SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include "hash.h"
  22 #include "crc32c.h"
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "print-tree.h"
  26 #include "transaction.h"
  27 #include "volumes.h"
  28 #include "locking.h"
  29
  30 #define BLOCK_GROUP_DATA     EXTENT_WRITEBACK
  31 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
  32 #define BLOCK_GROUP_SYSTEM   EXTENT_NEW
  33
  34 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
  35
  36 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  37                                  btrfs_root *extent_root);
  38 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  39                                btrfs_root *extent_root);
  40 static struct btrfs_block_group_cache *
  41 __btrfs_find_block_group(struct btrfs_root *root,
  42                          struct btrfs_block_group_cache *hint,
  43                          u64 search_start, int data, int owner);
  44
  45 void maybe_lock_mutex(struct btrfs_root *root)
  46 {
  47         if (root != root->fs_info->extent_root &&
  48             root != root->fs_info->chunk_root &&
  49             root != root->fs_info->dev_root) {
  50                 mutex_lock(&root->fs_info->alloc_mutex);
  51         }
  52 }
  53
  54 void maybe_unlock_mutex(struct btrfs_root *root)
  55 {
  56         if (root != root->fs_info->extent_root &&
  57             root != root->fs_info->chunk_root &&
  58             root != root->fs_info->dev_root) {
  59                 mutex_unlock(&root->fs_info->alloc_mutex);
  60         }
  61 }
  62
  63 static int cache_block_group(struct btrfs_root *root,
  64                              struct btrfs_block_group_cache *block_group)
  65 {
  66         struct btrfs_path *path;
  67         int ret;
  68         struct btrfs_key key;
  69         struct extent_buffer *leaf;
  70         struct extent_io_tree *free_space_cache;
  71         int slot;
  72         u64 last = 0;
  73         u64 hole_size;
  74         u64 first_free;
  75         int found = 0;
  76
  77         if (!block_group)
  78                 return 0;
  79
  80         root = root->fs_info->extent_root;
  81         free_space_cache = &root->fs_info->free_space_cache;
  82
  83         if (block_group->cached)
  84                 return 0;
  85
  86         path = btrfs_alloc_path();
  87         if (!path)
  88                 return -ENOMEM;
  89
  90         path->reada = 2;
  91         first_free = block_group->key.objectid;
  92         key.objectid = block_group->key.objectid;
  93         key.offset = 0;
  94         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  95         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  96         if (ret < 0)
  97                 return ret;
  98         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
  99         if (ret < 0)
 100                 return ret;
 101         if (ret == 0) {
 102                 leaf = path->nodes[0];
 103                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 104                 if (key.objectid + key.offset > first_free)
 105                         first_free = key.objectid + key.offset;
 106         }
 107         while(1) {
 108                 leaf = path->nodes[0];
 109                 slot = path->slots[0];
 110                 if (slot >= btrfs_header_nritems(leaf)) {
 111                         ret = btrfs_next_leaf(root, path);
 112                         if (ret < 0)
 113                                 goto err;
 114                         if (ret == 0) {
 115                                 continue;
 116                         } else {
 117                                 break;
 118                         }
 119                 }
 120                 btrfs_item_key_to_cpu(leaf, &key, slot);
 121                 if (key.objectid < block_group->key.objectid) {
 122                         goto next;
 123                 }
 124                 if (key.objectid >= block_group->key.objectid +
 125                     block_group->key.offset) {
 126                         break;
 127                 }
 128
 129                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 130                         if (!found) {
 131                                 last = first_free;
 132                                 found = 1;
 133                         }
 134                         if (key.objectid > last) {
 135                                 hole_size = key.objectid - last;
 136                                 set_extent_dirty(free_space_cache, last,
 137                                                  last + hole_size - 1,
 138                                                  GFP_NOFS);
 139                         }
 140                         last = key.objectid + key.offset;
 141                 }
 142 next:
 143                 path->slots[0]++;
 144         }
 145
 146         if (!found)
 147                 last = first_free;
 148         if (block_group->key.objectid +
 149             block_group->key.offset > last) {
 150                 hole_size = block_group->key.objectid +
 151                         block_group->key.offset - last;
 152                 set_extent_dirty(free_space_cache, last,
 153                                  last + hole_size - 1, GFP_NOFS);
 154         }
 155         block_group->cached = 1;
 156 err:
 157         btrfs_free_path(path);
 158         return 0;
 159 }
 160
 161 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 162                                                        btrfs_fs_info *info,
 163                                                          u64 bytenr)
 164 {
 165         struct extent_io_tree *block_group_cache;
 166         struct btrfs_block_group_cache *block_group = NULL;
 167         u64 ptr;
 168         u64 start;
 169         u64 end;
 170         int ret;
 171
 172         bytenr = max_t(u64, bytenr,
 173                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 174         block_group_cache = &info->block_group_cache;
 175         ret = find_first_extent_bit(block_group_cache,
 176                                     bytenr, &start, &end,
 177                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 178                                     BLOCK_GROUP_SYSTEM);
 179         if (ret) {
 180                 return NULL;
 181         }
 182         ret = get_state_private(block_group_cache, start, &ptr);
 183         if (ret)
 184                 return NULL;
 185
 186         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 187         return block_group;
 188 }
 189
 190 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 191                                                          btrfs_fs_info *info,
 192                                                          u64 bytenr)
 193 {
 194         struct extent_io_tree *block_group_cache;
 195         struct btrfs_block_group_cache *block_group = NULL;
 196         u64 ptr;
 197         u64 start;
 198         u64 end;
 199         int ret;
 200
 201         bytenr = max_t(u64, bytenr,
 202                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 203         block_group_cache = &info->block_group_cache;
 204         ret = find_first_extent_bit(block_group_cache,
 205                                     bytenr, &start, &end,
 206                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 207                                     BLOCK_GROUP_SYSTEM);
 208         if (ret) {
 209                 return NULL;
 210         }
 211         ret = get_state_private(block_group_cache, start, &ptr);
 212         if (ret)
 213                 return NULL;
 214
 215         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 216         if (block_group->key.objectid <= bytenr && bytenr <
 217             block_group->key.objectid + block_group->key.offset)
 218                 return block_group;
 219         return NULL;
 220 }
 221
 222 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 223 {
 224         return (cache->flags & bits) == bits;
 225 }
 226
 227 static int noinline find_search_start(struct btrfs_root *root,
 228                               struct btrfs_block_group_cache **cache_ret,
 229                               u64 *start_ret, u64 num, int data)
 230 {
 231         int ret;
 232         struct btrfs_block_group_cache *cache = *cache_ret;
 233         struct extent_io_tree *free_space_cache;
 234         struct extent_state *state;
 235         u64 last;
 236         u64 start = 0;
 237         u64 cache_miss = 0;
 238         u64 total_fs_bytes;
 239         u64 search_start = *start_ret;
 240         int wrapped = 0;
 241
 242         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 243         free_space_cache = &root->fs_info->free_space_cache;
 244
 245         if (!cache)
 246                 goto out;
 247
 248 again:
 249         ret = cache_block_group(root, cache);
 250         if (ret) {
 251                 goto out;
 252         }
 253
 254         last = max(search_start, cache->key.objectid);
 255         if (!block_group_bits(cache, data) || cache->ro)
 256                 goto new_group;
 257
 258         spin_lock_irq(&free_space_cache->lock);
 259         state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
 260         while(1) {
 261                 if (!state) {
 262                         if (!cache_miss)
 263                                 cache_miss = last;
 264                         spin_unlock_irq(&free_space_cache->lock);
 265                         goto new_group;
 266                 }
 267
 268                 start = max(last, state->start);
 269                 last = state->end + 1;
 270                 if (last - start < num) {
 271                         do {
 272                                 state = extent_state_next(state);
 273                         } while(state && !(state->state & EXTENT_DIRTY));
 274                         continue;
 275                 }
 276                 spin_unlock_irq(&free_space_cache->lock);
 277                 if (cache->ro) {
 278                         goto new_group;
 279                 }
 280                 if (start + num > cache->key.objectid + cache->key.offset)
 281                         goto new_group;
 282                 if (!block_group_bits(cache, data)) {
 283                         printk("block group bits don't match %Lu %d\n", cache->flags, data);
 284                 }
 285                 *start_ret = start;
 286                 return 0;
 287         }
 288 out:
 289         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 290         if (!cache) {
 291                 printk("Unable to find block group for %Lu\n", search_start);
 292                 WARN_ON(1);
 293         }
 294         return -ENOSPC;
 295
 296 new_group:
 297         last = cache->key.objectid + cache->key.offset;
 298 wrapped:
 299         cache = btrfs_lookup_first_block_group(root->fs_info, last);
 300         if (!cache || cache->key.objectid >= total_fs_bytes) {
 301 no_cache:
 302                 if (!wrapped) {
 303                         wrapped = 1;
 304                         last = search_start;
 305                         goto wrapped;
 306                 }
 307                 goto out;
 308         }
 309         if (cache_miss && !cache->cached) {
 310                 cache_block_group(root, cache);
 311                 last = cache_miss;
 312                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 313         }
 314         cache_miss = 0;
 315         cache = __btrfs_find_block_group(root, cache, last, data, 0);
 316         if (!cache)
 317                 goto no_cache;
 318         *cache_ret = cache;
 319         goto again;
 320 }
 321
 322 static u64 div_factor(u64 num, int factor)
 323 {
 324         if (factor == 10)
 325                 return num;
 326         num *= factor;
 327         do_div(num, 10);
 328         return num;
 329 }
 330
 331 static int block_group_state_bits(u64 flags)
 332 {
 333         int bits = 0;
 334         if (flags & BTRFS_BLOCK_GROUP_DATA)
 335                 bits |= BLOCK_GROUP_DATA;
 336         if (flags & BTRFS_BLOCK_GROUP_METADATA)
 337                 bits |= BLOCK_GROUP_METADATA;
 338         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
 339                 bits |= BLOCK_GROUP_SYSTEM;
 340         return bits;
 341 }
 342
 343 static struct btrfs_block_group_cache *
 344 __btrfs_find_block_group(struct btrfs_root *root,
 345                          struct btrfs_block_group_cache *hint,
 346                          u64 search_start, int data, int owner)
 347 {
 348         struct btrfs_block_group_cache *cache;
 349         struct extent_io_tree *block_group_cache;
 350         struct btrfs_block_group_cache *found_group = NULL;
 351         struct btrfs_fs_info *info = root->fs_info;
 352         u64 used;
 353         u64 last = 0;
 354         u64 start;
 355         u64 end;
 356         u64 free_check;
 357         u64 ptr;
 358         int bit;
 359         int ret;
 360         int full_search = 0;
 361         int factor = 10;
 362         int wrapped = 0;
 363
 364         block_group_cache = &info->block_group_cache;
 365
 366         if (data & BTRFS_BLOCK_GROUP_METADATA)
 367                 factor = 9;
 368
 369         bit = block_group_state_bits(data);
 370
 371         if (search_start) {
 372                 struct btrfs_block_group_cache *shint;
 373                 shint = btrfs_lookup_first_block_group(info, search_start);
 374                 if (shint && block_group_bits(shint, data) && !shint->ro) {
 375                         used = btrfs_block_group_used(&shint->item);
 376                         if (used + shint->pinned <
 377                             div_factor(shint->key.offset, factor)) {
 378                                 return shint;
 379                         }
 380                 }
 381         }
 382         if (hint && !hint->ro && block_group_bits(hint, data)) {
 383                 used = btrfs_block_group_used(&hint->item);
 384                 if (used + hint->pinned <
 385                     div_factor(hint->key.offset, factor)) {
 386                         return hint;
 387                 }
 388                 last = hint->key.objectid + hint->key.offset;
 389         } else {
 390                 if (hint)
 391                         last = max(hint->key.objectid, search_start);
 392                 else
 393                         last = search_start;
 394         }
 395 again:
 396         while(1) {
 397                 ret = find_first_extent_bit(block_group_cache, last,
 398                                             &start, &end, bit);
 399                 if (ret)
 400                         break;
 401
 402                 ret = get_state_private(block_group_cache, start, &ptr);
 403                 if (ret) {
 404                         last = end + 1;
 405                         continue;
 406                 }
 407
 408                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 409                 last = cache->key.objectid + cache->key.offset;
 410                 used = btrfs_block_group_used(&cache->item);
 411
 412                 if (!cache->ro && block_group_bits(cache, data)) {
 413                         free_check = div_factor(cache->key.offset, factor);
 414                         if (used + cache->pinned < free_check) {
 415                                 found_group = cache;
 416                                 goto found;
 417                         }
 418                 }
 419                 cond_resched();
 420         }
 421         if (!wrapped) {
 422                 last = search_start;
 423                 wrapped = 1;
 424                 goto again;
 425         }
 426         if (!full_search && factor < 10) {
 427                 last = search_start;
 428                 full_search = 1;
 429                 factor = 10;
 430                 goto again;
 431         }
 432 found:
 433         return found_group;
 434 }
 435
 436 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 437                                                  struct btrfs_block_group_cache
 438                                                  *hint, u64 search_start,
 439                                                  int data, int owner)
 440 {
 441
 442         struct btrfs_block_group_cache *ret;
 443         mutex_lock(&root->fs_info->alloc_mutex);
 444         ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
 445         mutex_unlock(&root->fs_info->alloc_mutex);
 446         return ret;
 447 }
 448 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
 449                            u64 owner, u64 owner_offset)
 450 {
 451         u32 high_crc = ~(u32)0;
 452         u32 low_crc = ~(u32)0;
 453         __le64 lenum;
 454         lenum = cpu_to_le64(root_objectid);
 455         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
 456         lenum = cpu_to_le64(ref_generation);
 457         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 458         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
 459                 lenum = cpu_to_le64(owner);
 460                 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 461                 lenum = cpu_to_le64(owner_offset);
 462                 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 463         }
 464         return ((u64)high_crc << 32) | (u64)low_crc;
 465 }
 466
 467 static int match_extent_ref(struct extent_buffer *leaf,
 468                             struct btrfs_extent_ref *disk_ref,
 469                             struct btrfs_extent_ref *cpu_ref)
 470 {
 471         int ret;
 472         int len;
 473
 474         if (cpu_ref->objectid)
 475                 len = sizeof(*cpu_ref);
 476         else
 477                 len = 2 * sizeof(u64);
 478         ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
 479                                    len);
 480         return ret == 0;
 481 }
 482
 483 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 484                                           struct btrfs_root *root,
 485                                           struct btrfs_path *path, u64 bytenr,
 486                                           u64 root_objectid,
 487                                           u64 ref_generation, u64 owner,
 488                                           u64 owner_offset, int del)
 489 {
 490         u64 hash;
 491         struct btrfs_key key;
 492         struct btrfs_key found_key;
 493         struct btrfs_extent_ref ref;
 494         struct extent_buffer *leaf;
 495         struct btrfs_extent_ref *disk_ref;
 496         int ret;
 497         int ret2;
 498
 499         btrfs_set_stack_ref_root(&ref, root_objectid);
 500         btrfs_set_stack_ref_generation(&ref, ref_generation);
 501         btrfs_set_stack_ref_objectid(&ref, owner);
 502         btrfs_set_stack_ref_offset(&ref, owner_offset);
 503
 504         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 505                                owner_offset);
 506         key.offset = hash;
 507         key.objectid = bytenr;
 508         key.type = BTRFS_EXTENT_REF_KEY;
 509
 510         while (1) {
 511                 ret = btrfs_search_slot(trans, root, &key, path,
 512                                         del ? -1 : 0, del);
 513                 if (ret < 0)
 514                         goto out;
 515                 leaf = path->nodes[0];
 516                 if (ret != 0) {
 517                         u32 nritems = btrfs_header_nritems(leaf);
 518                         if (path->slots[0] >= nritems) {
 519                                 ret2 = btrfs_next_leaf(root, path);
 520                                 if (ret2)
 521                                         goto out;
 522                                 leaf = path->nodes[0];
 523                         }
 524                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 525                         if (found_key.objectid != bytenr ||
 526                             found_key.type != BTRFS_EXTENT_REF_KEY)
 527                                 goto out;
 528                         key.offset = found_key.offset;
 529                         if (del) {
 530                                 btrfs_release_path(root, path);
 531                                 continue;
 532                         }
 533                 }
 534                 disk_ref = btrfs_item_ptr(path->nodes[0],
 535                                           path->slots[0],
 536                                           struct btrfs_extent_ref);
 537                 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
 538                         ret = 0;
 539                         goto out;
 540                 }
 541                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 542                 key.offset = found_key.offset + 1;
 543                 btrfs_release_path(root, path);
 544         }
 545 out:
 546         return ret;
 547 }
 548
 549 /*
 550  * Back reference rules.  Back refs have three main goals:
 551  *
 552  * 1) differentiate between all holders of references to an extent so that
 553  *    when a reference is dropped we can make sure it was a valid reference
 554  *    before freeing the extent.
 555  *
 556  * 2) Provide enough information to quickly find the holders of an extent
 557  *    if we notice a given block is corrupted or bad.
 558  *
 559  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 560  *    maintenance.  This is actually the same as #2, but with a slightly
 561  *    different use case.
 562  *
 563  * File extents can be referenced by:
 564  *
 565  * - multiple snapshots, subvolumes, or different generations in one subvol
 566  * - different files inside a single subvolume (in theory, not implemented yet)
 567  * - different offsets inside a file (bookend extents in file.c)
 568  *
 569  * The extent ref structure has fields for:
 570  *
 571  * - Objectid of the subvolume root
 572  * - Generation number of the tree holding the reference
 573  * - objectid of the file holding the reference
 574  * - offset in the file corresponding to the key holding the reference
 575  *
 576  * When a file extent is allocated the fields are filled in:
 577  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 578  *
 579  * When a leaf is cow'd new references are added for every file extent found
 580  * in the leaf.  It looks the same as the create case, but trans->transid
 581  * will be different when the block is cow'd.
 582  *
 583  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 584  *
 585  * When a file extent is removed either during snapshot deletion or file
 586  * truncation, the corresponding back reference is found
 587  * by searching for:
 588  *
 589  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 590  *      inode objectid, offset in file)
 591  *
 592  * Btree extents can be referenced by:
 593  *
 594  * - Different subvolumes
 595  * - Different generations of the same subvolume
 596  *
 597  * Storing sufficient information for a full reverse mapping of a btree
 598  * block would require storing the lowest key of the block in the backref,
 599  * and it would require updating that lowest key either before write out or
 600  * every time it changed.  Instead, the objectid of the lowest key is stored
 601  * along with the level of the tree block.  This provides a hint
 602  * about where in the btree the block can be found.  Searches through the
 603  * btree only need to look for a pointer to that block, so they stop one
 604  * level higher than the level recorded in the backref.
 605  *
 606  * Some btrees do not do reference counting on their extents.  These
 607  * include the extent tree and the tree of tree roots.  Backrefs for these
 608  * trees always have a generation of zero.
 609  *
 610  * When a tree block is created, back references are inserted:
 611  *
 612  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
 613  *
 614  * When a tree block is cow'd in a reference counted root,
 615  * new back references are added for all the blocks it points to.
 616  * These are of the form (trans->transid will have increased since creation):
 617  *
 618  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
 619  *
 620  * Because the lowest_key_objectid and the level are just hints
 621  * they are not used when backrefs are deleted.  When a backref is deleted:
 622  *
 623  * if backref was for a tree root:
 624  *     root_objectid = root->root_key.objectid
 625  * else
 626  *     root_objectid = btrfs_header_owner(parent)
 627  *
 628  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
 629  *
 630  * Back Reference Key hashing:
 631  *
 632  * Back references have four fields, each 64 bits long.  Unfortunately,
 633  * This is hashed into a single 64 bit number and placed into the key offset.
 634  * The key objectid corresponds to the first byte in the extent, and the
 635  * key type is set to BTRFS_EXTENT_REF_KEY
 636  */
 637 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
 638                                  struct btrfs_root *root,
 639                                  struct btrfs_path *path, u64 bytenr,
 640                                  u64 root_objectid, u64 ref_generation,
 641                                  u64 owner, u64 owner_offset)
 642 {
 643         u64 hash;
 644         struct btrfs_key key;
 645         struct btrfs_extent_ref ref;
 646         struct btrfs_extent_ref *disk_ref;
 647         int ret;
 648
 649         btrfs_set_stack_ref_root(&ref, root_objectid);
 650         btrfs_set_stack_ref_generation(&ref, ref_generation);
 651         btrfs_set_stack_ref_objectid(&ref, owner);
 652         btrfs_set_stack_ref_offset(&ref, owner_offset);
 653
 654         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 655                                owner_offset);
 656         key.offset = hash;
 657         key.objectid = bytenr;
 658         key.type = BTRFS_EXTENT_REF_KEY;
 659
 660         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
 661         while (ret == -EEXIST) {
 662                 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 663                                           struct btrfs_extent_ref);
 664                 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
 665                         goto out;
 666                 key.offset++;
 667                 btrfs_release_path(root, path);
 668                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 669                                               sizeof(ref));
 670         }
 671         if (ret)
 672                 goto out;
 673         disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 674                                   struct btrfs_extent_ref);
 675         write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
 676                             sizeof(ref));
 677         btrfs_mark_buffer_dirty(path->nodes[0]);
 678 out:
 679         btrfs_release_path(root, path);
 680         return ret;
 681 }
 682
 683 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 684                                 struct btrfs_root *root,
 685                                 u64 bytenr, u64 num_bytes,
 686                                 u64 root_objectid, u64 ref_generation,
 687                                 u64 owner, u64 owner_offset)
 688 {
 689         struct btrfs_path *path;
 690         int ret;
 691         struct btrfs_key key;
 692         struct extent_buffer *l;
 693         struct btrfs_extent_item *item;
 694         u32 refs;
 695
 696         WARN_ON(num_bytes < root->sectorsize);
 697         path = btrfs_alloc_path();
 698         if (!path)
 699                 return -ENOMEM;
 700
 701         path->reada = 1;
 702         key.objectid = bytenr;
 703         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 704         key.offset = num_bytes;
 705         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 706                                 0, 1);
 707         if (ret < 0)
 708                 return ret;
 709         if (ret != 0) {
 710                 BUG();
 711         }
 712         BUG_ON(ret != 0);
 713         l = path->nodes[0];
 714         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 715         refs = btrfs_extent_refs(l, item);
 716         btrfs_set_extent_refs(l, item, refs + 1);
 717         btrfs_mark_buffer_dirty(path->nodes[0]);
 718
 719         btrfs_release_path(root->fs_info->extent_root, path);
 720
 721         path->reada = 1;
 722         ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
 723                                           path, bytenr, root_objectid,
 724                                           ref_generation, owner, owner_offset);
 725         BUG_ON(ret);
 726         finish_current_insert(trans, root->fs_info->extent_root);
 727         del_pending_extents(trans, root->fs_info->extent_root);
 728
 729         btrfs_free_path(path);
 730         return 0;
 731 }
 732
 733 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 734                                 struct btrfs_root *root,
 735                                 u64 bytenr, u64 num_bytes,
 736                                 u64 root_objectid, u64 ref_generation,
 737                                 u64 owner, u64 owner_offset)
 738 {
 739         int ret;
 740
 741         mutex_lock(&root->fs_info->alloc_mutex);
 742         ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
 743                                      root_objectid, ref_generation,
 744                                      owner, owner_offset);
 745         mutex_unlock(&root->fs_info->alloc_mutex);
 746         return ret;
 747 }
 748
 749 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 750                          struct btrfs_root *root)
 751 {
 752         finish_current_insert(trans, root->fs_info->extent_root);
 753         del_pending_extents(trans, root->fs_info->extent_root);
 754         return 0;
 755 }
 756
 757 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 758                              struct btrfs_root *root, u64 bytenr,
 759                              u64 num_bytes, u32 *refs)
 760 {
 761         struct btrfs_path *path;
 762         int ret;
 763         struct btrfs_key key;
 764         struct extent_buffer *l;
 765         struct btrfs_extent_item *item;
 766
 767         WARN_ON(num_bytes < root->sectorsize);
 768         path = btrfs_alloc_path();
 769         path->reada = 1;
 770         key.objectid = bytenr;
 771         key.offset = num_bytes;
 772         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 773         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 774                                 0, 0);
 775         if (ret < 0)
 776                 goto out;
 777         if (ret != 0) {
 778                 btrfs_print_leaf(root, path->nodes[0]);
 779                 printk("failed to find block number %Lu\n", bytenr);
 780                 BUG();
 781         }
 782         l = path->nodes[0];
 783         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 784         *refs = btrfs_extent_refs(l, item);
 785 out:
 786         btrfs_free_path(path);
 787         return 0;
 788 }
 789
 790 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
 791                                   struct btrfs_path *count_path,
 792                                   u64 expected_owner,
 793                                   u64 first_extent)
 794 {
 795         struct btrfs_root *extent_root = root->fs_info->extent_root;
 796         struct btrfs_path *path;
 797         u64 bytenr;
 798         u64 found_objectid;
 799         u64 found_owner;
 800         u64 root_objectid = root->root_key.objectid;
 801         u32 total_count = 0;
 802         u32 extent_refs;
 803         u32 cur_count;
 804         u32 nritems;
 805         int ret;
 806         struct btrfs_key key;
 807         struct btrfs_key found_key;
 808         struct extent_buffer *l;
 809         struct btrfs_extent_item *item;
 810         struct btrfs_extent_ref *ref_item;
 811         int level = -1;
 812
 813         /* FIXME, needs locking */
 814         BUG();
 815
 816         mutex_lock(&root->fs_info->alloc_mutex);
 817         path = btrfs_alloc_path();
 818 again:
 819         if (level == -1)
 820                 bytenr = first_extent;
 821         else
 822                 bytenr = count_path->nodes[level]->start;
 823
 824         cur_count = 0;
 825         key.objectid = bytenr;
 826         key.offset = 0;
 827
 828         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 829         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 830         if (ret < 0)
 831                 goto out;
 832         BUG_ON(ret == 0);
 833
 834         l = path->nodes[0];
 835         btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 836
 837         if (found_key.objectid != bytenr ||
 838             found_key.type != BTRFS_EXTENT_ITEM_KEY) {
 839                 goto out;
 840         }
 841
 842         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 843         extent_refs = btrfs_extent_refs(l, item);
 844         while (1) {
 845                 l = path->nodes[0];
 846                 nritems = btrfs_header_nritems(l);
 847                 if (path->slots[0] >= nritems) {
 848                         ret = btrfs_next_leaf(extent_root, path);
 849                         if (ret == 0)
 850                                 continue;
 851                         break;
 852                 }
 853                 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
 854                 if (found_key.objectid != bytenr)
 855                         break;
 856
 857                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 858                         path->slots[0]++;
 859                         continue;
 860                 }
 861
 862                 cur_count++;
 863                 ref_item = btrfs_item_ptr(l, path->slots[0],
 864                                           struct btrfs_extent_ref);
 865                 found_objectid = btrfs_ref_root(l, ref_item);
 866
 867                 if (found_objectid != root_objectid) {
 868                         total_count = 2;
 869                         goto out;
 870                 }
 871                 if (level == -1) {
 872                         found_owner = btrfs_ref_objectid(l, ref_item);
 873                         if (found_owner != expected_owner) {
 874                                 total_count = 2;
 875                                 goto out;
 876                         }
 877                         /*
 878                          * nasty.  we don't count a reference held by
 879                          * the running transaction.  This allows nodatacow
 880                          * to avoid cow most of the time
 881                          */
 882                         if (found_owner >= BTRFS_FIRST_FREE_OBJECTID &&
 883                             btrfs_ref_generation(l, ref_item) ==
 884                             root->fs_info->generation) {
 885                                 extent_refs--;
 886                         }
 887                 }
 888                 total_count = 1;
 889                 path->slots[0]++;
 890         }
 891         /*
 892          * if there is more than one reference against a data extent,
 893          * we have to assume the other ref is another snapshot
 894          */
 895         if (level == -1 && extent_refs > 1) {
 896                 total_count = 2;
 897                 goto out;
 898         }
 899         if (cur_count == 0) {
 900                 total_count = 0;
 901                 goto out;
 902         }
 903         if (level >= 0 && root->node == count_path->nodes[level])
 904                 goto out;
 905         level++;
 906         btrfs_release_path(root, path);
 907         goto again;
 908
 909 out:
 910         btrfs_free_path(path);
 911         mutex_unlock(&root->fs_info->alloc_mutex);
 912         return total_count;
 913 }
 914
 915 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 916                   struct extent_buffer *buf)
 917 {
 918         u64 bytenr;
 919         u32 nritems;
 920         struct btrfs_key key;
 921         struct btrfs_file_extent_item *fi;
 922         int i;
 923         int level;
 924         int ret;
 925         int faili;
 926
 927         if (!root->ref_cows)
 928                 return 0;
 929
 930         mutex_lock(&root->fs_info->alloc_mutex);
 931         level = btrfs_header_level(buf);
 932         nritems = btrfs_header_nritems(buf);
 933         for (i = 0; i < nritems; i++) {
 934                 if (level == 0) {
 935                         u64 disk_bytenr;
 936                         btrfs_item_key_to_cpu(buf, &key, i);
 937                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 938                                 continue;
 939                         fi = btrfs_item_ptr(buf, i,
 940                                             struct btrfs_file_extent_item);
 941                         if (btrfs_file_extent_type(buf, fi) ==
 942                             BTRFS_FILE_EXTENT_INLINE)
 943                                 continue;
 944                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 945                         if (disk_bytenr == 0)
 946                                 continue;
 947                         ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr,
 948                                     btrfs_file_extent_disk_num_bytes(buf, fi),
 949                                     root->root_key.objectid, trans->transid,
 950                                     key.objectid, key.offset);
 951                         if (ret) {
 952                                 faili = i;
 953                                 goto fail;
 954                         }
 955                 } else {
 956                         bytenr = btrfs_node_blockptr(buf, i);
 957                         btrfs_node_key_to_cpu(buf, &key, i);
 958                         ret = __btrfs_inc_extent_ref(trans, root, bytenr,
 959                                            btrfs_level_size(root, level - 1),
 960                                            root->root_key.objectid,
 961                                            trans->transid,
 962                                            level - 1, key.objectid);
 963                         if (ret) {
 964                                 faili = i;
 965                                 goto fail;
 966                         }
 967                 }
 968         }
 969         mutex_unlock(&root->fs_info->alloc_mutex);
 970         return 0;
 971 fail:
 972         WARN_ON(1);
 973 #if 0
 974         for (i =0; i < faili; i++) {
 975                 if (level == 0) {
 976                         u64 disk_bytenr;
 977                         btrfs_item_key_to_cpu(buf, &key, i);
 978                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 979                                 continue;
 980                         fi = btrfs_item_ptr(buf, i,
 981                                             struct btrfs_file_extent_item);
 982                         if (btrfs_file_extent_type(buf, fi) ==
 983                             BTRFS_FILE_EXTENT_INLINE)
 984                                 continue;
 985                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 986                         if (disk_bytenr == 0)
 987                                 continue;
 988                         err = btrfs_free_extent(trans, root, disk_bytenr,
 989                                     btrfs_file_extent_disk_num_bytes(buf,
 990                                                                       fi), 0);
 991                         BUG_ON(err);
 992                 } else {
 993                         bytenr = btrfs_node_blockptr(buf, i);
 994                         err = btrfs_free_extent(trans, root, bytenr,
 995                                         btrfs_level_size(root, level - 1), 0);
 996                         BUG_ON(err);
 997                 }
 998         }
 999 #endif
1000         mutex_unlock(&root->fs_info->alloc_mutex);
1001         return ret;
1002 }
1003
1004 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1005                                  struct btrfs_root *root,
1006                                  struct btrfs_path *path,
1007                                  struct btrfs_block_group_cache *cache)
1008 {
1009         int ret;
1010         int pending_ret;
1011         struct btrfs_root *extent_root = root->fs_info->extent_root;
1012         unsigned long bi;
1013         struct extent_buffer *leaf;
1014
1015         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1016         if (ret < 0)
1017                 goto fail;
1018         BUG_ON(ret);
1019
1020         leaf = path->nodes[0];
1021         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1022         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1023         btrfs_mark_buffer_dirty(leaf);
1024         btrfs_release_path(extent_root, path);
1025 fail:
1026         finish_current_insert(trans, extent_root);
1027         pending_ret = del_pending_extents(trans, extent_root);
1028         if (ret)
1029                 return ret;
1030         if (pending_ret)
1031                 return pending_ret;
1032         return 0;
1033
1034 }
1035
1036 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1037                                    struct btrfs_root *root)
1038 {
1039         struct extent_io_tree *block_group_cache;
1040         struct btrfs_block_group_cache *cache;
1041         int ret;
1042         int err = 0;
1043         int werr = 0;
1044         struct btrfs_path *path;
1045         u64 last = 0;
1046         u64 start;
1047         u64 end;
1048         u64 ptr;
1049
1050         block_group_cache = &root->fs_info->block_group_cache;
1051         path = btrfs_alloc_path();
1052         if (!path)
1053                 return -ENOMEM;
1054
1055         mutex_lock(&root->fs_info->alloc_mutex);
1056         while(1) {
1057                 ret = find_first_extent_bit(block_group_cache, last,
1058                                             &start, &end, BLOCK_GROUP_DIRTY);
1059                 if (ret)
1060                         break;
1061
1062                 last = end + 1;
1063                 ret = get_state_private(block_group_cache, start, &ptr);
1064                 if (ret)
1065                         break;
1066                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1067                 err = write_one_cache_group(trans, root,
1068                                             path, cache);
1069                 /*
1070                  * if we fail to write the cache group, we want
1071                  * to keep it marked dirty in hopes that a later
1072                  * write will work
1073                  */
1074                 if (err) {
1075                         werr = err;
1076                         continue;
1077                 }
1078                 clear_extent_bits(block_group_cache, start, end,
1079                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
1080         }
1081         btrfs_free_path(path);
1082         mutex_unlock(&root->fs_info->alloc_mutex);
1083         return werr;
1084 }
1085
1086 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1087                                                   u64 flags)
1088 {
1089         struct list_head *head = &info->space_info;
1090         struct list_head *cur;
1091         struct btrfs_space_info *found;
1092         list_for_each(cur, head) {
1093                 found = list_entry(cur, struct btrfs_space_info, list);
1094                 if (found->flags == flags)
1095                         return found;
1096         }
1097         return NULL;
1098
1099 }
1100
1101 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1102                              u64 total_bytes, u64 bytes_used,
1103                              struct btrfs_space_info **space_info)
1104 {
1105         struct btrfs_space_info *found;
1106
1107         found = __find_space_info(info, flags);
1108         if (found) {
1109                 found->total_bytes += total_bytes;
1110                 found->bytes_used += bytes_used;
1111                 found->full = 0;
1112                 WARN_ON(found->total_bytes < found->bytes_used);
1113                 *space_info = found;
1114                 return 0;
1115         }
1116         found = kmalloc(sizeof(*found), GFP_NOFS);
1117         if (!found)
1118                 return -ENOMEM;
1119
1120         list_add(&found->list, &info->space_info);
1121         found->flags = flags;
1122         found->total_bytes = total_bytes;
1123         found->bytes_used = bytes_used;
1124         found->bytes_pinned = 0;
1125         found->full = 0;
1126         found->force_alloc = 0;
1127         *space_info = found;
1128         return 0;
1129 }
1130
1131 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1132 {
1133         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1134                                    BTRFS_BLOCK_GROUP_RAID1 |
1135                                    BTRFS_BLOCK_GROUP_RAID10 |
1136                                    BTRFS_BLOCK_GROUP_DUP);
1137         if (extra_flags) {
1138                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1139                         fs_info->avail_data_alloc_bits |= extra_flags;
1140                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1141                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1142                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1143                         fs_info->avail_system_alloc_bits |= extra_flags;
1144         }
1145 }
1146
1147 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1148 {
1149         u64 num_devices = root->fs_info->fs_devices->num_devices;
1150
1151         if (num_devices == 1)
1152                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1153         if (num_devices < 4)
1154                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1155
1156         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1157             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1158                       BTRFS_BLOCK_GROUP_RAID10))) {
1159                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1160         }
1161
1162         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1163             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1164                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1165         }
1166
1167         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1168             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1169              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1170              (flags & BTRFS_BLOCK_GROUP_DUP)))
1171                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1172         return flags;
1173 }
1174
1175 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1176                           struct btrfs_root *extent_root, u64 alloc_bytes,
1177                           u64 flags, int force)
1178 {
1179         struct btrfs_space_info *space_info;
1180         u64 thresh;
1181         u64 start;
1182         u64 num_bytes;
1183         int ret;
1184
1185         flags = reduce_alloc_profile(extent_root, flags);
1186
1187         space_info = __find_space_info(extent_root->fs_info, flags);
1188         if (!space_info) {
1189                 ret = update_space_info(extent_root->fs_info, flags,
1190                                         0, 0, &space_info);
1191                 BUG_ON(ret);
1192         }
1193         BUG_ON(!space_info);
1194
1195         if (space_info->force_alloc) {
1196                 force = 1;
1197                 space_info->force_alloc = 0;
1198         }
1199         if (space_info->full)
1200                 goto out;
1201
1202         thresh = div_factor(space_info->total_bytes, 6);
1203         if (!force &&
1204            (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1205             thresh)
1206                 goto out;
1207
1208         mutex_lock(&extent_root->fs_info->chunk_mutex);
1209         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1210         if (ret == -ENOSPC) {
1211 printk("space info full %Lu\n", flags);
1212                 space_info->full = 1;
1213                 goto out;
1214         }
1215         BUG_ON(ret);
1216
1217         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1218                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1219         BUG_ON(ret);
1220         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1221 out:
1222         return 0;
1223 }
1224
1225 static int update_block_group(struct btrfs_trans_handle *trans,
1226                               struct btrfs_root *root,
1227                               u64 bytenr, u64 num_bytes, int alloc,
1228                               int mark_free)
1229 {
1230         struct btrfs_block_group_cache *cache;
1231         struct btrfs_fs_info *info = root->fs_info;
1232         u64 total = num_bytes;
1233         u64 old_val;
1234         u64 byte_in_group;
1235         u64 start;
1236         u64 end;
1237
1238         while(total) {
1239                 cache = btrfs_lookup_block_group(info, bytenr);
1240                 if (!cache) {
1241                         return -1;
1242                 }
1243                 byte_in_group = bytenr - cache->key.objectid;
1244                 WARN_ON(byte_in_group > cache->key.offset);
1245                 start = cache->key.objectid;
1246                 end = start + cache->key.offset - 1;
1247                 set_extent_bits(&info->block_group_cache, start, end,
1248                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1249
1250                 old_val = btrfs_block_group_used(&cache->item);
1251                 num_bytes = min(total, cache->key.offset - byte_in_group);
1252                 if (alloc) {
1253                         old_val += num_bytes;
1254                         cache->space_info->bytes_used += num_bytes;
1255                 } else {
1256                         old_val -= num_bytes;
1257                         cache->space_info->bytes_used -= num_bytes;
1258                         if (mark_free) {
1259                                 set_extent_dirty(&info->free_space_cache,
1260                                                  bytenr, bytenr + num_bytes - 1,
1261                                                  GFP_NOFS);
1262                         }
1263                 }
1264                 btrfs_set_block_group_used(&cache->item, old_val);
1265                 total -= num_bytes;
1266                 bytenr += num_bytes;
1267         }
1268         return 0;
1269 }
1270
1271 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1272 {
1273         u64 start;
1274         u64 end;
1275         int ret;
1276         ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1277                                     search_start, &start, &end,
1278                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1279                                     BLOCK_GROUP_SYSTEM);
1280         if (ret)
1281                 return 0;
1282         return start;
1283 }
1284
1285
1286 static int update_pinned_extents(struct btrfs_root *root,
1287                                 u64 bytenr, u64 num, int pin)
1288 {
1289         u64 len;
1290         struct btrfs_block_group_cache *cache;
1291         struct btrfs_fs_info *fs_info = root->fs_info;
1292
1293         if (pin) {
1294                 set_extent_dirty(&fs_info->pinned_extents,
1295                                 bytenr, bytenr + num - 1, GFP_NOFS);
1296         } else {
1297                 clear_extent_dirty(&fs_info->pinned_extents,
1298                                 bytenr, bytenr + num - 1, GFP_NOFS);
1299         }
1300         while (num > 0) {
1301                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1302                 if (!cache) {
1303                         u64 first = first_logical_byte(root, bytenr);
1304                         WARN_ON(first < bytenr);
1305                         len = min(first - bytenr, num);
1306                 } else {
1307                         len = min(num, cache->key.offset -
1308                                   (bytenr - cache->key.objectid));
1309                 }
1310                 if (pin) {
1311                         if (cache) {
1312                                 cache->pinned += len;
1313                                 cache->space_info->bytes_pinned += len;
1314                         }
1315                         fs_info->total_pinned += len;
1316                 } else {
1317                         if (cache) {
1318                                 cache->pinned -= len;
1319                                 cache->space_info->bytes_pinned -= len;
1320                         }
1321                         fs_info->total_pinned -= len;
1322                 }
1323                 bytenr += len;
1324                 num -= len;
1325         }
1326         return 0;
1327 }
1328
1329 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1330 {
1331         u64 last = 0;
1332         u64 start;
1333         u64 end;
1334         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1335         int ret;
1336
1337         while(1) {
1338                 ret = find_first_extent_bit(pinned_extents, last,
1339                                             &start, &end, EXTENT_DIRTY);
1340                 if (ret)
1341                         break;
1342                 set_extent_dirty(copy, start, end, GFP_NOFS);
1343                 last = end + 1;
1344         }
1345         return 0;
1346 }
1347
1348 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1349                                struct btrfs_root *root,
1350                                struct extent_io_tree *unpin)
1351 {
1352         u64 start;
1353         u64 end;
1354         int ret;
1355         struct extent_io_tree *free_space_cache;
1356         free_space_cache = &root->fs_info->free_space_cache;
1357
1358         mutex_lock(&root->fs_info->alloc_mutex);
1359         while(1) {
1360                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1361                                             EXTENT_DIRTY);
1362                 if (ret)
1363                         break;
1364                 update_pinned_extents(root, start, end + 1 - start, 0);
1365                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1366                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1367         }
1368         mutex_unlock(&root->fs_info->alloc_mutex);
1369         return 0;
1370 }
1371
1372 static int finish_current_insert(struct btrfs_trans_handle *trans,
1373                                  struct btrfs_root *extent_root)
1374 {
1375         u64 start;
1376         u64 end;
1377         struct btrfs_fs_info *info = extent_root->fs_info;
1378         struct extent_buffer *eb;
1379         struct btrfs_path *path;
1380         struct btrfs_key ins;
1381         struct btrfs_disk_key first;
1382         struct btrfs_extent_item extent_item;
1383         int ret;
1384         int level;
1385         int err = 0;
1386
1387         btrfs_set_stack_extent_refs(&extent_item, 1);
1388         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1389         path = btrfs_alloc_path();
1390
1391         while(1) {
1392                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1393                                             &end, EXTENT_LOCKED);
1394                 if (ret)
1395                         break;
1396
1397                 ins.objectid = start;
1398                 ins.offset = end + 1 - start;
1399                 err = btrfs_insert_item(trans, extent_root, &ins,
1400                                         &extent_item, sizeof(extent_item));
1401                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1402                                   GFP_NOFS);
1403                 eb = read_tree_block(extent_root, ins.objectid, ins.offset,
1404                                      trans->transid);
1405                 btrfs_tree_lock(eb);
1406                 level = btrfs_header_level(eb);
1407                 if (level == 0) {
1408                         btrfs_item_key(eb, &first, 0);
1409                 } else {
1410                         btrfs_node_key(eb, &first, 0);
1411                 }
1412                 btrfs_tree_unlock(eb);
1413                 free_extent_buffer(eb);
1414                 /*
1415                  * the first key is just a hint, so the race we've created
1416                  * against reading it is fine
1417                  */
1418                 err = btrfs_insert_extent_backref(trans, extent_root, path,
1419                                           start, extent_root->root_key.objectid,
1420                                           0, level,
1421                                           btrfs_disk_key_objectid(&first));
1422                 BUG_ON(err);
1423         }
1424         btrfs_free_path(path);
1425         return 0;
1426 }
1427
1428 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1429                           int pending)
1430 {
1431         int err = 0;
1432
1433         if (!pending) {
1434 #if 0
1435                 struct extent_buffer *buf;
1436                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1437                 if (buf) {
1438                         if (!btrfs_try_tree_lock(buf) &&
1439                             btrfs_buffer_uptodate(buf, 0)) {
1440                                 u64 transid =
1441                                     root->fs_info->running_transaction->transid;
1442                                 u64 header_transid =
1443                                         btrfs_header_generation(buf);
1444                                 if (header_transid == transid &&
1445                                     !btrfs_header_flag(buf,
1446                                                BTRFS_HEADER_FLAG_WRITTEN)) {
1447                                         clean_tree_block(NULL, root, buf);
1448                                         btrfs_tree_unlock(buf);
1449                                         free_extent_buffer(buf);
1450                                         return 1;
1451                                 }
1452                                 btrfs_tree_unlock(buf);
1453                         }
1454                         free_extent_buffer(buf);
1455                 }
1456 #endif
1457                 update_pinned_extents(root, bytenr, num_bytes, 1);
1458         } else {
1459                 set_extent_bits(&root->fs_info->pending_del,
1460                                 bytenr, bytenr + num_bytes - 1,
1461                                 EXTENT_LOCKED, GFP_NOFS);
1462         }
1463         BUG_ON(err < 0);
1464         return 0;
1465 }
1466
1467 /*
1468  * remove an extent from the root, returns 0 on success
1469  */
1470 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1471                          *root, u64 bytenr, u64 num_bytes,
1472                          u64 root_objectid, u64 ref_generation,
1473                          u64 owner_objectid, u64 owner_offset, int pin,
1474                          int mark_free)
1475 {
1476         struct btrfs_path *path;
1477         struct btrfs_key key;
1478         struct btrfs_fs_info *info = root->fs_info;
1479         struct btrfs_root *extent_root = info->extent_root;
1480         struct extent_buffer *leaf;
1481         int ret;
1482         int extent_slot = 0;
1483         int found_extent = 0;
1484         int num_to_del = 1;
1485         struct btrfs_extent_item *ei;
1486         u32 refs;
1487
1488         key.objectid = bytenr;
1489         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1490         key.offset = num_bytes;
1491         path = btrfs_alloc_path();
1492         if (!path)
1493                 return -ENOMEM;
1494
1495         path->reada = 1;
1496         ret = lookup_extent_backref(trans, extent_root, path,
1497                                     bytenr, root_objectid,
1498                                     ref_generation,
1499                                     owner_objectid, owner_offset, 1);
1500         if (ret == 0) {
1501                 struct btrfs_key found_key;
1502                 extent_slot = path->slots[0];
1503                 while(extent_slot > 0) {
1504                         extent_slot--;
1505                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1506                                               extent_slot);
1507                         if (found_key.objectid != bytenr)
1508                                 break;
1509                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1510                             found_key.offset == num_bytes) {
1511                                 found_extent = 1;
1512                                 break;
1513                         }
1514                         if (path->slots[0] - extent_slot > 5)
1515                                 break;
1516                 }
1517                 if (!found_extent)
1518                         ret = btrfs_del_item(trans, extent_root, path);
1519         } else {
1520                 btrfs_print_leaf(extent_root, path->nodes[0]);
1521                 WARN_ON(1);
1522                 printk("Unable to find ref byte nr %Lu root %Lu "
1523                        " gen %Lu owner %Lu offset %Lu\n", bytenr,
1524                        root_objectid, ref_generation, owner_objectid,
1525                        owner_offset);
1526         }
1527         if (!found_extent) {
1528                 btrfs_release_path(extent_root, path);
1529                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1530                 if (ret < 0)
1531                         return ret;
1532                 BUG_ON(ret);
1533                 extent_slot = path->slots[0];
1534         }
1535
1536         leaf = path->nodes[0];
1537         ei = btrfs_item_ptr(leaf, extent_slot,
1538                             struct btrfs_extent_item);
1539         refs = btrfs_extent_refs(leaf, ei);
1540         BUG_ON(refs == 0);
1541         refs -= 1;
1542         btrfs_set_extent_refs(leaf, ei, refs);
1543
1544         btrfs_mark_buffer_dirty(leaf);
1545
1546         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1547                 /* if the back ref and the extent are next to each other
1548                  * they get deleted below in one shot
1549                  */
1550                 path->slots[0] = extent_slot;
1551                 num_to_del = 2;
1552         } else if (found_extent) {
1553                 /* otherwise delete the extent back ref */
1554                 ret = btrfs_del_item(trans, extent_root, path);
1555                 BUG_ON(ret);
1556                 /* if refs are 0, we need to setup the path for deletion */
1557                 if (refs == 0) {
1558                         btrfs_release_path(extent_root, path);
1559                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1560                                                 -1, 1);
1561                         if (ret < 0)
1562                                 return ret;
1563                         BUG_ON(ret);
1564                 }
1565         }
1566
1567         if (refs == 0) {
1568                 u64 super_used;
1569                 u64 root_used;
1570
1571                 if (pin) {
1572                         ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1573                         if (ret > 0)
1574                                 mark_free = 1;
1575                         BUG_ON(ret < 0);
1576                 }
1577
1578                 /* block accounting for super block */
1579                 spin_lock_irq(&info->delalloc_lock);
1580                 super_used = btrfs_super_bytes_used(&info->super_copy);
1581                 btrfs_set_super_bytes_used(&info->super_copy,
1582                                            super_used - num_bytes);
1583                 spin_unlock_irq(&info->delalloc_lock);
1584
1585                 /* block accounting for root item */
1586                 root_used = btrfs_root_used(&root->root_item);
1587                 btrfs_set_root_used(&root->root_item,
1588                                            root_used - num_bytes);
1589                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1590                                       num_to_del);
1591                 if (ret) {
1592                         return ret;
1593                 }
1594                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1595                                          mark_free);
1596                 BUG_ON(ret);
1597         }
1598         btrfs_free_path(path);
1599         finish_current_insert(trans, extent_root);
1600         return ret;
1601 }
1602
1603 /*
1604  * find all the blocks marked as pending in the radix tree and remove
1605  * them from the extent map
1606  */
1607 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1608                                btrfs_root *extent_root)
1609 {
1610         int ret;
1611         int err = 0;
1612         u64 start;
1613         u64 end;
1614         struct extent_io_tree *pending_del;
1615         struct extent_io_tree *pinned_extents;
1616
1617         pending_del = &extent_root->fs_info->pending_del;
1618         pinned_extents = &extent_root->fs_info->pinned_extents;
1619
1620         while(1) {
1621                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1622                                             EXTENT_LOCKED);
1623                 if (ret)
1624                         break;
1625                 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1626                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1627                                   GFP_NOFS);
1628                 ret = __free_extent(trans, extent_root,
1629                                      start, end + 1 - start,
1630                                      extent_root->root_key.objectid,
1631                                      0, 0, 0, 0, 0);
1632                 if (ret)
1633                         err = ret;
1634         }
1635         return err;
1636 }
1637
1638 /*
1639  * remove an extent from the root, returns 0 on success
1640  */
1641 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1642                                struct btrfs_root *root, u64 bytenr,
1643                                u64 num_bytes, u64 root_objectid,
1644                                u64 ref_generation, u64 owner_objectid,
1645                                u64 owner_offset, int pin)
1646 {
1647         struct btrfs_root *extent_root = root->fs_info->extent_root;
1648         int pending_ret;
1649         int ret;
1650
1651         WARN_ON(num_bytes < root->sectorsize);
1652         if (!root->ref_cows)
1653                 ref_generation = 0;
1654
1655         if (root == extent_root) {
1656                 pin_down_bytes(root, bytenr, num_bytes, 1);
1657                 return 0;
1658         }
1659         ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1660                             ref_generation, owner_objectid, owner_offset,
1661                             pin, pin == 0);
1662         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1663         return ret ? ret : pending_ret;
1664 }
1665
1666 int btrfs_free_extent(struct btrfs_trans_handle *trans,
1667                       struct btrfs_root *root, u64 bytenr,
1668                       u64 num_bytes, u64 root_objectid,
1669                       u64 ref_generation, u64 owner_objectid,
1670                       u64 owner_offset, int pin)
1671 {
1672         int ret;
1673
1674         maybe_lock_mutex(root);
1675         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1676                                   root_objectid, ref_generation,
1677                                   owner_objectid, owner_offset, pin);
1678         maybe_unlock_mutex(root);
1679         return ret;
1680 }
1681
1682 static u64 stripe_align(struct btrfs_root *root, u64 val)
1683 {
1684         u64 mask = ((u64)root->stripesize - 1);
1685         u64 ret = (val + mask) & ~mask;
1686         return ret;
1687 }
1688
1689 /*
1690  * walks the btree of allocated extents and find a hole of a given size.
1691  * The key ins is changed to record the hole:
1692  * ins->objectid == block start
1693  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1694  * ins->offset == number of blocks
1695  * Any available blocks before search_start are skipped.
1696  */
1697 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1698                                      struct btrfs_root *orig_root,
1699                                      u64 num_bytes, u64 empty_size,
1700                                      u64 search_start, u64 search_end,
1701                                      u64 hint_byte, struct btrfs_key *ins,
1702                                      u64 exclude_start, u64 exclude_nr,
1703                                      int data)
1704 {
1705         int ret;
1706         u64 orig_search_start;
1707         struct btrfs_root * root = orig_root->fs_info->extent_root;
1708         struct btrfs_fs_info *info = root->fs_info;
1709         u64 total_needed = num_bytes;
1710         u64 *last_ptr = NULL;
1711         struct btrfs_block_group_cache *block_group;
1712         int full_scan = 0;
1713         int wrapped = 0;
1714         int chunk_alloc_done = 0;
1715         int empty_cluster = 2 * 1024 * 1024;
1716         int allowed_chunk_alloc = 0;
1717
1718         WARN_ON(num_bytes < root->sectorsize);
1719         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1720
1721         if (orig_root->ref_cows || empty_size)
1722                 allowed_chunk_alloc = 1;
1723
1724         if (data & BTRFS_BLOCK_GROUP_METADATA) {
1725                 last_ptr = &root->fs_info->last_alloc;
1726                 empty_cluster = 256 * 1024;
1727         }
1728
1729         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1730                 last_ptr = &root->fs_info->last_data_alloc;
1731         }
1732
1733         if (last_ptr) {
1734                 if (*last_ptr)
1735                         hint_byte = *last_ptr;
1736                 else {
1737                         empty_size += empty_cluster;
1738                 }
1739         }
1740
1741         search_start = max(search_start, first_logical_byte(root, 0));
1742         orig_search_start = search_start;
1743
1744         if (search_end == (u64)-1)
1745                 search_end = btrfs_super_total_bytes(&info->super_copy);
1746
1747         if (hint_byte) {
1748                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
1749                 if (!block_group)
1750                         hint_byte = search_start;
1751                 block_group = __btrfs_find_block_group(root, block_group,
1752                                                      hint_byte, data, 1);
1753                 if (last_ptr && *last_ptr == 0 && block_group)
1754                         hint_byte = block_group->key.objectid;
1755         } else {
1756                 block_group = __btrfs_find_block_group(root,
1757                                                      trans->block_group,
1758                                                      search_start, data, 1);
1759         }
1760         search_start = max(search_start, hint_byte);
1761
1762         total_needed += empty_size;
1763
1764 check_failed:
1765         if (!block_group) {
1766                 block_group = btrfs_lookup_first_block_group(info,
1767                                                              search_start);
1768                 if (!block_group)
1769                         block_group = btrfs_lookup_first_block_group(info,
1770                                                        orig_search_start);
1771         }
1772         if (full_scan && !chunk_alloc_done) {
1773                 if (allowed_chunk_alloc) {
1774                         do_chunk_alloc(trans, root,
1775                                      num_bytes + 2 * 1024 * 1024, data, 1);
1776                         allowed_chunk_alloc = 0;
1777                 } else if (block_group && block_group_bits(block_group, data)) {
1778                         block_group->space_info->force_alloc = 1;
1779                 }
1780                 chunk_alloc_done = 1;
1781         }
1782         ret = find_search_start(root, &block_group, &search_start,
1783                                 total_needed, data);
1784         if (ret == -ENOSPC && last_ptr && *last_ptr) {
1785                 *last_ptr = 0;
1786                 block_group = btrfs_lookup_first_block_group(info,
1787                                                              orig_search_start);
1788                 search_start = orig_search_start;
1789                 ret = find_search_start(root, &block_group, &search_start,
1790                                         total_needed, data);
1791         }
1792         if (ret == -ENOSPC)
1793                 goto enospc;
1794         if (ret)
1795                 goto error;
1796
1797         if (last_ptr && *last_ptr && search_start != *last_ptr) {
1798                 *last_ptr = 0;
1799                 if (!empty_size) {
1800                         empty_size += empty_cluster;
1801                         total_needed += empty_size;
1802                 }
1803                 block_group = btrfs_lookup_first_block_group(info,
1804                                                        orig_search_start);
1805                 search_start = orig_search_start;
1806                 ret = find_search_start(root, &block_group,
1807                                         &search_start, total_needed, data);
1808                 if (ret == -ENOSPC)
1809                         goto enospc;
1810                 if (ret)
1811                         goto error;
1812         }
1813
1814         search_start = stripe_align(root, search_start);
1815         ins->objectid = search_start;
1816         ins->offset = num_bytes;
1817
1818         if (ins->objectid + num_bytes >= search_end)
1819                 goto enospc;
1820
1821         if (ins->objectid + num_bytes >
1822             block_group->key.objectid + block_group->key.offset) {
1823                 search_start = block_group->key.objectid +
1824                         block_group->key.offset;
1825                 goto new_group;
1826         }
1827
1828         if (test_range_bit(&info->extent_ins, ins->objectid,
1829                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1830                 search_start = ins->objectid + num_bytes;
1831                 goto new_group;
1832         }
1833
1834         if (test_range_bit(&info->pinned_extents, ins->objectid,
1835                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1836                 search_start = ins->objectid + num_bytes;
1837                 goto new_group;
1838         }
1839
1840         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1841             ins->objectid < exclude_start + exclude_nr)) {
1842                 search_start = exclude_start + exclude_nr;
1843                 goto new_group;
1844         }
1845
1846         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1847                 block_group = btrfs_lookup_block_group(info, ins->objectid);
1848                 if (block_group)
1849                         trans->block_group = block_group;
1850         }
1851         ins->offset = num_bytes;
1852         if (last_ptr) {
1853                 *last_ptr = ins->objectid + ins->offset;
1854                 if (*last_ptr ==
1855                     btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1856                         *last_ptr = 0;
1857                 }
1858         }
1859         return 0;
1860
1861 new_group:
1862         if (search_start + num_bytes >= search_end) {
1863 enospc:
1864                 search_start = orig_search_start;
1865                 if (full_scan) {
1866                         ret = -ENOSPC;
1867                         goto error;
1868                 }
1869                 if (wrapped) {
1870                         if (!full_scan)
1871                                 total_needed -= empty_size;
1872                         full_scan = 1;
1873                 } else
1874                         wrapped = 1;
1875         }
1876         block_group = btrfs_lookup_first_block_group(info, search_start);
1877         cond_resched();
1878         block_group = __btrfs_find_block_group(root, block_group,
1879                                              search_start, data, 0);
1880         goto check_failed;
1881
1882 error:
1883         return ret;
1884 }
1885
1886 /*
1887  * finds a free extent and does all the dirty work required for allocation
1888  * returns the key for the extent through ins, and a tree buffer for
1889  * the first block of the extent through buf.
1890  *
1891  * returns 0 if everything worked, non-zero otherwise.
1892  */
1893 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1894                        struct btrfs_root *root,
1895                        u64 num_bytes, u64 min_alloc_size,
1896                        u64 root_objectid, u64 ref_generation,
1897                        u64 owner, u64 owner_offset,
1898                        u64 empty_size, u64 hint_byte,
1899                        u64 search_end, struct btrfs_key *ins, u64 data)
1900 {
1901         int ret;
1902         int pending_ret;
1903         u64 super_used;
1904         u64 root_used;
1905         u64 search_start = 0;
1906         u64 alloc_profile;
1907         u32 sizes[2];
1908         struct btrfs_fs_info *info = root->fs_info;
1909         struct btrfs_root *extent_root = info->extent_root;
1910         struct btrfs_extent_item *extent_item;
1911         struct btrfs_extent_ref *ref;
1912         struct btrfs_path *path;
1913         struct btrfs_key keys[2];
1914
1915         maybe_lock_mutex(root);
1916
1917         if (data) {
1918                 alloc_profile = info->avail_data_alloc_bits &
1919                                 info->data_alloc_profile;
1920                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1921         } else if (root == root->fs_info->chunk_root) {
1922                 alloc_profile = info->avail_system_alloc_bits &
1923                                 info->system_alloc_profile;
1924                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1925         } else {
1926                 alloc_profile = info->avail_metadata_alloc_bits &
1927                                 info->metadata_alloc_profile;
1928                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1929         }
1930 again:
1931         data = reduce_alloc_profile(root, data);
1932         /*
1933          * the only place that sets empty_size is btrfs_realloc_node, which
1934          * is not called recursively on allocations
1935          */
1936         if (empty_size || root->ref_cows) {
1937                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1938                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1939                                      2 * 1024 * 1024,
1940                                      BTRFS_BLOCK_GROUP_METADATA |
1941                                      (info->metadata_alloc_profile &
1942                                       info->avail_metadata_alloc_bits), 0);
1943                         BUG_ON(ret);
1944                 }
1945                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1946                                      num_bytes + 2 * 1024 * 1024, data, 0);
1947                 BUG_ON(ret);
1948         }
1949
1950         WARN_ON(num_bytes < root->sectorsize);
1951         ret = find_free_extent(trans, root, num_bytes, empty_size,
1952                                search_start, search_end, hint_byte, ins,
1953                                trans->alloc_exclude_start,
1954                                trans->alloc_exclude_nr, data);
1955
1956         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
1957                 num_bytes = num_bytes >> 1;
1958                 num_bytes = max(num_bytes, min_alloc_size);
1959                 do_chunk_alloc(trans, root->fs_info->extent_root,
1960                                num_bytes, data, 1);
1961                 goto again;
1962         }
1963         if (ret) {
1964                 printk("allocation failed flags %Lu\n", data);
1965         }
1966         if (ret) {
1967                 BUG();
1968                 goto out;
1969         }
1970
1971         /* block accounting for super block */
1972         spin_lock_irq(&info->delalloc_lock);
1973         super_used = btrfs_super_bytes_used(&info->super_copy);
1974         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1975         spin_unlock_irq(&info->delalloc_lock);
1976
1977         /* block accounting for root item */
1978         root_used = btrfs_root_used(&root->root_item);
1979         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1980
1981         clear_extent_dirty(&root->fs_info->free_space_cache,
1982                            ins->objectid, ins->objectid + ins->offset - 1,
1983                            GFP_NOFS);
1984
1985         if (root == extent_root) {
1986                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1987                                 ins->objectid + ins->offset - 1,
1988                                 EXTENT_LOCKED, GFP_NOFS);
1989                 goto update_block;
1990         }
1991
1992         WARN_ON(trans->alloc_exclude_nr);
1993         trans->alloc_exclude_start = ins->objectid;
1994         trans->alloc_exclude_nr = ins->offset;
1995
1996         memcpy(&keys[0], ins, sizeof(*ins));
1997         keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1998                                          owner, owner_offset);
1999         keys[1].objectid = ins->objectid;
2000         keys[1].type = BTRFS_EXTENT_REF_KEY;
2001         sizes[0] = sizeof(*extent_item);
2002         sizes[1] = sizeof(*ref);
2003
2004         path = btrfs_alloc_path();
2005         BUG_ON(!path);
2006
2007         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2008                                        sizes, 2);
2009
2010         BUG_ON(ret);
2011         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2012                                      struct btrfs_extent_item);
2013         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2014         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2015                              struct btrfs_extent_ref);
2016
2017         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2018         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2019         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2020         btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2021
2022         btrfs_mark_buffer_dirty(path->nodes[0]);
2023
2024         trans->alloc_exclude_start = 0;
2025         trans->alloc_exclude_nr = 0;
2026         btrfs_free_path(path);
2027         finish_current_insert(trans, extent_root);
2028         pending_ret = del_pending_extents(trans, extent_root);
2029
2030         if (ret)
2031                 goto out;
2032         if (pending_ret) {
2033                 ret = pending_ret;
2034                 goto out;
2035         }
2036
2037 update_block:
2038         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2039         if (ret) {
2040                 printk("update block group failed for %Lu %Lu\n",
2041                        ins->objectid, ins->offset);
2042                 BUG();
2043         }
2044 out:
2045         maybe_unlock_mutex(root);
2046         return ret;
2047 }
2048 /*
2049  * helper function to allocate a block for a given tree
2050  * returns the tree buffer or NULL.
2051  */
2052 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2053                                              struct btrfs_root *root,
2054                                              u32 blocksize,
2055                                              u64 root_objectid,
2056                                              u64 ref_generation,
2057                                              u64 first_objectid,
2058                                              int level,
2059                                              u64 hint,
2060                                              u64 empty_size)
2061 {
2062         struct btrfs_key ins;
2063         int ret;
2064         struct extent_buffer *buf;
2065
2066         ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2067                                  root_objectid, ref_generation,
2068                                  level, first_objectid, empty_size, hint,
2069                                  (u64)-1, &ins, 0);
2070         if (ret) {
2071                 BUG_ON(ret > 0);
2072                 return ERR_PTR(ret);
2073         }
2074         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2075         if (!buf) {
2076                 btrfs_free_extent(trans, root, ins.objectid, blocksize,
2077                                   root->root_key.objectid, ref_generation,
2078                                   0, 0, 0);
2079                 return ERR_PTR(-ENOMEM);
2080         }
2081         btrfs_set_header_generation(buf, trans->transid);
2082         btrfs_tree_lock(buf);
2083         clean_tree_block(trans, root, buf);
2084         btrfs_set_buffer_uptodate(buf);
2085
2086         if (PageDirty(buf->first_page)) {
2087                 printk("page %lu dirty\n", buf->first_page->index);
2088                 WARN_ON(1);
2089         }
2090
2091         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2092                          buf->start + buf->len - 1, GFP_NOFS);
2093         if (!btrfs_test_opt(root, SSD))
2094                 btrfs_set_buffer_defrag(buf);
2095         trans->blocks_used++;
2096         return buf;
2097 }
2098
2099 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2100                                   struct btrfs_root *root,
2101                                   struct extent_buffer *leaf)
2102 {
2103         u64 leaf_owner;
2104         u64 leaf_generation;
2105         struct btrfs_key key;
2106         struct btrfs_file_extent_item *fi;
2107         int i;
2108         int nritems;
2109         int ret;
2110
2111         BUG_ON(!btrfs_is_leaf(leaf));
2112         nritems = btrfs_header_nritems(leaf);
2113         leaf_owner = btrfs_header_owner(leaf);
2114         leaf_generation = btrfs_header_generation(leaf);
2115
2116         for (i = 0; i < nritems; i++) {
2117                 u64 disk_bytenr;
2118
2119                 btrfs_item_key_to_cpu(leaf, &key, i);
2120                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2121                         continue;
2122                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2123                 if (btrfs_file_extent_type(leaf, fi) ==
2124                     BTRFS_FILE_EXTENT_INLINE)
2125                         continue;
2126                 /*
2127                  * FIXME make sure to insert a trans record that
2128                  * repeats the snapshot del on crash
2129                  */
2130                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2131                 if (disk_bytenr == 0)
2132                         continue;
2133                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2134                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2135                                 leaf_owner, leaf_generation,
2136                                 key.objectid, key.offset, 0);
2137                 BUG_ON(ret);
2138         }
2139         return 0;
2140 }
2141
2142 static void noinline reada_walk_down(struct btrfs_root *root,
2143                                      struct extent_buffer *node,
2144                                      int slot)
2145 {
2146         u64 bytenr;
2147         u64 last = 0;
2148         u32 nritems;
2149         u32 refs;
2150         u32 blocksize;
2151         int ret;
2152         int i;
2153         int level;
2154         int skipped = 0;
2155
2156         nritems = btrfs_header_nritems(node);
2157         level = btrfs_header_level(node);
2158         if (level)
2159                 return;
2160
2161         for (i = slot; i < nritems && skipped < 32; i++) {
2162                 bytenr = btrfs_node_blockptr(node, i);
2163                 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2164                              (last > bytenr && last - bytenr > 32 * 1024))) {
2165                         skipped++;
2166                         continue;
2167                 }
2168                 blocksize = btrfs_level_size(root, level - 1);
2169                 if (i != slot) {
2170                         ret = lookup_extent_ref(NULL, root, bytenr,
2171                                                 blocksize, &refs);
2172                         BUG_ON(ret);
2173                         if (refs != 1) {
2174                                 skipped++;
2175                                 continue;
2176                         }
2177                 }
2178                 mutex_unlock(&root->fs_info->alloc_mutex);
2179                 ret = readahead_tree_block(root, bytenr, blocksize,
2180                                            btrfs_node_ptr_generation(node, i));
2181                 last = bytenr + blocksize;
2182                 cond_resched();
2183                 mutex_lock(&root->fs_info->alloc_mutex);
2184                 if (ret)
2185                         break;
2186         }
2187 }
2188
2189 /*
2190  * helper function for drop_snapshot, this walks down the tree dropping ref
2191  * counts as it goes.
2192  */
2193 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2194                                    struct btrfs_root *root,
2195                                    struct btrfs_path *path, int *level)
2196 {
2197         u64 root_owner;
2198         u64 root_gen;
2199         u64 bytenr;
2200         u64 ptr_gen;
2201         struct extent_buffer *next;
2202         struct extent_buffer *cur;
2203         struct extent_buffer *parent;
2204         u32 blocksize;
2205         int ret;
2206         u32 refs;
2207
2208         mutex_lock(&root->fs_info->alloc_mutex);
2209
2210         WARN_ON(*level < 0);
2211         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2212         ret = lookup_extent_ref(trans, root,
2213                                 path->nodes[*level]->start,
2214                                 path->nodes[*level]->len, &refs);
2215         BUG_ON(ret);
2216         if (refs > 1)
2217                 goto out;
2218
2219         /*
2220          * walk down to the last node level and free all the leaves
2221          */
2222         while(*level >= 0) {
2223                 WARN_ON(*level < 0);
2224                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2225                 cur = path->nodes[*level];
2226
2227                 if (btrfs_header_level(cur) != *level)
2228                         WARN_ON(1);
2229
2230                 if (path->slots[*level] >=
2231                     btrfs_header_nritems(cur))
2232                         break;
2233                 if (*level == 0) {
2234                         ret = drop_leaf_ref(trans, root, cur);
2235                         BUG_ON(ret);
2236                         break;
2237                 }
2238                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2239                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2240                 blocksize = btrfs_level_size(root, *level - 1);
2241
2242                 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
2243                 BUG_ON(ret);
2244                 if (refs != 1) {
2245                         parent = path->nodes[*level];
2246                         root_owner = btrfs_header_owner(parent);
2247                         root_gen = btrfs_header_generation(parent);
2248                         path->slots[*level]++;
2249                         ret = __btrfs_free_extent(trans, root, bytenr,
2250                                                 blocksize, root_owner,
2251                                                 root_gen, 0, 0, 1);
2252                         BUG_ON(ret);
2253                         continue;
2254                 }
2255                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2256                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2257                         free_extent_buffer(next);
2258                         reada_walk_down(root, cur, path->slots[*level]);
2259
2260                         mutex_unlock(&root->fs_info->alloc_mutex);
2261                         next = read_tree_block(root, bytenr, blocksize,
2262                                                ptr_gen);
2263                         mutex_lock(&root->fs_info->alloc_mutex);
2264
2265                         /* we've dropped the lock, double check */
2266                         ret = lookup_extent_ref(trans, root, bytenr,
2267                                                 blocksize, &refs);
2268                         BUG_ON(ret);
2269                         if (refs != 1) {
2270                                 parent = path->nodes[*level];
2271                                 root_owner = btrfs_header_owner(parent);
2272                                 root_gen = btrfs_header_generation(parent);
2273
2274                                 path->slots[*level]++;
2275                                 free_extent_buffer(next);
2276                                 ret = __btrfs_free_extent(trans, root, bytenr,
2277                                                         blocksize,
2278                                                         root_owner,
2279                                                         root_gen, 0, 0, 1);
2280                                 BUG_ON(ret);
2281                                 continue;
2282                         }
2283                 }
2284                 WARN_ON(*level <= 0);
2285                 if (path->nodes[*level-1])
2286                         free_extent_buffer(path->nodes[*level-1]);
2287                 path->nodes[*level-1] = next;
2288                 *level = btrfs_header_level(next);
2289                 path->slots[*level] = 0;
2290         }
2291 out:
2292         WARN_ON(*level < 0);
2293         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2294
2295         if (path->nodes[*level] == root->node) {
2296                 root_owner = root->root_key.objectid;
2297                 parent = path->nodes[*level];
2298         } else {
2299                 parent = path->nodes[*level + 1];
2300                 root_owner = btrfs_header_owner(parent);
2301         }
2302
2303         root_gen = btrfs_header_generation(parent);
2304         ret = __btrfs_free_extent(trans, root, path->nodes[*level]->start,
2305                                 path->nodes[*level]->len,
2306                                 root_owner, root_gen, 0, 0, 1);
2307         free_extent_buffer(path->nodes[*level]);
2308         path->nodes[*level] = NULL;
2309         *level += 1;
2310         BUG_ON(ret);
2311         mutex_unlock(&root->fs_info->alloc_mutex);
2312         return 0;
2313 }
2314
2315 /*
2316  * helper for dropping snapshots.  This walks back up the tree in the path
2317  * to find the first node higher up where we haven't yet gone through
2318  * all the slots
2319  */
2320 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2321                                  struct btrfs_root *root,
2322                                  struct btrfs_path *path, int *level)
2323 {
2324         u64 root_owner;
2325         u64 root_gen;
2326         struct btrfs_root_item *root_item = &root->root_item;
2327         int i;
2328         int slot;
2329         int ret;
2330
2331         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2332                 slot = path->slots[i];
2333                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2334                         struct extent_buffer *node;
2335                         struct btrfs_disk_key disk_key;
2336                         node = path->nodes[i];
2337                         path->slots[i]++;
2338                         *level = i;
2339                         WARN_ON(*level == 0);
2340                         btrfs_node_key(node, &disk_key, path->slots[i]);
2341                         memcpy(&root_item->drop_progress,
2342                                &disk_key, sizeof(disk_key));
2343                         root_item->drop_level = i;
2344                         return 0;
2345                 } else {
2346                         if (path->nodes[*level] == root->node) {
2347                                 root_owner = root->root_key.objectid;
2348                                 root_gen =
2349                                    btrfs_header_generation(path->nodes[*level]);
2350                         } else {
2351                                 struct extent_buffer *node;
2352                                 node = path->nodes[*level + 1];
2353                                 root_owner = btrfs_header_owner(node);
2354                                 root_gen = btrfs_header_generation(node);
2355                         }
2356                         ret = btrfs_free_extent(trans, root,
2357                                                 path->nodes[*level]->start,
2358                                                 path->nodes[*level]->len,
2359                                                 root_owner, root_gen, 0, 0, 1);
2360                         BUG_ON(ret);
2361                         free_extent_buffer(path->nodes[*level]);
2362                         path->nodes[*level] = NULL;
2363                         *level = i + 1;
2364                 }
2365         }
2366         return 1;
2367 }
2368
2369 /*
2370  * drop the reference count on the tree rooted at 'snap'.  This traverses
2371  * the tree freeing any blocks that have a ref count of zero after being
2372  * decremented.
2373  */
2374 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2375                         *root)
2376 {
2377         int ret = 0;
2378         int wret;
2379         int level;
2380         struct btrfs_path *path;
2381         int i;
2382         int orig_level;
2383         struct btrfs_root_item *root_item = &root->root_item;
2384
2385         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2386         path = btrfs_alloc_path();
2387         BUG_ON(!path);
2388
2389         level = btrfs_header_level(root->node);
2390         orig_level = level;
2391         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2392                 path->nodes[level] = root->node;
2393                 extent_buffer_get(root->node);
2394                 path->slots[level] = 0;
2395         } else {
2396                 struct btrfs_key key;
2397                 struct btrfs_disk_key found_key;
2398                 struct extent_buffer *node;
2399
2400                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2401                 level = root_item->drop_level;
2402                 path->lowest_level = level;
2403                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2404                 if (wret < 0) {
2405                         ret = wret;
2406                         goto out;
2407                 }
2408                 node = path->nodes[level];
2409                 btrfs_node_key(node, &found_key, path->slots[level]);
2410                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2411                                sizeof(found_key)));
2412                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2413                         if (path->nodes[i] && path->locks[i]) {
2414                                 path->locks[i] = 0;
2415                                 btrfs_tree_unlock(path->nodes[i]);
2416                         }
2417                 }
2418         }
2419         while(1) {
2420                 wret = walk_down_tree(trans, root, path, &level);
2421                 if (wret > 0)
2422                         break;
2423                 if (wret < 0)
2424                         ret = wret;
2425
2426                 wret = walk_up_tree(trans, root, path, &level);
2427                 if (wret > 0)
2428                         break;
2429                 if (wret < 0)
2430                         ret = wret;
2431                 ret = -EAGAIN;
2432                 break;
2433         }
2434         for (i = 0; i <= orig_level; i++) {
2435                 if (path->nodes[i]) {
2436                         free_extent_buffer(path->nodes[i]);
2437                         path->nodes[i] = NULL;
2438                 }
2439         }
2440 out:
2441         btrfs_free_path(path);
2442         return ret;
2443 }
2444
2445 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2446 {
2447         u64 start;
2448         u64 end;
2449         u64 ptr;
2450         int ret;
2451
2452         mutex_lock(&info->alloc_mutex);
2453         while(1) {
2454                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2455                                             &start, &end, (unsigned int)-1);
2456                 if (ret)
2457                         break;
2458                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2459                 if (!ret)
2460                         kfree((void *)(unsigned long)ptr);
2461                 clear_extent_bits(&info->block_group_cache, start,
2462                                   end, (unsigned int)-1, GFP_NOFS);
2463         }
2464         while(1) {
2465                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2466                                             &start, &end, EXTENT_DIRTY);
2467                 if (ret)
2468                         break;
2469                 clear_extent_dirty(&info->free_space_cache, start,
2470                                    end, GFP_NOFS);
2471         }
2472         mutex_unlock(&info->alloc_mutex);
2473         return 0;
2474 }
2475
2476 static unsigned long calc_ra(unsigned long start, unsigned long last,
2477                              unsigned long nr)
2478 {
2479         return min(last, start + nr - 1);
2480 }
2481
2482 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2483                                          u64 len)
2484 {
2485         u64 page_start;
2486         u64 page_end;
2487         unsigned long last_index;
2488         unsigned long i;
2489         struct page *page;
2490         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2491         struct file_ra_state *ra;
2492         unsigned long total_read = 0;
2493         unsigned long ra_pages;
2494         struct btrfs_trans_handle *trans;
2495
2496         ra = kzalloc(sizeof(*ra), GFP_NOFS);
2497
2498         mutex_lock(&inode->i_mutex);
2499         i = start >> PAGE_CACHE_SHIFT;
2500         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2501
2502         ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2503
2504         file_ra_state_init(ra, inode->i_mapping);
2505
2506         for (; i <= last_index; i++) {
2507                 if (total_read % ra_pages == 0) {
2508                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2509                                        calc_ra(i, last_index, ra_pages));
2510                 }
2511                 total_read++;
2512                 if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
2513                         goto truncate_racing;
2514
2515                 page = grab_cache_page(inode->i_mapping, i);
2516                 if (!page) {
2517                         goto out_unlock;
2518                 }
2519                 if (!PageUptodate(page)) {
2520                         btrfs_readpage(NULL, page);
2521                         lock_page(page);
2522                         if (!PageUptodate(page)) {
2523                                 unlock_page(page);
2524                                 page_cache_release(page);
2525                                 goto out_unlock;
2526                         }
2527                 }
2528 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2529                 ClearPageDirty(page);
2530 #else
2531                 cancel_dirty_page(page, PAGE_CACHE_SIZE);
2532 #endif
2533                 wait_on_page_writeback(page);
2534                 set_page_extent_mapped(page);
2535                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2536                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2537
2538                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2539
2540                 set_extent_delalloc(io_tree, page_start,
2541                                     page_end, GFP_NOFS);
2542                 set_page_dirty(page);
2543
2544                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2545                 unlock_page(page);
2546                 page_cache_release(page);
2547         }
2548         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2549                                            total_read);
2550
2551 out_unlock:
2552         kfree(ra);
2553         trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2554         if (trans) {
2555                 btrfs_add_ordered_inode(inode);
2556                 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2557                 mark_inode_dirty(inode);
2558         }
2559         mutex_unlock(&inode->i_mutex);
2560         return 0;
2561
2562 truncate_racing:
2563         vmtruncate(inode, inode->i_size);
2564         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2565                                            total_read);
2566         goto out_unlock;
2567 }
2568
2569 /*
2570  * The back references tell us which tree holds a ref on a block,
2571  * but it is possible for the tree root field in the reference to
2572  * reflect the original root before a snapshot was made.  In this
2573  * case we should search through all the children of a given root
2574  * to find potential holders of references on a block.
2575  *
2576  * Instead, we do something a little less fancy and just search
2577  * all the roots for a given key/block combination.
2578  */
2579 static int find_root_for_ref(struct btrfs_root *root,
2580                              struct btrfs_path *path,
2581                              struct btrfs_key *key0,
2582                              int level,
2583                              int file_key,
2584                              struct btrfs_root **found_root,
2585                              u64 bytenr)
2586 {
2587         struct btrfs_key root_location;
2588         struct btrfs_root *cur_root = *found_root;
2589         struct btrfs_file_extent_item *file_extent;
2590         u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2591         u64 found_bytenr;
2592         int ret;
2593         int i;
2594
2595         root_location.offset = (u64)-1;
2596         root_location.type = BTRFS_ROOT_ITEM_KEY;
2597         path->lowest_level = level;
2598         path->reada = 0;
2599         while(1) {
2600                 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2601                 found_bytenr = 0;
2602                 if (ret == 0 && file_key) {
2603                         struct extent_buffer *leaf = path->nodes[0];
2604                         file_extent = btrfs_item_ptr(leaf, path->slots[0],
2605                                              struct btrfs_file_extent_item);
2606                         if (btrfs_file_extent_type(leaf, file_extent) ==
2607                             BTRFS_FILE_EXTENT_REG) {
2608                                 found_bytenr =
2609                                         btrfs_file_extent_disk_bytenr(leaf,
2610                                                                file_extent);
2611                        }
2612                 } else if (!file_key) {
2613                         if (path->nodes[level])
2614                                 found_bytenr = path->nodes[level]->start;
2615                 }
2616
2617                 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2618                         if (!path->nodes[i])
2619                                 break;
2620                         free_extent_buffer(path->nodes[i]);
2621                         path->nodes[i] = NULL;
2622                 }
2623                 btrfs_release_path(cur_root, path);
2624
2625                 if (found_bytenr == bytenr) {
2626                         *found_root = cur_root;
2627                         ret = 0;
2628                         goto out;
2629                 }
2630                 ret = btrfs_search_root(root->fs_info->tree_root,
2631                                         root_search_start, &root_search_start);
2632                 if (ret)
2633                         break;
2634
2635                 root_location.objectid = root_search_start;
2636                 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2637                                                       &root_location);
2638                 if (!cur_root) {
2639                         ret = 1;
2640                         break;
2641                 }
2642         }
2643 out:
2644         path->lowest_level = 0;
2645         return ret;
2646 }
2647
2648 /*
2649  * note, this releases the path
2650  */
2651 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2652                                   struct btrfs_path *path,
2653                                   struct btrfs_key *extent_key,
2654                                   u64 *last_file_objectid,
2655                                   u64 *last_file_offset,
2656                                   u64 *last_file_root,
2657                                   u64 last_extent)
2658 {
2659         struct inode *inode;
2660         struct btrfs_root *found_root;
2661         struct btrfs_key root_location;
2662         struct btrfs_key found_key;
2663         struct btrfs_extent_ref *ref;
2664         u64 ref_root;
2665         u64 ref_gen;
2666         u64 ref_objectid;
2667         u64 ref_offset;
2668         int ret;
2669         int level;
2670
2671         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2672                              struct btrfs_extent_ref);
2673         ref_root = btrfs_ref_root(path->nodes[0], ref);
2674         ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2675         ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2676         ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2677         btrfs_release_path(extent_root, path);
2678
2679         root_location.objectid = ref_root;
2680         if (ref_gen == 0)
2681                 root_location.offset = 0;
2682         else
2683                 root_location.offset = (u64)-1;
2684         root_location.type = BTRFS_ROOT_ITEM_KEY;
2685
2686         found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2687                                                 &root_location);
2688         BUG_ON(!found_root);
2689
2690         if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2691                 found_key.objectid = ref_objectid;
2692                 found_key.type = BTRFS_EXTENT_DATA_KEY;
2693                 found_key.offset = ref_offset;
2694                 level = 0;
2695
2696                 if (last_extent == extent_key->objectid &&
2697                     *last_file_objectid == ref_objectid &&
2698                     *last_file_offset == ref_offset &&
2699                     *last_file_root == ref_root)
2700                         goto out;
2701
2702                 ret = find_root_for_ref(extent_root, path, &found_key,
2703                                         level, 1, &found_root,
2704                                         extent_key->objectid);
2705
2706                 if (ret)
2707                         goto out;
2708
2709                 if (last_extent == extent_key->objectid &&
2710                     *last_file_objectid == ref_objectid &&
2711                     *last_file_offset == ref_offset &&
2712                     *last_file_root == ref_root)
2713                         goto out;
2714
2715                 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2716                                           ref_objectid, found_root);
2717                 if (inode->i_state & I_NEW) {
2718                         /* the inode and parent dir are two different roots */
2719                         BTRFS_I(inode)->root = found_root;
2720                         BTRFS_I(inode)->location.objectid = ref_objectid;
2721                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2722                         BTRFS_I(inode)->location.offset = 0;
2723                         btrfs_read_locked_inode(inode);
2724                         unlock_new_inode(inode);
2725
2726                 }
2727                 /* this can happen if the reference is not against
2728                  * the latest version of the tree root
2729                  */
2730                 if (is_bad_inode(inode)) {
2731                         goto out;
2732                 }
2733                 *last_file_objectid = inode->i_ino;
2734                 *last_file_root = found_root->root_key.objectid;
2735                 *last_file_offset = ref_offset;
2736
2737                 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2738                 iput(inode);
2739         } else {
2740                 struct btrfs_trans_handle *trans;
2741                 struct extent_buffer *eb;
2742                 int i;
2743
2744                 eb = read_tree_block(found_root, extent_key->objectid,
2745                                      extent_key->offset, 0);
2746                 btrfs_tree_lock(eb);
2747                 level = btrfs_header_level(eb);
2748
2749                 if (level == 0)
2750                         btrfs_item_key_to_cpu(eb, &found_key, 0);
2751                 else
2752                         btrfs_node_key_to_cpu(eb, &found_key, 0);
2753
2754                 btrfs_tree_unlock(eb);
2755                 free_extent_buffer(eb);
2756
2757                 ret = find_root_for_ref(extent_root, path, &found_key,
2758                                         level, 0, &found_root,
2759                                         extent_key->objectid);
2760
2761                 if (ret)
2762                         goto out;
2763
2764                 trans = btrfs_start_transaction(found_root, 1);
2765
2766                 path->lowest_level = level;
2767                 path->reada = 2;
2768                 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2769                                         0, 1);
2770                 path->lowest_level = 0;
2771                 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2772                         if (!path->nodes[i])
2773                                 break;
2774                         free_extent_buffer(path->nodes[i]);
2775                         path->nodes[i] = NULL;
2776                 }
2777                 btrfs_release_path(found_root, path);
2778                 if (found_root == found_root->fs_info->extent_root)
2779                         btrfs_extent_post_op(trans, found_root);
2780                 btrfs_end_transaction(trans, found_root);
2781         }
2782
2783 out:
2784         return 0;
2785 }
2786
2787 static int noinline del_extent_zero(struct btrfs_root *extent_root,
2788                                     struct btrfs_path *path,
2789                                     struct btrfs_key *extent_key)
2790 {
2791         int ret;
2792         struct btrfs_trans_handle *trans;
2793
2794         trans = btrfs_start_transaction(extent_root, 1);
2795         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
2796         if (ret > 0) {
2797                 ret = -EIO;
2798                 goto out;
2799         }
2800         if (ret < 0)
2801                 goto out;
2802         ret = btrfs_del_item(trans, extent_root, path);
2803 out:
2804         btrfs_end_transaction(trans, extent_root);
2805         return ret;
2806 }
2807
2808 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2809                                         struct btrfs_path *path,
2810                                         struct btrfs_key *extent_key)
2811 {
2812         struct btrfs_key key;
2813         struct btrfs_key found_key;
2814         struct extent_buffer *leaf;
2815         u64 last_file_objectid = 0;
2816         u64 last_file_root = 0;
2817         u64 last_file_offset = (u64)-1;
2818         u64 last_extent = 0;
2819         u32 nritems;
2820         u32 item_size;
2821         int ret = 0;
2822
2823         if (extent_key->objectid == 0) {
2824                 ret = del_extent_zero(extent_root, path, extent_key);
2825                 goto out;
2826         }
2827         key.objectid = extent_key->objectid;
2828         key.type = BTRFS_EXTENT_REF_KEY;
2829         key.offset = 0;
2830
2831         while(1) {
2832                 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2833
2834                 if (ret < 0)
2835                         goto out;
2836
2837                 ret = 0;
2838                 leaf = path->nodes[0];
2839                 nritems = btrfs_header_nritems(leaf);
2840                 if (path->slots[0] == nritems) {
2841                         ret = btrfs_next_leaf(extent_root, path);
2842                         if (ret > 0) {
2843                                 ret = 0;
2844                                 goto out;
2845                         }
2846                         if (ret < 0)
2847                                 goto out;
2848                         leaf = path->nodes[0];
2849                 }
2850
2851                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2852                 if (found_key.objectid != extent_key->objectid) {
2853                         break;
2854                 }
2855
2856                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
2857                         break;
2858                 }
2859
2860                 key.offset = found_key.offset + 1;
2861                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2862
2863                 ret = relocate_one_reference(extent_root, path, extent_key,
2864                                              &last_file_objectid,
2865                                              &last_file_offset,
2866                                              &last_file_root, last_extent);
2867                 if (ret)
2868                         goto out;
2869                 last_extent = extent_key->objectid;
2870         }
2871         ret = 0;
2872 out:
2873         btrfs_release_path(extent_root, path);
2874         return ret;
2875 }
2876
2877 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
2878 {
2879         u64 num_devices;
2880         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
2881                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
2882
2883         num_devices = root->fs_info->fs_devices->num_devices;
2884         if (num_devices == 1) {
2885                 stripped |= BTRFS_BLOCK_GROUP_DUP;
2886                 stripped = flags & ~stripped;
2887
2888                 /* turn raid0 into single device chunks */
2889                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
2890                         return stripped;
2891
2892                 /* turn mirroring into duplication */
2893                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2894                              BTRFS_BLOCK_GROUP_RAID10))
2895                         return stripped | BTRFS_BLOCK_GROUP_DUP;
2896                 return flags;
2897         } else {
2898                 /* they already had raid on here, just return */
2899                 if (flags & stripped)
2900                         return flags;
2901
2902                 stripped |= BTRFS_BLOCK_GROUP_DUP;
2903                 stripped = flags & ~stripped;
2904
2905                 /* switch duplicated blocks with raid1 */
2906                 if (flags & BTRFS_BLOCK_GROUP_DUP)
2907                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
2908
2909                 /* turn single device chunks into raid0 */
2910                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
2911         }
2912         return flags;
2913 }
2914
2915 int __alloc_chunk_for_shrink(struct btrfs_root *root,
2916                      struct btrfs_block_group_cache *shrink_block_group,
2917                      int force)
2918 {
2919         struct btrfs_trans_handle *trans;
2920         u64 new_alloc_flags;
2921         u64 calc;
2922
2923         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
2924
2925                 trans = btrfs_start_transaction(root, 1);
2926                 new_alloc_flags = update_block_group_flags(root,
2927                                                    shrink_block_group->flags);
2928                 if (new_alloc_flags != shrink_block_group->flags) {
2929                         calc =
2930                              btrfs_block_group_used(&shrink_block_group->item);
2931                 } else {
2932                         calc = shrink_block_group->key.offset;
2933                 }
2934                 do_chunk_alloc(trans, root->fs_info->extent_root,
2935                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
2936                 btrfs_end_transaction(trans, root);
2937         }
2938         return 0;
2939 }
2940
2941 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
2942 {
2943         struct btrfs_trans_handle *trans;
2944         struct btrfs_root *tree_root = root->fs_info->tree_root;
2945         struct btrfs_path *path;
2946         u64 cur_byte;
2947         u64 total_found;
2948         u64 shrink_last_byte;
2949         struct btrfs_block_group_cache *shrink_block_group;
2950         struct btrfs_fs_info *info = root->fs_info;
2951         struct btrfs_key key;
2952         struct btrfs_key found_key;
2953         struct extent_buffer *leaf;
2954         u32 nritems;
2955         int ret;
2956         int progress;
2957
2958         mutex_lock(&root->fs_info->alloc_mutex);
2959         shrink_block_group = btrfs_lookup_block_group(root->fs_info,
2960                                                       shrink_start);
2961         BUG_ON(!shrink_block_group);
2962
2963         shrink_last_byte = shrink_block_group->key.objectid +
2964                 shrink_block_group->key.offset;
2965
2966         shrink_block_group->space_info->total_bytes -=
2967                 shrink_block_group->key.offset;
2968         path = btrfs_alloc_path();
2969         root = root->fs_info->extent_root;
2970         path->reada = 2;
2971
2972         printk("btrfs relocating block group %llu flags %llu\n",
2973                (unsigned long long)shrink_start,
2974                (unsigned long long)shrink_block_group->flags);
2975
2976         __alloc_chunk_for_shrink(root, shrink_block_group, 1);
2977
2978 again:
2979
2980         shrink_block_group->ro = 1;
2981
2982         total_found = 0;
2983         progress = 0;
2984         key.objectid = shrink_start;
2985         key.offset = 0;
2986         key.type = 0;
2987         cur_byte = key.objectid;
2988
2989         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2990         if (ret < 0)
2991                 goto out;
2992
2993         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
2994         if (ret < 0)
2995                 goto out;
2996
2997         if (ret == 0) {
2998                 leaf = path->nodes[0];
2999                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3000                 if (found_key.objectid + found_key.offset > shrink_start &&
3001                     found_key.objectid < shrink_last_byte) {
3002                         cur_byte = found_key.objectid;
3003                         key.objectid = cur_byte;
3004                 }
3005         }
3006         btrfs_release_path(root, path);
3007
3008         while(1) {
3009                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3010                 if (ret < 0)
3011                         goto out;
3012
3013                 leaf = path->nodes[0];
3014                 nritems = btrfs_header_nritems(leaf);
3015 next:
3016                 if (path->slots[0] >= nritems) {
3017                         ret = btrfs_next_leaf(root, path);
3018                         if (ret < 0)
3019                                 goto out;
3020                         if (ret == 1) {
3021                                 ret = 0;
3022                                 break;
3023                         }
3024                         leaf = path->nodes[0];
3025                         nritems = btrfs_header_nritems(leaf);
3026                 }
3027
3028                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3029
3030                 if (found_key.objectid >= shrink_last_byte)
3031                         break;
3032
3033                 if (progress && need_resched()) {
3034                         memcpy(&key, &found_key, sizeof(key));
3035                         cond_resched();
3036                         btrfs_release_path(root, path);
3037                         btrfs_search_slot(NULL, root, &key, path, 0, 0);
3038                         progress = 0;
3039                         goto next;
3040                 }
3041                 progress = 1;
3042
3043                 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3044                     found_key.objectid + found_key.offset <= cur_byte) {
3045                         memcpy(&key, &found_key, sizeof(key));
3046                         key.offset++;
3047                         path->slots[0]++;
3048                         goto next;
3049                 }
3050
3051                 total_found++;
3052                 cur_byte = found_key.objectid + found_key.offset;
3053                 key.objectid = cur_byte;
3054                 btrfs_release_path(root, path);
3055                 ret = relocate_one_extent(root, path, &found_key);
3056                 __alloc_chunk_for_shrink(root, shrink_block_group, 0);
3057         }
3058
3059         btrfs_release_path(root, path);
3060
3061         if (total_found > 0) {
3062                 printk("btrfs relocate found %llu last extent was %llu\n",
3063                        (unsigned long long)total_found,
3064                        (unsigned long long)found_key.objectid);
3065                 trans = btrfs_start_transaction(tree_root, 1);
3066                 btrfs_commit_transaction(trans, tree_root);
3067
3068                 btrfs_clean_old_snapshots(tree_root);
3069
3070                 trans = btrfs_start_transaction(tree_root, 1);
3071                 btrfs_commit_transaction(trans, tree_root);
3072                 goto again;
3073         }
3074
3075         /*
3076          * we've freed all the extents, now remove the block
3077          * group item from the tree
3078          */
3079         trans = btrfs_start_transaction(root, 1);
3080         memcpy(&key, &shrink_block_group->key, sizeof(key));
3081
3082         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3083         if (ret > 0)
3084                 ret = -EIO;
3085         if (ret < 0)
3086                 goto out;
3087
3088         clear_extent_bits(&info->block_group_cache, key.objectid,
3089                           key.objectid + key.offset - 1,
3090                           (unsigned int)-1, GFP_NOFS);
3091
3092
3093         clear_extent_bits(&info->free_space_cache,
3094                            key.objectid, key.objectid + key.offset - 1,
3095                            (unsigned int)-1, GFP_NOFS);
3096
3097         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3098         kfree(shrink_block_group);
3099
3100         btrfs_del_item(trans, root, path);
3101         btrfs_commit_transaction(trans, root);
3102
3103         /* the code to unpin extents might set a few bits in the free
3104          * space cache for this range again
3105          */
3106         clear_extent_bits(&info->free_space_cache,
3107                            key.objectid, key.objectid + key.offset - 1,
3108                            (unsigned int)-1, GFP_NOFS);
3109 out:
3110         btrfs_free_path(path);
3111         mutex_unlock(&root->fs_info->alloc_mutex);
3112         return ret;
3113 }
3114
3115 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3116                            struct btrfs_key *key)
3117 {
3118         int ret = 0;
3119         struct btrfs_key found_key;
3120         struct extent_buffer *leaf;
3121         int slot;
3122
3123         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3124         if (ret < 0)
3125                 goto out;
3126
3127         while(1) {
3128                 slot = path->slots[0];
3129                 leaf = path->nodes[0];
3130                 if (slot >= btrfs_header_nritems(leaf)) {
3131                         ret = btrfs_next_leaf(root, path);
3132                         if (ret == 0)
3133                                 continue;
3134                         if (ret < 0)
3135                                 goto out;
3136                         break;
3137                 }
3138                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3139
3140                 if (found_key.objectid >= key->objectid &&
3141                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3142                         ret = 0;
3143                         goto out;
3144                 }
3145                 path->slots[0]++;
3146         }
3147         ret = -ENOENT;
3148 out:
3149         return ret;
3150 }
3151
3152 int btrfs_read_block_groups(struct btrfs_root *root)
3153 {
3154         struct btrfs_path *path;
3155         int ret;
3156         int bit;
3157         struct btrfs_block_group_cache *cache;
3158         struct btrfs_fs_info *info = root->fs_info;
3159         struct btrfs_space_info *space_info;
3160         struct extent_io_tree *block_group_cache;
3161         struct btrfs_key key;
3162         struct btrfs_key found_key;
3163         struct extent_buffer *leaf;
3164
3165         block_group_cache = &info->block_group_cache;
3166         root = info->extent_root;
3167         key.objectid = 0;
3168         key.offset = 0;
3169         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3170         path = btrfs_alloc_path();
3171         if (!path)
3172                 return -ENOMEM;
3173
3174         mutex_lock(&root->fs_info->alloc_mutex);
3175         while(1) {
3176                 ret = find_first_block_group(root, path, &key);
3177                 if (ret > 0) {
3178                         ret = 0;
3179                         goto error;
3180                 }
3181                 if (ret != 0)
3182                         goto error;
3183
3184                 leaf = path->nodes[0];
3185                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3186                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3187                 if (!cache) {
3188                         ret = -ENOMEM;
3189                         break;
3190                 }
3191
3192                 read_extent_buffer(leaf, &cache->item,
3193                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3194                                    sizeof(cache->item));
3195                 memcpy(&cache->key, &found_key, sizeof(found_key));
3196
3197                 key.objectid = found_key.objectid + found_key.offset;
3198                 btrfs_release_path(root, path);
3199                 cache->flags = btrfs_block_group_flags(&cache->item);
3200                 bit = 0;
3201                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3202                         bit = BLOCK_GROUP_DATA;
3203                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3204                         bit = BLOCK_GROUP_SYSTEM;
3205                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3206                         bit = BLOCK_GROUP_METADATA;
3207                 }
3208                 set_avail_alloc_bits(info, cache->flags);
3209
3210                 ret = update_space_info(info, cache->flags, found_key.offset,
3211                                         btrfs_block_group_used(&cache->item),
3212                                         &space_info);
3213                 BUG_ON(ret);
3214                 cache->space_info = space_info;
3215
3216                 /* use EXTENT_LOCKED to prevent merging */
3217                 set_extent_bits(block_group_cache, found_key.objectid,
3218                                 found_key.objectid + found_key.offset - 1,
3219                                 bit | EXTENT_LOCKED, GFP_NOFS);
3220                 set_state_private(block_group_cache, found_key.objectid,
3221                                   (unsigned long)cache);
3222
3223                 if (key.objectid >=
3224                     btrfs_super_total_bytes(&info->super_copy))
3225                         break;
3226         }
3227         ret = 0;
3228 error:
3229         btrfs_free_path(path);
3230         mutex_unlock(&root->fs_info->alloc_mutex);
3231         return ret;
3232 }
3233
3234 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3235                            struct btrfs_root *root, u64 bytes_used,
3236                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3237                            u64 size)
3238 {
3239         int ret;
3240         int bit = 0;
3241         struct btrfs_root *extent_root;
3242         struct btrfs_block_group_cache *cache;
3243         struct extent_io_tree *block_group_cache;
3244
3245         extent_root = root->fs_info->extent_root;
3246         block_group_cache = &root->fs_info->block_group_cache;
3247
3248         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3249         BUG_ON(!cache);
3250         cache->key.objectid = chunk_offset;
3251         cache->key.offset = size;
3252         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3253
3254         btrfs_set_block_group_used(&cache->item, bytes_used);
3255         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3256         cache->flags = type;
3257         btrfs_set_block_group_flags(&cache->item, type);
3258
3259         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3260                                 &cache->space_info);
3261         BUG_ON(ret);
3262
3263         bit = block_group_state_bits(type);
3264         set_extent_bits(block_group_cache, chunk_offset,
3265                         chunk_offset + size - 1,
3266                         bit | EXTENT_LOCKED, GFP_NOFS);
3267
3268         set_state_private(block_group_cache, chunk_offset,
3269                           (unsigned long)cache);
3270         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3271                                 sizeof(cache->item));
3272         BUG_ON(ret);
3273
3274         finish_current_insert(trans, extent_root);
3275         ret = del_pending_extents(trans, extent_root);
3276         BUG_ON(ret);
3277         set_avail_alloc_bits(extent_root->fs_info, type);
3278
3279         return 0;
3280 }