5410         struct btrfs_trans_handle *trans;
5411         u64 new_alloc_flags;
5412         u64 calc;
5413
5414         spin_lock(&shrink_block_group->lock);
5415         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5416                 spin_unlock(&shrink_block_group->lock);
5417
5418                 trans = btrfs_start_transaction(root, 1);
5419                 spin_lock(&shrink_block_group->lock);
5420
5421                 new_alloc_flags = update_block_group_flags(root,
5422                                                    shrink_block_group->flags);
5423                 if (new_alloc_flags != shrink_block_group->flags) {
5424                         calc =
5425                              btrfs_block_group_used(&shrink_block_group->item);
5426                 } else {
5427                         calc = shrink_block_group->key.offset;
5428                 }
5429                 spin_unlock(&shrink_block_group->lock);
5430
5431                 do_chunk_alloc(trans, root->fs_info->extent_root,
5432                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5433
5434                 btrfs_end_transaction(trans, root);
5435         } else
5436                 spin_unlock(&shrink_block_group->lock);
5437         return 0;
5438 }
5439
5440 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5441                                  struct btrfs_root *root,
5442                                  u64 objectid, u64 size)
5443 {
5444         struct btrfs_path *path;
5445         struct btrfs_inode_item *item;
5446         struct extent_buffer *leaf;
5447         int ret;
5448
5449         path = btrfs_alloc_path();
5450         if (!path)
5451                 return -ENOMEM;
5452
5453         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5454         if (ret)
5455                 goto out;
5456
5457         leaf = path->nodes[0];
5458         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5459         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5460         btrfs_set_inode_generation(leaf, item, 1);
5461         btrfs_set_inode_size(leaf, item, size);
5462         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5463         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM |
5464                                           BTRFS_INODE_NOCOMPRESS);
5465         btrfs_mark_buffer_dirty(leaf);
5466         btrfs_release_path(root, path);
5467 out:
5468         btrfs_free_path(path);
5469         return ret;
5470 }
5471
5472 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
5473                                         struct btrfs_block_group_cache *group)
5474 {
5475         struct inode *inode = NULL;
5476         struct btrfs_trans_handle *trans;
5477         struct btrfs_root *root;
5478         struct btrfs_key root_key;
5479         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5480         int err = 0;
5481
5482         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5483         root_key.type = BTRFS_ROOT_ITEM_KEY;
5484         root_key.offset = (u64)-1;
5485         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5486         if (IS_ERR(root))
5487                 return ERR_CAST(root);
5488
5489         trans = btrfs_start_transaction(root, 1);
5490         BUG_ON(!trans);
5491
5492         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5493         if (err)
5494                 goto out;
5495
5496         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5497         BUG_ON(err);
5498
5499         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5500                                        group->key.offset, 0, group->key.offset,
5501                                        0, 0, 0);
5502         BUG_ON(err);
5503
5504         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5505         if (inode->i_state & I_NEW) {
5506                 BTRFS_I(inode)->root = root;
5507                 BTRFS_I(inode)->location.objectid = objectid;
5508                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5509                 BTRFS_I(inode)->location.offset = 0;
5510                 btrfs_read_locked_inode(inode);
5511                 unlock_new_inode(inode);
5512                 BUG_ON(is_bad_inode(inode));
5513         } else {
5514                 BUG_ON(1);
5515         }
5516
5517         err = btrfs_orphan_add(trans, inode);
5518 out:
5519         btrfs_end_transaction(trans, root);
5520         if (err) {
5521                 if (inode)
5522                         iput(inode);
5523                 inode = ERR_PTR(err);
5524         }
5525         return inode;
5526 }
5527
5528 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5529 {
5530         struct btrfs_trans_handle *trans;
5531         struct btrfs_path *path;
5532         struct btrfs_fs_info *info = root->fs_info;
5533         struct extent_buffer *leaf;
5534         struct inode *reloc_inode;
5535         struct btrfs_block_group_cache *block_group;
5536         struct btrfs_key key;
5537         u64 skipped;
5538         u64 cur_byte;
5539         u64 total_found;
5540         u32 nritems;
5541         int ret;
5542         int progress;
5543         int pass = 0;
5544
5545         root = root->fs_info->extent_root;
5546
5547         block_group = btrfs_lookup_block_group(info, group_start);
5548         BUG_ON(!block_group);
5549
5550         printk("btrfs relocating block group %llu flags %llu\n",
5551                (unsigned long long)block_group->key.objectid,
5552                (unsigned long long)block_group->flags);
5553
5554         path = btrfs_alloc_path();
5555         BUG_ON(!path);
5556
5557         reloc_inode = create_reloc_inode(info, block_group);
5558         BUG_ON(IS_ERR(reloc_inode));
5559
5560         __alloc_chunk_for_shrink(root, block_group, 1);
5561         set_block_group_readonly(block_group);
5562
5563         btrfs_start_delalloc_inodes(info->tree_root);
5564         btrfs_wait_ordered_extents(info->tree_root, 0);
5565 again:
5566         skipped = 0;
5567         total_found = 0;
5568         progress = 0;
5569         key.objectid = block_group->key.objectid;
5570         key.offset = 0;
5571         key.type = 0;
5572         cur_byte = key.objectid;
5573
5574         trans = btrfs_start_transaction(info->tree_root, 1);
5575         btrfs_commit_transaction(trans, info->tree_root);
5576
5577         mutex_lock(&root->fs_info->cleaner_mutex);
5578         btrfs_clean_old_snapshots(info->tree_root);
5579         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5580         mutex_unlock(&root->fs_info->cleaner_mutex);
5581
5582         while(1) {
5583                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5584                 if (ret < 0)
5585                         goto out;
5586 next:
5587                 leaf = path->nodes[0];
5588                 nritems = btrfs_header_nritems(leaf);
5589                 if (path->slots[0] >= nritems) {
5590                         ret = btrfs_next_leaf(root, path);
5591                         if (ret < 0)
5592                                 goto out;
5593                         if (ret == 1) {
5594                                 ret = 0;
5595                                 break;
5596                         }
5597                         leaf = path->nodes[0];
5598                         nritems = btrfs_header_nritems(leaf);
5599                 }
5600
5601                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5602
5603                 if (key.objectid >= block_group->key.objectid +
5604                     block_group->key.offset)
5605                         break;
5606
5607                 if (progress && need_resched()) {
5608                         btrfs_release_path(root, path);
5609                         cond_resched();
5610                         progress = 0;
5611                         continue;
5612                 }
5613                 progress = 1;
5614
5615                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5616                     key.objectid + key.offset <= cur_byte) {
5617                         path->slots[0]++;
5618                         goto next;
5619                 }
5620
5621                 total_found++;
5622                 cur_byte = key.objectid + key.offset;
5623                 btrfs_release_path(root, path);
5624
5625                 __alloc_chunk_for_shrink(root, block_group, 0);
5626                 ret = relocate_one_extent(root, path, &key, block_group,
5627                                           reloc_inode, pass);
5628                 BUG_ON(ret < 0);
5629                 if (ret > 0)
5630                         skipped++;
5631
5632                 key.objectid = cur_byte;
5633                 key.type = 0;
5634                 key.offset = 0;
5635         }
5636
5637         btrfs_release_path(root, path);
5638
5639         if (pass == 0) {
5640                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5641                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5642                 WARN_ON(reloc_inode->i_mapping->nrpages);
5643         }
5644
5645         if (total_found > 0) {
5646                 printk("btrfs found %llu extents in pass %d\n",
5647                        (unsigned long long)total_found, pass);
5648                 pass++;
5649                 if (total_found == skipped && pass > 2) {
5650                         iput(reloc_inode);
5651                         reloc_inode = create_reloc_inode(info, block_group);
5652                         pass = 0;
5653                 }
5654                 goto again;
5655         }
5656
5657         /* delete reloc_inode */
5658         iput(reloc_inode);
5659
5660         /* unpin extents in this range */
5661         trans = btrfs_start_transaction(info->tree_root, 1);
5662         btrfs_commit_transaction(trans, info->tree_root);
5663
5664         spin_lock(&block_group->lock);
5665         WARN_ON(block_group->pinned > 0);
5666         WARN_ON(block_group->reserved > 0);
5667         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5668         spin_unlock(&block_group->lock);
5669         ret = 0;
5670 out:
5671         btrfs_free_path(path);
5672         return ret;
5673 }
5674
5675 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5676                            struct btrfs_key *key)
5677 {
5678         int ret = 0;
5679         struct btrfs_key found_key;
5680         struct extent_buffer *leaf;
5681         int slot;
5682
5683         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5684         if (ret < 0)
5685                 goto out;
5686
5687         while(1) {
5688                 slot = path->slots[0];
5689                 leaf = path->nodes[0];
5690                 if (slot >= btrfs_header_nritems(leaf)) {
5691                         ret = btrfs_next_leaf(root, path);
5692                         if (ret == 0)
5693                                 continue;
5694                         if (ret < 0)
5695                                 goto out;
5696                         break;
5697                 }
5698                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5699
5700                 if (found_key.objectid >= key->objectid &&
5701                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5702                         ret = 0;
5703                         goto out;
5704                 }
5705                 path->slots[0]++;
5706         }
5707         ret = -ENOENT;
5708 out:
5709         return ret;
5710 }
5711
5712 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5713 {
5714         struct btrfs_block_group_cache *block_group;
5715         struct rb_node *n;
5716
5717         spin_lock(&info->block_group_cache_lock);
5718         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5719                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5720                                        cache_node);
5721                 rb_erase(&block_group->cache_node,
5722                          &info->block_group_cache_tree);
5723                 spin_unlock(&info->block_group_cache_lock);
5724
5725                 btrfs_remove_free_space_cache(block_group);
5726                 down_write(&block_group->space_info->groups_sem);
5727                 list_del(&block_group->list);
5728                 up_write(&block_group->space_info->groups_sem);
5729                 kfree(block_group);
5730
5731                 spin_lock(&info->block_group_cache_lock);
5732         }
5733         spin_unlock(&info->block_group_cache_lock);
5734         return 0;
5735 }
5736
5737 int btrfs_read_block_groups(struct btrfs_root *root)
5738 {
5739         struct btrfs_path *path;
5740         int ret;
5741         struct btrfs_block_group_cache *cache;
5742         struct btrfs_fs_info *info = root->fs_info;
5743         struct btrfs_space_info *space_info;
5744         struct btrfs_key key;
5745         struct btrfs_key found_key;
5746         struct extent_buffer *leaf;
5747
5748         root = info->extent_root;
5749         key.objectid = 0;
5750         key.offset = 0;
5751         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5752         path = btrfs_alloc_path();
5753         if (!path)
5754                 return -ENOMEM;
5755
5756         while(1) {
5757                 ret = find_first_block_group(root, path, &key);
5758                 if (ret > 0) {
5759                         ret = 0;
5760                         goto error;
5761                 }
5762                 if (ret != 0)
5763                         goto error;
5764
5765                 leaf = path->nodes[0];
5766                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5767                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5768                 if (!cache) {
5769                         ret = -ENOMEM;
5770                         break;
5771                 }
5772
5773                 spin_lock_init(&cache->lock);
5774                 mutex_init(&cache->alloc_mutex);
5775                 INIT_LIST_HEAD(&cache->list);
5776                 read_extent_buffer(leaf, &cache->item,
5777                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5778                                    sizeof(cache->item));
5779                 memcpy(&cache->key, &found_key, sizeof(found_key));
5780
5781                 key.objectid = found_key.objectid + found_key.offset;
5782                 btrfs_release_path(root, path);
5783                 cache->flags = btrfs_block_group_flags(&cache->item);
5784
5785                 ret = update_space_info(info, cache->flags, found_key.offset,
5786                                         btrfs_block_group_used(&cache->item),
5787                                         &space_info);
5788                 BUG_ON(ret);
5789                 cache->space_info = space_info;
5790                 down_write(&space_info->groups_sem);
5791                 list_add_tail(&cache->list, &space_info->block_groups);
5792                 up_write(&space_info->groups_sem);
5793
5794                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5795                 BUG_ON(ret);
5796
5797                 set_avail_alloc_bits(root->fs_info, cache->flags);
5798                 if (btrfs_chunk_readonly(root, cache->key.objectid))
5799                         set_block_group_readonly(cache);
5800         }
5801         ret = 0;
5802 error:
5803         btrfs_free_path(path);
5804         return ret;
5805 }
5806
5807 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5808                            struct btrfs_root *root, u64 bytes_used,
5809                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5810                            u64 size)
5811 {
5812         int ret;
5813         struct btrfs_root *extent_root;
5814         struct btrfs_block_group_cache *cache;
5815
5816         extent_root = root->fs_info->extent_root;
5817
5818         root->fs_info->last_trans_new_blockgroup = trans->transid;
5819
5820         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5821         if (!cache)
5822                 return -ENOMEM;
5823
5824         cache->key.objectid = chunk_offset;
5825         cache->key.offset = size;
5826         spin_lock_init(&cache->lock);
5827         mutex_init(&cache->alloc_mutex);
5828         INIT_LIST_HEAD(&cache->list);
5829         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5830
5831         btrfs_set_block_group_used(&cache->item, bytes_used);
5832         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5833         cache->flags = type;
5834         btrfs_set_block_group_flags(&cache->item, type);
5835
5836         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5837                                 &cache->space_info);
5838         BUG_ON(ret);
5839         down_write(&cache->space_info->groups_sem);
5840         list_add_tail(&cache->list, &cache->space_info->block_groups);
5841         up_write(&cache->space_info->groups_sem);
5842
5843         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5844         BUG_ON(ret);
5845
5846         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5847                                 sizeof(cache->item));
5848         BUG_ON(ret);
5849
5850         finish_current_insert(trans, extent_root, 0);
5851         ret = del_pending_extents(trans, extent_root, 0);
5852         BUG_ON(ret);
5853         set_avail_alloc_bits(extent_root->fs_info, type);
5854
5855         return 0;
5856 }
5857
5858 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5859                              struct btrfs_root *root, u64 group_start)
5860 {
5861         struct btrfs_path *path;
5862         struct btrfs_block_group_cache *block_group;
5863         struct btrfs_key key;
5864         int ret;
5865
5866         root = root->fs_info->extent_root;
5867
5868         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5869         BUG_ON(!block_group);
5870         BUG_ON(!block_group->ro);
5871
5872         memcpy(&key, &block_group->key, sizeof(key));
5873
5874         path = btrfs_alloc_path();
5875         BUG_ON(!path);
5876
5877         btrfs_remove_free_space_cache(block_group);
5878         rb_erase(&block_group->cache_node,
5879                  &root->fs_info->block_group_cache_tree);
5880         down_write(&block_group->space_info->groups_sem);
5881         list_del(&block_group->list);
5882         up_write(&block_group->space_info->groups_sem);
5883
5884         spin_lock(&block_group->space_info->lock);
5885         block_group->space_info->total_bytes -= block_group->key.offset;
5886         block_group->space_info->bytes_readonly -= block_group->key.offset;
5887         spin_unlock(&block_group->space_info->lock);
5888         block_group->space_info->full = 0;
5889
5890         /*
5891         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5892         kfree(shrink_block_group);
5893         */
5894
5895         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5896         if (ret > 0)
5897                 ret = -EIO;
5898         if (ret < 0)
5899                 goto out;
5900
5901         ret = btrfs_del_item(trans, root, path);
5902 out:
5903         btrfs_free_path(path);
5904         return ret;
5905 }