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 #include "ref-cache.h"
  30
  31 #define BLOCK_GROUP_DATA     EXTENT_WRITEBACK
  32 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
  33 #define BLOCK_GROUP_SYSTEM   EXTENT_NEW
  34
  35 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
  36
  37 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  38                                  btrfs_root *extent_root);
  39 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  40                                btrfs_root *extent_root);
  41 static struct btrfs_block_group_cache *
  42 __btrfs_find_block_group(struct btrfs_root *root,
  43                          struct btrfs_block_group_cache *hint,
  44                          u64 search_start, int data, int owner);
  45
  46 void maybe_lock_mutex(struct btrfs_root *root)
  47 {
  48         if (root != root->fs_info->extent_root &&
  49             root != root->fs_info->chunk_root &&
  50             root != root->fs_info->dev_root) {
  51                 mutex_lock(&root->fs_info->alloc_mutex);
  52         }
  53 }
  54
  55 void maybe_unlock_mutex(struct btrfs_root *root)
  56 {
  57         if (root != root->fs_info->extent_root &&
  58             root != root->fs_info->chunk_root &&
  59             root != root->fs_info->dev_root) {
  60                 mutex_unlock(&root->fs_info->alloc_mutex);
  61         }
  62 }
  63
  64 static int cache_block_group(struct btrfs_root *root,
  65                              struct btrfs_block_group_cache *block_group)
  66 {
  67         struct btrfs_path *path;
  68         int ret;
  69         struct btrfs_key key;
  70         struct extent_buffer *leaf;
  71         struct extent_io_tree *free_space_cache;
  72         int slot;
  73         u64 last = 0;
  74         u64 hole_size;
  75         u64 first_free;
  76         int found = 0;
  77
  78         if (!block_group)
  79                 return 0;
  80
  81         root = root->fs_info->extent_root;
  82         free_space_cache = &root->fs_info->free_space_cache;
  83
  84         if (block_group->cached)
  85                 return 0;
  86
  87         path = btrfs_alloc_path();
  88         if (!path)
  89                 return -ENOMEM;
  90
  91         path->reada = 2;
  92         /*
  93          * we get into deadlocks with paths held by callers of this function.
  94          * since the alloc_mutex is protecting things right now, just
  95          * skip the locking here
  96          */
  97         path->skip_locking = 1;
  98         first_free = block_group->key.objectid;
  99         key.objectid = block_group->key.objectid;
 100         key.offset = 0;
 101         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 102         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 103         if (ret < 0)
 104                 return ret;
 105         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
 106         if (ret < 0)
 107                 return ret;
 108         if (ret == 0) {
 109                 leaf = path->nodes[0];
 110                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 111                 if (key.objectid + key.offset > first_free)
 112                         first_free = key.objectid + key.offset;
 113         }
 114         while(1) {
 115                 leaf = path->nodes[0];
 116                 slot = path->slots[0];
 117                 if (slot >= btrfs_header_nritems(leaf)) {
 118                         ret = btrfs_next_leaf(root, path);
 119                         if (ret < 0)
 120                                 goto err;
 121                         if (ret == 0) {
 122                                 continue;
 123                         } else {
 124                                 break;
 125                         }
 126                 }
 127                 btrfs_item_key_to_cpu(leaf, &key, slot);
 128                 if (key.objectid < block_group->key.objectid) {
 129                         goto next;
 130                 }
 131                 if (key.objectid >= block_group->key.objectid +
 132                     block_group->key.offset) {
 133                         break;
 134                 }
 135
 136                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 137                         if (!found) {
 138                                 last = first_free;
 139                                 found = 1;
 140                         }
 141                         if (key.objectid > last) {
 142                                 hole_size = key.objectid - last;
 143                                 set_extent_dirty(free_space_cache, last,
 144                                                  last + hole_size - 1,
 145                                                  GFP_NOFS);
 146                         }
 147                         last = key.objectid + key.offset;
 148                 }
 149 next:
 150                 path->slots[0]++;
 151         }
 152
 153         if (!found)
 154                 last = first_free;
 155         if (block_group->key.objectid +
 156             block_group->key.offset > last) {
 157                 hole_size = block_group->key.objectid +
 158                         block_group->key.offset - last;
 159                 set_extent_dirty(free_space_cache, last,
 160                                  last + hole_size - 1, GFP_NOFS);
 161         }
 162         block_group->cached = 1;
 163 err:
 164         btrfs_free_path(path);
 165         return 0;
 166 }
 167
 168 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 169                                                        btrfs_fs_info *info,
 170                                                          u64 bytenr)
 171 {
 172         struct extent_io_tree *block_group_cache;
 173         struct btrfs_block_group_cache *block_group = NULL;
 174         u64 ptr;
 175         u64 start;
 176         u64 end;
 177         int ret;
 178
 179         bytenr = max_t(u64, bytenr,
 180                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 181         block_group_cache = &info->block_group_cache;
 182         ret = find_first_extent_bit(block_group_cache,
 183                                     bytenr, &start, &end,
 184                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 185                                     BLOCK_GROUP_SYSTEM);
 186         if (ret) {
 187                 return NULL;
 188         }
 189         ret = get_state_private(block_group_cache, start, &ptr);
 190         if (ret)
 191                 return NULL;
 192
 193         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 194         return block_group;
 195 }
 196
 197 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 198                                                          btrfs_fs_info *info,
 199                                                          u64 bytenr)
 200 {
 201         struct extent_io_tree *block_group_cache;
 202         struct btrfs_block_group_cache *block_group = NULL;
 203         u64 ptr;
 204         u64 start;
 205         u64 end;
 206         int ret;
 207
 208         bytenr = max_t(u64, bytenr,
 209                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 210         block_group_cache = &info->block_group_cache;
 211         ret = find_first_extent_bit(block_group_cache,
 212                                     bytenr, &start, &end,
 213                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 214                                     BLOCK_GROUP_SYSTEM);
 215         if (ret) {
 216                 return NULL;
 217         }
 218         ret = get_state_private(block_group_cache, start, &ptr);
 219         if (ret)
 220                 return NULL;
 221
 222         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 223         if (block_group->key.objectid <= bytenr && bytenr <
 224             block_group->key.objectid + block_group->key.offset)
 225                 return block_group;
 226         return NULL;
 227 }
 228
 229 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 230 {
 231         return (cache->flags & bits) == bits;
 232 }
 233
 234 static int noinline find_search_start(struct btrfs_root *root,
 235                               struct btrfs_block_group_cache **cache_ret,
 236                               u64 *start_ret, u64 num, int data)
 237 {
 238         int ret;
 239         struct btrfs_block_group_cache *cache = *cache_ret;
 240         struct extent_io_tree *free_space_cache;
 241         struct extent_state *state;
 242         u64 last;
 243         u64 start = 0;
 244         u64 cache_miss = 0;
 245         u64 total_fs_bytes;
 246         u64 search_start = *start_ret;
 247         int wrapped = 0;
 248
 249         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
 250         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 251         free_space_cache = &root->fs_info->free_space_cache;
 252
 253         if (!cache)
 254                 goto out;
 255
 256 again:
 257         ret = cache_block_group(root, cache);
 258         if (ret) {
 259                 goto out;
 260         }
 261
 262         last = max(search_start, cache->key.objectid);
 263         if (!block_group_bits(cache, data) || cache->ro)
 264                 goto new_group;
 265
 266         spin_lock_irq(&free_space_cache->lock);
 267         state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
 268         while(1) {
 269                 if (!state) {
 270                         if (!cache_miss)
 271                                 cache_miss = last;
 272                         spin_unlock_irq(&free_space_cache->lock);
 273                         goto new_group;
 274                 }
 275
 276                 start = max(last, state->start);
 277                 last = state->end + 1;
 278                 if (last - start < num) {
 279                         do {
 280                                 state = extent_state_next(state);
 281                         } while(state && !(state->state & EXTENT_DIRTY));
 282                         continue;
 283                 }
 284                 spin_unlock_irq(&free_space_cache->lock);
 285                 if (cache->ro) {
 286                         goto new_group;
 287                 }
 288                 if (start + num > cache->key.objectid + cache->key.offset)
 289                         goto new_group;
 290                 if (!block_group_bits(cache, data)) {
 291                         printk("block group bits don't match %Lu %d\n", cache->flags, data);
 292                 }
 293                 *start_ret = start;
 294                 return 0;
 295         }
 296 out:
 297         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 298         if (!cache) {
 299                 printk("Unable to find block group for %Lu\n", search_start);
 300                 WARN_ON(1);
 301         }
 302         return -ENOSPC;
 303
 304 new_group:
 305         last = cache->key.objectid + cache->key.offset;
 306 wrapped:
 307         cache = btrfs_lookup_first_block_group(root->fs_info, last);
 308         if (!cache || cache->key.objectid >= total_fs_bytes) {
 309 no_cache:
 310                 if (!wrapped) {
 311                         wrapped = 1;
 312                         last = search_start;
 313                         goto wrapped;
 314                 }
 315                 goto out;
 316         }
 317         if (cache_miss && !cache->cached) {
 318                 cache_block_group(root, cache);
 319                 last = cache_miss;
 320                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 321         }
 322         cache_miss = 0;
 323         cache = btrfs_find_block_group(root, cache, last, data, 0);
 324         if (!cache)
 325                 goto no_cache;
 326         *cache_ret = cache;
 327         goto again;
 328 }
 329
 330 static u64 div_factor(u64 num, int factor)
 331 {
 332         if (factor == 10)
 333                 return num;
 334         num *= factor;
 335         do_div(num, 10);
 336         return num;
 337 }
 338
 339 static int block_group_state_bits(u64 flags)
 340 {
 341         int bits = 0;
 342         if (flags & BTRFS_BLOCK_GROUP_DATA)
 343                 bits |= BLOCK_GROUP_DATA;
 344         if (flags & BTRFS_BLOCK_GROUP_METADATA)
 345                 bits |= BLOCK_GROUP_METADATA;
 346         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
 347                 bits |= BLOCK_GROUP_SYSTEM;
 348         return bits;
 349 }
 350
 351 static struct btrfs_block_group_cache *
 352 __btrfs_find_block_group(struct btrfs_root *root,
 353                          struct btrfs_block_group_cache *hint,
 354                          u64 search_start, int data, int owner)
 355 {
 356         struct btrfs_block_group_cache *cache;
 357         struct extent_io_tree *block_group_cache;
 358         struct btrfs_block_group_cache *found_group = NULL;
 359         struct btrfs_fs_info *info = root->fs_info;
 360         u64 used;
 361         u64 last = 0;
 362         u64 start;
 363         u64 end;
 364         u64 free_check;
 365         u64 ptr;
 366         int bit;
 367         int ret;
 368         int full_search = 0;
 369         int factor = 10;
 370         int wrapped = 0;
 371
 372         block_group_cache = &info->block_group_cache;
 373
 374         if (data & BTRFS_BLOCK_GROUP_METADATA)
 375                 factor = 9;
 376
 377         bit = block_group_state_bits(data);
 378
 379         if (search_start) {
 380                 struct btrfs_block_group_cache *shint;
 381                 shint = btrfs_lookup_first_block_group(info, search_start);
 382                 if (shint && block_group_bits(shint, data) && !shint->ro) {
 383                         spin_lock(&shint->lock);
 384                         used = btrfs_block_group_used(&shint->item);
 385                         if (used + shint->pinned <
 386                             div_factor(shint->key.offset, factor)) {
 387                                 spin_unlock(&shint->lock);
 388                                 return shint;
 389                         }
 390                         spin_unlock(&shint->lock);
 391                 }
 392         }
 393         if (hint && !hint->ro && block_group_bits(hint, data)) {
 394                 spin_lock(&hint->lock);
 395                 used = btrfs_block_group_used(&hint->item);
 396                 if (used + hint->pinned <
 397                     div_factor(hint->key.offset, factor)) {
 398                         spin_unlock(&hint->lock);
 399                         return hint;
 400                 }
 401                 spin_unlock(&hint->lock);
 402                 last = hint->key.objectid + hint->key.offset;
 403         } else {
 404                 if (hint)
 405                         last = max(hint->key.objectid, search_start);
 406                 else
 407                         last = search_start;
 408         }
 409 again:
 410         while(1) {
 411                 ret = find_first_extent_bit(block_group_cache, last,
 412                                             &start, &end, bit);
 413                 if (ret)
 414                         break;
 415
 416                 ret = get_state_private(block_group_cache, start, &ptr);
 417                 if (ret) {
 418                         last = end + 1;
 419                         continue;
 420                 }
 421
 422                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 423                 spin_lock(&cache->lock);
 424                 last = cache->key.objectid + cache->key.offset;
 425                 used = btrfs_block_group_used(&cache->item);
 426
 427                 if (!cache->ro && block_group_bits(cache, data)) {
 428                         free_check = div_factor(cache->key.offset, factor);
 429                         if (used + cache->pinned < free_check) {
 430                                 found_group = cache;
 431                                 spin_unlock(&cache->lock);
 432                                 goto found;
 433                         }
 434                 }
 435                 spin_unlock(&cache->lock);
 436                 cond_resched();
 437         }
 438         if (!wrapped) {
 439                 last = search_start;
 440                 wrapped = 1;
 441                 goto again;
 442         }
 443         if (!full_search && factor < 10) {
 444                 last = search_start;
 445                 full_search = 1;
 446                 factor = 10;
 447                 goto again;
 448         }
 449 found:
 450         return found_group;
 451 }
 452
 453 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 454                                                  struct btrfs_block_group_cache
 455                                                  *hint, u64 search_start,
 456                                                  int data, int owner)
 457 {
 458
 459         struct btrfs_block_group_cache *ret;
 460         ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
 461         return ret;
 462 }
 463 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
 464                            u64 owner, u64 owner_offset)
 465 {
 466         u32 high_crc = ~(u32)0;
 467         u32 low_crc = ~(u32)0;
 468         __le64 lenum;
 469         lenum = cpu_to_le64(root_objectid);
 470         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
 471         lenum = cpu_to_le64(ref_generation);
 472         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 473         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
 474                 lenum = cpu_to_le64(owner);
 475                 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 476                 lenum = cpu_to_le64(owner_offset);
 477                 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 478         }
 479         return ((u64)high_crc << 32) | (u64)low_crc;
 480 }
 481
 482 static int match_extent_ref(struct extent_buffer *leaf,
 483                             struct btrfs_extent_ref *disk_ref,
 484                             struct btrfs_extent_ref *cpu_ref)
 485 {
 486         int ret;
 487         int len;
 488
 489         if (cpu_ref->objectid)
 490                 len = sizeof(*cpu_ref);
 491         else
 492                 len = 2 * sizeof(u64);
 493         ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
 494                                    len);
 495         return ret == 0;
 496 }
 497
 498 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 499                                           struct btrfs_root *root,
 500                                           struct btrfs_path *path, u64 bytenr,
 501                                           u64 root_objectid,
 502                                           u64 ref_generation, u64 owner,
 503                                           u64 owner_offset, int del)
 504 {
 505         u64 hash;
 506         struct btrfs_key key;
 507         struct btrfs_key found_key;
 508         struct btrfs_extent_ref ref;
 509         struct extent_buffer *leaf;
 510         struct btrfs_extent_ref *disk_ref;
 511         int ret;
 512         int ret2;
 513
 514         btrfs_set_stack_ref_root(&ref, root_objectid);
 515         btrfs_set_stack_ref_generation(&ref, ref_generation);
 516         btrfs_set_stack_ref_objectid(&ref, owner);
 517         btrfs_set_stack_ref_offset(&ref, owner_offset);
 518
 519         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 520                                owner_offset);
 521         key.offset = hash;
 522         key.objectid = bytenr;
 523         key.type = BTRFS_EXTENT_REF_KEY;
 524
 525         while (1) {
 526                 ret = btrfs_search_slot(trans, root, &key, path,
 527                                         del ? -1 : 0, del);
 528                 if (ret < 0)
 529                         goto out;
 530                 leaf = path->nodes[0];
 531                 if (ret != 0) {
 532                         u32 nritems = btrfs_header_nritems(leaf);
 533                         if (path->slots[0] >= nritems) {
 534                                 ret2 = btrfs_next_leaf(root, path);
 535                                 if (ret2)
 536                                         goto out;
 537                                 leaf = path->nodes[0];
 538                         }
 539                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 540                         if (found_key.objectid != bytenr ||
 541                             found_key.type != BTRFS_EXTENT_REF_KEY)
 542                                 goto out;
 543                         key.offset = found_key.offset;
 544                         if (del) {
 545                                 btrfs_release_path(root, path);
 546                                 continue;
 547                         }
 548                 }
 549                 disk_ref = btrfs_item_ptr(path->nodes[0],
 550                                           path->slots[0],
 551                                           struct btrfs_extent_ref);
 552                 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
 553                         ret = 0;
 554                         goto out;
 555                 }
 556                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 557                 key.offset = found_key.offset + 1;
 558                 btrfs_release_path(root, path);
 559         }
 560 out:
 561         return ret;
 562 }
 563
 564 /*
 565  * Back reference rules.  Back refs have three main goals:
 566  *
 567  * 1) differentiate between all holders of references to an extent so that
 568  *    when a reference is dropped we can make sure it was a valid reference
 569  *    before freeing the extent.
 570  *
 571  * 2) Provide enough information to quickly find the holders of an extent
 572  *    if we notice a given block is corrupted or bad.
 573  *
 574  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 575  *    maintenance.  This is actually the same as #2, but with a slightly
 576  *    different use case.
 577  *
 578  * File extents can be referenced by:
 579  *
 580  * - multiple snapshots, subvolumes, or different generations in one subvol
 581  * - different files inside a single subvolume (in theory, not implemented yet)
 582  * - different offsets inside a file (bookend extents in file.c)
 583  *
 584  * The extent ref structure has fields for:
 585  *
 586  * - Objectid of the subvolume root
 587  * - Generation number of the tree holding the reference
 588  * - objectid of the file holding the reference
 589  * - offset in the file corresponding to the key holding the reference
 590  *
 591  * When a file extent is allocated the fields are filled in:
 592  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 593  *
 594  * When a leaf is cow'd new references are added for every file extent found
 595  * in the leaf.  It looks the same as the create case, but trans->transid
 596  * will be different when the block is cow'd.
 597  *
 598  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 599  *
 600  * When a file extent is removed either during snapshot deletion or file
 601  * truncation, the corresponding back reference is found
 602  * by searching for:
 603  *
 604  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 605  *      inode objectid, offset in file)
 606  *
 607  * Btree extents can be referenced by:
 608  *
 609  * - Different subvolumes
 610  * - Different generations of the same subvolume
 611  *
 612  * Storing sufficient information for a full reverse mapping of a btree
 613  * block would require storing the lowest key of the block in the backref,
 614  * and it would require updating that lowest key either before write out or
 615  * every time it changed.  Instead, the objectid of the lowest key is stored
 616  * along with the level of the tree block.  This provides a hint
 617  * about where in the btree the block can be found.  Searches through the
 618  * btree only need to look for a pointer to that block, so they stop one
 619  * level higher than the level recorded in the backref.
 620  *
 621  * Some btrees do not do reference counting on their extents.  These
 622  * include the extent tree and the tree of tree roots.  Backrefs for these
 623  * trees always have a generation of zero.
 624  *
 625  * When a tree block is created, back references are inserted:
 626  *
 627  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
 628  *
 629  * When a tree block is cow'd in a reference counted root,
 630  * new back references are added for all the blocks it points to.
 631  * These are of the form (trans->transid will have increased since creation):
 632  *
 633  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
 634  *
 635  * Because the lowest_key_objectid and the level are just hints
 636  * they are not used when backrefs are deleted.  When a backref is deleted:
 637  *
 638  * if backref was for a tree root:
 639  *     root_objectid = root->root_key.objectid
 640  * else
 641  *     root_objectid = btrfs_header_owner(parent)
 642  *
 643  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
 644  *
 645  * Back Reference Key hashing:
 646  *
 647  * Back references have four fields, each 64 bits long.  Unfortunately,
 648  * This is hashed into a single 64 bit number and placed into the key offset.
 649  * The key objectid corresponds to the first byte in the extent, and the
 650  * key type is set to BTRFS_EXTENT_REF_KEY
 651  */
 652 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
 653                                  struct btrfs_root *root,
 654                                  struct btrfs_path *path, u64 bytenr,
 655                                  u64 root_objectid, u64 ref_generation,
 656                                  u64 owner, u64 owner_offset)
 657 {
 658         u64 hash;
 659         struct btrfs_key key;
 660         struct btrfs_extent_ref ref;
 661         struct btrfs_extent_ref *disk_ref;
 662         int ret;
 663
 664         btrfs_set_stack_ref_root(&ref, root_objectid);
 665         btrfs_set_stack_ref_generation(&ref, ref_generation);
 666         btrfs_set_stack_ref_objectid(&ref, owner);
 667         btrfs_set_stack_ref_offset(&ref, owner_offset);
 668
 669         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 670                                owner_offset);
 671         key.offset = hash;
 672         key.objectid = bytenr;
 673         key.type = BTRFS_EXTENT_REF_KEY;
 674
 675         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
 676         while (ret == -EEXIST) {
 677                 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 678                                           struct btrfs_extent_ref);
 679                 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
 680                         goto out;
 681                 key.offset++;
 682                 btrfs_release_path(root, path);
 683                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 684                                               sizeof(ref));
 685         }
 686         if (ret)
 687                 goto out;
 688         disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 689                                   struct btrfs_extent_ref);
 690         write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
 691                             sizeof(ref));
 692         btrfs_mark_buffer_dirty(path->nodes[0]);
 693 out:
 694         btrfs_release_path(root, path);
 695         return ret;
 696 }
 697
 698 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 699                                 struct btrfs_root *root,
 700                                 u64 bytenr, u64 num_bytes,
 701                                 u64 root_objectid, u64 ref_generation,
 702                                 u64 owner, u64 owner_offset)
 703 {
 704         struct btrfs_path *path;
 705         int ret;
 706         struct btrfs_key key;
 707         struct extent_buffer *l;
 708         struct btrfs_extent_item *item;
 709         u32 refs;
 710
 711         WARN_ON(num_bytes < root->sectorsize);
 712         path = btrfs_alloc_path();
 713         if (!path)
 714                 return -ENOMEM;
 715
 716         path->reada = 1;
 717         key.objectid = bytenr;
 718         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 719         key.offset = num_bytes;
 720         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 721                                 0, 1);
 722         if (ret < 0)
 723                 return ret;
 724         if (ret != 0) {
 725                 BUG();
 726         }
 727         BUG_ON(ret != 0);
 728         l = path->nodes[0];
 729         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 730         refs = btrfs_extent_refs(l, item);
 731         btrfs_set_extent_refs(l, item, refs + 1);
 732         btrfs_mark_buffer_dirty(path->nodes[0]);
 733
 734         btrfs_release_path(root->fs_info->extent_root, path);
 735
 736         path->reada = 1;
 737         ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
 738                                           path, bytenr, root_objectid,
 739                                           ref_generation, owner, owner_offset);
 740         BUG_ON(ret);
 741         finish_current_insert(trans, root->fs_info->extent_root);
 742         del_pending_extents(trans, root->fs_info->extent_root);
 743
 744         btrfs_free_path(path);
 745         return 0;
 746 }
 747
 748 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 749                                 struct btrfs_root *root,
 750                                 u64 bytenr, u64 num_bytes,
 751                                 u64 root_objectid, u64 ref_generation,
 752                                 u64 owner, u64 owner_offset)
 753 {
 754         int ret;
 755
 756         mutex_lock(&root->fs_info->alloc_mutex);
 757         ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
 758                                      root_objectid, ref_generation,
 759                                      owner, owner_offset);
 760         mutex_unlock(&root->fs_info->alloc_mutex);
 761         return ret;
 762 }
 763
 764 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 765                          struct btrfs_root *root)
 766 {
 767         finish_current_insert(trans, root->fs_info->extent_root);
 768         del_pending_extents(trans, root->fs_info->extent_root);
 769         return 0;
 770 }
 771
 772 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 773                              struct btrfs_root *root, u64 bytenr,
 774                              u64 num_bytes, u32 *refs)
 775 {
 776         struct btrfs_path *path;
 777         int ret;
 778         struct btrfs_key key;
 779         struct extent_buffer *l;
 780         struct btrfs_extent_item *item;
 781
 782         WARN_ON(num_bytes < root->sectorsize);
 783         path = btrfs_alloc_path();
 784         path->reada = 1;
 785         key.objectid = bytenr;
 786         key.offset = num_bytes;
 787         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 788         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 789                                 0, 0);
 790         if (ret < 0)
 791                 goto out;
 792         if (ret != 0) {
 793                 btrfs_print_leaf(root, path->nodes[0]);
 794                 printk("failed to find block number %Lu\n", bytenr);
 795                 BUG();
 796         }
 797         l = path->nodes[0];
 798         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 799         *refs = btrfs_extent_refs(l, item);
 800 out:
 801         btrfs_free_path(path);
 802         return 0;
 803 }
 804
 805
 806 static int get_reference_status(struct btrfs_root *root, u64 bytenr,
 807                                 u64 parent_gen, u64 ref_objectid,
 808                                 u64 *min_generation, u32 *ref_count)
 809 {
 810         struct btrfs_root *extent_root = root->fs_info->extent_root;
 811         struct btrfs_path *path;
 812         struct extent_buffer *leaf;
 813         struct btrfs_extent_ref *ref_item;
 814         struct btrfs_key key;
 815         struct btrfs_key found_key;
 816         u64 root_objectid = root->root_key.objectid;
 817         u64 ref_generation;
 818         u32 nritems;
 819         int ret;
 820
 821         key.objectid = bytenr;
 822         key.offset = 0;
 823         key.type = BTRFS_EXTENT_ITEM_KEY;
 824
 825         path = btrfs_alloc_path();
 826         mutex_lock(&root->fs_info->alloc_mutex);
 827         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 828         if (ret < 0)
 829                 goto out;
 830         BUG_ON(ret == 0);
 831
 832         leaf = path->nodes[0];
 833         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 834
 835         if (found_key.objectid != bytenr ||
 836             found_key.type != BTRFS_EXTENT_ITEM_KEY) {
 837                 ret = 1;
 838                 goto out;
 839         }
 840
 841         *ref_count = 0;
 842         *min_generation = (u64)-1;
 843
 844         while (1) {
 845                 leaf = path->nodes[0];
 846                 nritems = btrfs_header_nritems(leaf);
 847                 if (path->slots[0] >= nritems) {
 848                         ret = btrfs_next_leaf(extent_root, path);
 849                         if (ret < 0)
 850                                 goto out;
 851                         if (ret == 0)
 852                                 continue;
 853                         break;
 854                 }
 855                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 856                 if (found_key.objectid != bytenr)
 857                         break;
 858
 859                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 860                         path->slots[0]++;
 861                         continue;
 862                 }
 863
 864                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
 865                                           struct btrfs_extent_ref);
 866                 ref_generation = btrfs_ref_generation(leaf, ref_item);
 867                 /*
 868                  * For (parent_gen > 0 && parent_gen > ref_gen):
 869                  *
 870                  * we reach here through the oldest root, therefore
 871                  * all other reference from same snapshot should have
 872                  * a larger generation.
 873                  */
 874                 if ((root_objectid != btrfs_ref_root(leaf, ref_item)) ||
 875                     (parent_gen > 0 && parent_gen > ref_generation) ||
 876                     (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
 877                      ref_objectid != btrfs_ref_objectid(leaf, ref_item))) {
 878                         if (ref_count)
 879                                 *ref_count = 2;
 880                         break;
 881                 }
 882
 883                 *ref_count = 1;
 884                 if (*min_generation > ref_generation)
 885                         *min_generation = ref_generation;
 886
 887                 path->slots[0]++;
 888         }
 889         ret = 0;
 890 out:
 891         mutex_unlock(&root->fs_info->alloc_mutex);
 892         btrfs_free_path(path);
 893         return ret;
 894 }
 895
 896 int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans,
 897                            struct btrfs_root *root,
 898                            struct btrfs_key *key, u64 bytenr)
 899 {
 900         struct btrfs_root *old_root;
 901         struct btrfs_path *path = NULL;
 902         struct extent_buffer *eb;
 903         struct btrfs_file_extent_item *item;
 904         u64 ref_generation;
 905         u64 min_generation;
 906         u64 extent_start;
 907         u32 ref_count;
 908         int level;
 909         int ret;
 910
 911         BUG_ON(trans == NULL);
 912         BUG_ON(key->type != BTRFS_EXTENT_DATA_KEY);
 913         ret = get_reference_status(root, bytenr, 0, key->objectid,
 914                                    &min_generation, &ref_count);
 915         if (ret)
 916                 return ret;
 917
 918         if (ref_count != 1)
 919                 return 1;
 920
 921         old_root = root->dirty_root->root;
 922         ref_generation = old_root->root_key.offset;
 923
 924         /* all references are created in running transaction */
 925         if (min_generation > ref_generation) {
 926                 ret = 0;
 927                 goto out;
 928         }
 929
 930         path = btrfs_alloc_path();
 931         if (!path) {
 932                 ret = -ENOMEM;
 933                 goto out;
 934         }
 935
 936         path->skip_locking = 1;
 937         /* if no item found, the extent is referenced by other snapshot */
 938         ret = btrfs_search_slot(NULL, old_root, key, path, 0, 0);
 939         if (ret)
 940                 goto out;
 941
 942         eb = path->nodes[0];
 943         item = btrfs_item_ptr(eb, path->slots[0],
 944                               struct btrfs_file_extent_item);
 945         if (btrfs_file_extent_type(eb, item) != BTRFS_FILE_EXTENT_REG ||
 946             btrfs_file_extent_disk_bytenr(eb, item) != bytenr) {
 947                 ret = 1;
 948                 goto out;
 949         }
 950
 951         for (level = BTRFS_MAX_LEVEL - 1; level >= -1; level--) {
 952                 if (level >= 0) {
 953                         eb = path->nodes[level];
 954                         if (!eb)
 955                                 continue;
 956                         extent_start = eb->start;
 957                 } else
 958                         extent_start = bytenr;
 959
 960                 ret = get_reference_status(root, extent_start, ref_generation,
 961                                            0, &min_generation, &ref_count);
 962                 if (ret)
 963                         goto out;
 964
 965                 if (ref_count != 1) {
 966                         ret = 1;
 967                         goto out;
 968                 }
 969                 if (level >= 0)
 970                         ref_generation = btrfs_header_generation(eb);
 971         }
 972         ret = 0;
 973 out:
 974         if (path)
 975                 btrfs_free_path(path);
 976         return ret;
 977 }
 978
 979 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 980                   struct extent_buffer *buf, int cache_ref)
 981 {
 982         u64 bytenr;
 983         u32 nritems;
 984         struct btrfs_key key;
 985         struct btrfs_file_extent_item *fi;
 986         int i;
 987         int level;
 988         int ret;
 989         int faili;
 990         int nr_file_extents = 0;
 991
 992         if (!root->ref_cows)
 993                 return 0;
 994
 995         level = btrfs_header_level(buf);
 996         nritems = btrfs_header_nritems(buf);
 997         for (i = 0; i < nritems; i++) {
 998                 cond_resched();
 999                 if (level == 0) {
1000                         u64 disk_bytenr;
1001                         btrfs_item_key_to_cpu(buf, &key, i);
1002                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1003                                 continue;
1004                         fi = btrfs_item_ptr(buf, i,
1005                                             struct btrfs_file_extent_item);
1006                         if (btrfs_file_extent_type(buf, fi) ==
1007                             BTRFS_FILE_EXTENT_INLINE)
1008                                 continue;
1009                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1010                         if (disk_bytenr == 0)
1011                                 continue;
1012
1013                         if (buf != root->commit_root)
1014                                 nr_file_extents++;
1015
1016                         mutex_lock(&root->fs_info->alloc_mutex);
1017                         ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr,
1018                                     btrfs_file_extent_disk_num_bytes(buf, fi),
1019                                     root->root_key.objectid, trans->transid,
1020                                     key.objectid, key.offset);
1021                         mutex_unlock(&root->fs_info->alloc_mutex);
1022                         if (ret) {
1023                                 faili = i;
1024                                 WARN_ON(1);
1025                                 goto fail;
1026                         }
1027                 } else {
1028                         bytenr = btrfs_node_blockptr(buf, i);
1029                         btrfs_node_key_to_cpu(buf, &key, i);
1030
1031                         mutex_lock(&root->fs_info->alloc_mutex);
1032                         ret = __btrfs_inc_extent_ref(trans, root, bytenr,
1033                                            btrfs_level_size(root, level - 1),
1034                                            root->root_key.objectid,
1035                                            trans->transid,
1036                                            level - 1, key.objectid);
1037                         mutex_unlock(&root->fs_info->alloc_mutex);
1038                         if (ret) {
1039                                 faili = i;
1040                                 WARN_ON(1);
1041                                 goto fail;
1042                         }
1043                 }
1044         }
1045         /* cache orignal leaf block's references */
1046         if (level == 0 && cache_ref && buf != root->commit_root) {
1047                 struct btrfs_leaf_ref *ref;
1048                 struct btrfs_extent_info *info;
1049
1050                 ref = btrfs_alloc_leaf_ref(root, nr_file_extents);
1051                 if (!ref) {
1052                         WARN_ON(1);
1053                         goto out;
1054                 }
1055
1056                 ref->root_gen = root->root_key.offset;
1057                 ref->bytenr = buf->start;
1058                 ref->owner = btrfs_header_owner(buf);
1059                 ref->generation = btrfs_header_generation(buf);
1060                 ref->nritems = nr_file_extents;
1061                 info = ref->extents;
1062
1063                 for (i = 0; nr_file_extents > 0 && i < nritems; i++) {
1064                         u64 disk_bytenr;
1065                         btrfs_item_key_to_cpu(buf, &key, i);
1066                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1067                                 continue;
1068                         fi = btrfs_item_ptr(buf, i,
1069                                             struct btrfs_file_extent_item);
1070                         if (btrfs_file_extent_type(buf, fi) ==
1071                             BTRFS_FILE_EXTENT_INLINE)
1072                                 continue;
1073                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1074                         if (disk_bytenr == 0)
1075                                 continue;
1076
1077                         info->bytenr = disk_bytenr;
1078                         info->num_bytes =
1079                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1080                         info->objectid = key.objectid;
1081                         info->offset = key.offset;
1082                         info++;
1083                 }
1084
1085                 BUG_ON(!root->ref_tree);
1086                 ret = btrfs_add_leaf_ref(root, ref);
1087                 WARN_ON(ret);
1088                 btrfs_free_leaf_ref(root, ref);
1089         }
1090 out:
1091         return 0;
1092 fail:
1093         WARN_ON(1);
1094 #if 0
1095         for (i =0; i < faili; i++) {
1096                 if (level == 0) {
1097                         u64 disk_bytenr;
1098                         btrfs_item_key_to_cpu(buf, &key, i);
1099                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1100                                 continue;
1101                         fi = btrfs_item_ptr(buf, i,
1102                                             struct btrfs_file_extent_item);
1103                         if (btrfs_file_extent_type(buf, fi) ==
1104                             BTRFS_FILE_EXTENT_INLINE)
1105                                 continue;
1106                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1107                         if (disk_bytenr == 0)
1108                                 continue;
1109                         err = btrfs_free_extent(trans, root, disk_bytenr,
1110                                     btrfs_file_extent_disk_num_bytes(buf,
1111                                                                       fi), 0);
1112                         BUG_ON(err);
1113                 } else {
1114                         bytenr = btrfs_node_blockptr(buf, i);
1115                         err = btrfs_free_extent(trans, root, bytenr,
1116                                         btrfs_level_size(root, level - 1), 0);
1117                         BUG_ON(err);
1118                 }
1119         }
1120 #endif
1121         return ret;
1122 }
1123
1124 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1125                                  struct btrfs_root *root,
1126                                  struct btrfs_path *path,
1127                                  struct btrfs_block_group_cache *cache)
1128 {
1129         int ret;
1130         int pending_ret;
1131         struct btrfs_root *extent_root = root->fs_info->extent_root;
1132         unsigned long bi;
1133         struct extent_buffer *leaf;
1134
1135         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1136         if (ret < 0)
1137                 goto fail;
1138         BUG_ON(ret);
1139
1140         leaf = path->nodes[0];
1141         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1142         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1143         btrfs_mark_buffer_dirty(leaf);
1144         btrfs_release_path(extent_root, path);
1145 fail:
1146         finish_current_insert(trans, extent_root);
1147         pending_ret = del_pending_extents(trans, extent_root);
1148         if (ret)
1149                 return ret;
1150         if (pending_ret)
1151                 return pending_ret;
1152         return 0;
1153
1154 }
1155
1156 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1157                                    struct btrfs_root *root)
1158 {
1159         struct extent_io_tree *block_group_cache;
1160         struct btrfs_block_group_cache *cache;
1161         int ret;
1162         int err = 0;
1163         int werr = 0;
1164         struct btrfs_path *path;
1165         u64 last = 0;
1166         u64 start;
1167         u64 end;
1168         u64 ptr;
1169
1170         block_group_cache = &root->fs_info->block_group_cache;
1171         path = btrfs_alloc_path();
1172         if (!path)
1173                 return -ENOMEM;
1174
1175         mutex_lock(&root->fs_info->alloc_mutex);
1176         while(1) {
1177                 ret = find_first_extent_bit(block_group_cache, last,
1178                                             &start, &end, BLOCK_GROUP_DIRTY);
1179                 if (ret)
1180                         break;
1181
1182                 last = end + 1;
1183                 ret = get_state_private(block_group_cache, start, &ptr);
1184                 if (ret)
1185                         break;
1186                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1187                 err = write_one_cache_group(trans, root,
1188                                             path, cache);
1189                 /*
1190                  * if we fail to write the cache group, we want
1191                  * to keep it marked dirty in hopes that a later
1192                  * write will work
1193                  */
1194                 if (err) {
1195                         werr = err;
1196                         continue;
1197                 }
1198                 clear_extent_bits(block_group_cache, start, end,
1199                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
1200         }
1201         btrfs_free_path(path);
1202         mutex_unlock(&root->fs_info->alloc_mutex);
1203         return werr;
1204 }
1205
1206 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1207                                                   u64 flags)
1208 {
1209         struct list_head *head = &info->space_info;
1210         struct list_head *cur;
1211         struct btrfs_space_info *found;
1212         list_for_each(cur, head) {
1213                 found = list_entry(cur, struct btrfs_space_info, list);
1214                 if (found->flags == flags)
1215                         return found;
1216         }
1217         return NULL;
1218
1219 }
1220
1221 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1222                              u64 total_bytes, u64 bytes_used,
1223                              struct btrfs_space_info **space_info)
1224 {
1225         struct btrfs_space_info *found;
1226
1227         found = __find_space_info(info, flags);
1228         if (found) {
1229                 found->total_bytes += total_bytes;
1230                 found->bytes_used += bytes_used;
1231                 found->full = 0;
1232                 *space_info = found;
1233                 return 0;
1234         }
1235         found = kmalloc(sizeof(*found), GFP_NOFS);
1236         if (!found)
1237                 return -ENOMEM;
1238
1239         list_add(&found->list, &info->space_info);
1240         found->flags = flags;
1241         found->total_bytes = total_bytes;
1242         found->bytes_used = bytes_used;
1243         found->bytes_pinned = 0;
1244         found->full = 0;
1245         found->force_alloc = 0;
1246         *space_info = found;
1247         return 0;
1248 }
1249
1250 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1251 {
1252         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1253                                    BTRFS_BLOCK_GROUP_RAID1 |
1254                                    BTRFS_BLOCK_GROUP_RAID10 |
1255                                    BTRFS_BLOCK_GROUP_DUP);
1256         if (extra_flags) {
1257                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1258                         fs_info->avail_data_alloc_bits |= extra_flags;
1259                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1260                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1261                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1262                         fs_info->avail_system_alloc_bits |= extra_flags;
1263         }
1264 }
1265
1266 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1267 {
1268         u64 num_devices = root->fs_info->fs_devices->num_devices;
1269
1270         if (num_devices == 1)
1271                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1272         if (num_devices < 4)
1273                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1274
1275         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1276             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1277                       BTRFS_BLOCK_GROUP_RAID10))) {
1278                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1279         }
1280
1281         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1282             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1283                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1284         }
1285
1286         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1287             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1288              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1289              (flags & BTRFS_BLOCK_GROUP_DUP)))
1290                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1291         return flags;
1292 }
1293
1294 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1295                           struct btrfs_root *extent_root, u64 alloc_bytes,
1296                           u64 flags, int force)
1297 {
1298         struct btrfs_space_info *space_info;
1299         u64 thresh;
1300         u64 start;
1301         u64 num_bytes;
1302         int ret;
1303
1304         flags = reduce_alloc_profile(extent_root, flags);
1305
1306         space_info = __find_space_info(extent_root->fs_info, flags);
1307         if (!space_info) {
1308                 ret = update_space_info(extent_root->fs_info, flags,
1309                                         0, 0, &space_info);
1310                 BUG_ON(ret);
1311         }
1312         BUG_ON(!space_info);
1313
1314         if (space_info->force_alloc) {
1315                 force = 1;
1316                 space_info->force_alloc = 0;
1317         }
1318         if (space_info->full)
1319                 goto out;
1320
1321         thresh = div_factor(space_info->total_bytes, 6);
1322         if (!force &&
1323            (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1324             thresh)
1325                 goto out;
1326
1327         mutex_lock(&extent_root->fs_info->chunk_mutex);
1328         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1329         if (ret == -ENOSPC) {
1330 printk("space info full %Lu\n", flags);
1331                 space_info->full = 1;
1332                 goto out_unlock;
1333         }
1334         BUG_ON(ret);
1335
1336         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1337                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1338         BUG_ON(ret);
1339 out_unlock:
1340         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1341 out:
1342         return 0;
1343 }
1344
1345 static int update_block_group(struct btrfs_trans_handle *trans,
1346                               struct btrfs_root *root,
1347                               u64 bytenr, u64 num_bytes, int alloc,
1348                               int mark_free)
1349 {
1350         struct btrfs_block_group_cache *cache;
1351         struct btrfs_fs_info *info = root->fs_info;
1352         u64 total = num_bytes;
1353         u64 old_val;
1354         u64 byte_in_group;
1355         u64 start;
1356         u64 end;
1357
1358         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1359         while(total) {
1360                 cache = btrfs_lookup_block_group(info, bytenr);
1361                 if (!cache) {
1362                         return -1;
1363                 }
1364                 byte_in_group = bytenr - cache->key.objectid;
1365                 WARN_ON(byte_in_group > cache->key.offset);
1366                 start = cache->key.objectid;
1367                 end = start + cache->key.offset - 1;
1368                 set_extent_bits(&info->block_group_cache, start, end,
1369                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1370
1371                 spin_lock(&cache->lock);
1372                 old_val = btrfs_block_group_used(&cache->item);
1373                 num_bytes = min(total, cache->key.offset - byte_in_group);
1374                 if (alloc) {
1375                         old_val += num_bytes;
1376                         cache->space_info->bytes_used += num_bytes;
1377                         btrfs_set_block_group_used(&cache->item, old_val);
1378                         spin_unlock(&cache->lock);
1379                 } else {
1380                         old_val -= num_bytes;
1381                         cache->space_info->bytes_used -= num_bytes;
1382                         btrfs_set_block_group_used(&cache->item, old_val);
1383                         spin_unlock(&cache->lock);
1384                         if (mark_free) {
1385                                 set_extent_dirty(&info->free_space_cache,
1386                                                  bytenr, bytenr + num_bytes - 1,
1387                                                  GFP_NOFS);
1388                         }
1389                 }
1390                 total -= num_bytes;
1391                 bytenr += num_bytes;
1392         }
1393         return 0;
1394 }
1395
1396 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1397 {
1398         u64 start;
1399         u64 end;
1400         int ret;
1401         ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1402                                     search_start, &start, &end,
1403                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1404                                     BLOCK_GROUP_SYSTEM);
1405         if (ret)
1406                 return 0;
1407         return start;
1408 }
1409
1410
1411 static int update_pinned_extents(struct btrfs_root *root,
1412                                 u64 bytenr, u64 num, int pin)
1413 {
1414         u64 len;
1415         struct btrfs_block_group_cache *cache;
1416         struct btrfs_fs_info *fs_info = root->fs_info;
1417
1418         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1419         if (pin) {
1420                 set_extent_dirty(&fs_info->pinned_extents,
1421                                 bytenr, bytenr + num - 1, GFP_NOFS);
1422         } else {
1423                 clear_extent_dirty(&fs_info->pinned_extents,
1424                                 bytenr, bytenr + num - 1, GFP_NOFS);
1425         }
1426         while (num > 0) {
1427                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1428                 if (!cache) {
1429                         u64 first = first_logical_byte(root, bytenr);
1430                         WARN_ON(first < bytenr);
1431                         len = min(first - bytenr, num);
1432                 } else {
1433                         len = min(num, cache->key.offset -
1434                                   (bytenr - cache->key.objectid));
1435                 }
1436                 if (pin) {
1437                         if (cache) {
1438                                 spin_lock(&cache->lock);
1439                                 cache->pinned += len;
1440                                 cache->space_info->bytes_pinned += len;
1441                                 spin_unlock(&cache->lock);
1442                         }
1443                         fs_info->total_pinned += len;
1444                 } else {
1445                         if (cache) {
1446                                 spin_lock(&cache->lock);
1447                                 cache->pinned -= len;
1448                                 cache->space_info->bytes_pinned -= len;
1449                                 spin_unlock(&cache->lock);
1450                         }
1451                         fs_info->total_pinned -= len;
1452                 }
1453                 bytenr += len;
1454                 num -= len;
1455         }
1456         return 0;
1457 }
1458
1459 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1460 {
1461         u64 last = 0;
1462         u64 start;
1463         u64 end;
1464         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1465         int ret;
1466
1467         while(1) {
1468                 ret = find_first_extent_bit(pinned_extents, last,
1469                                             &start, &end, EXTENT_DIRTY);
1470                 if (ret)
1471                         break;
1472                 set_extent_dirty(copy, start, end, GFP_NOFS);
1473                 last = end + 1;
1474         }
1475         return 0;
1476 }
1477
1478 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1479                                struct btrfs_root *root,
1480                                struct extent_io_tree *unpin)
1481 {
1482         u64 start;
1483         u64 end;
1484         int ret;
1485         struct extent_io_tree *free_space_cache;
1486         free_space_cache = &root->fs_info->free_space_cache;
1487
1488         mutex_lock(&root->fs_info->alloc_mutex);
1489         while(1) {
1490                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1491                                             EXTENT_DIRTY);
1492                 if (ret)
1493                         break;
1494                 update_pinned_extents(root, start, end + 1 - start, 0);
1495                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1496                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1497                 if (need_resched()) {
1498                         mutex_unlock(&root->fs_info->alloc_mutex);
1499                         cond_resched();
1500                         mutex_lock(&root->fs_info->alloc_mutex);
1501                 }
1502         }
1503         mutex_unlock(&root->fs_info->alloc_mutex);
1504         return 0;
1505 }
1506
1507 static int finish_current_insert(struct btrfs_trans_handle *trans,
1508                                  struct btrfs_root *extent_root)
1509 {
1510         u64 start;
1511         u64 end;
1512         struct btrfs_fs_info *info = extent_root->fs_info;
1513         struct extent_buffer *eb;
1514         struct btrfs_path *path;
1515         struct btrfs_key ins;
1516         struct btrfs_disk_key first;
1517         struct btrfs_extent_item extent_item;
1518         int ret;
1519         int level;
1520         int err = 0;
1521
1522         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1523         btrfs_set_stack_extent_refs(&extent_item, 1);
1524         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1525         path = btrfs_alloc_path();
1526
1527         while(1) {
1528                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1529                                             &end, EXTENT_LOCKED);
1530                 if (ret)
1531                         break;
1532
1533                 ins.objectid = start;
1534                 ins.offset = end + 1 - start;
1535                 err = btrfs_insert_item(trans, extent_root, &ins,
1536                                         &extent_item, sizeof(extent_item));
1537                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1538                                   GFP_NOFS);
1539
1540                 eb = btrfs_find_tree_block(extent_root, ins.objectid,
1541                                            ins.offset);
1542
1543                 if (!btrfs_buffer_uptodate(eb, trans->transid)) {
1544                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1545                         btrfs_read_buffer(eb, trans->transid);
1546                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1547                 }
1548
1549                 btrfs_tree_lock(eb);
1550                 level = btrfs_header_level(eb);
1551                 if (level == 0) {
1552                         btrfs_item_key(eb, &first, 0);
1553                 } else {
1554                         btrfs_node_key(eb, &first, 0);
1555                 }
1556                 btrfs_tree_unlock(eb);
1557                 free_extent_buffer(eb);
1558                 /*
1559                  * the first key is just a hint, so the race we've created
1560                  * against reading it is fine
1561                  */
1562                 err = btrfs_insert_extent_backref(trans, extent_root, path,
1563                                           start, extent_root->root_key.objectid,
1564                                           0, level,
1565                                           btrfs_disk_key_objectid(&first));
1566                 BUG_ON(err);
1567                 if (need_resched()) {
1568                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1569                         cond_resched();
1570                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1571                 }
1572         }
1573         btrfs_free_path(path);
1574         return 0;
1575 }
1576
1577 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1578                           int pending)
1579 {
1580         int err = 0;
1581
1582         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1583         if (!pending) {
1584                 struct extent_buffer *buf;
1585                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1586                 if (buf) {
1587                         if (btrfs_buffer_uptodate(buf, 0) &&
1588                             btrfs_try_tree_lock(buf)) {
1589                                 u64 transid =
1590                                     root->fs_info->running_transaction->transid;
1591                                 u64 header_transid =
1592                                         btrfs_header_generation(buf);
1593                                 if (header_transid == transid &&
1594                                     !btrfs_header_flag(buf,
1595                                                BTRFS_HEADER_FLAG_WRITTEN)) {
1596                                         clean_tree_block(NULL, root, buf);
1597                                         btrfs_tree_unlock(buf);
1598                                         free_extent_buffer(buf);
1599                                         return 1;
1600                                 }
1601                                 btrfs_tree_unlock(buf);
1602                         }
1603                         free_extent_buffer(buf);
1604                 }
1605                 update_pinned_extents(root, bytenr, num_bytes, 1);
1606         } else {
1607                 set_extent_bits(&root->fs_info->pending_del,
1608                                 bytenr, bytenr + num_bytes - 1,
1609                                 EXTENT_LOCKED, GFP_NOFS);
1610         }
1611         BUG_ON(err < 0);
1612         return 0;
1613 }
1614
1615 /*
1616  * remove an extent from the root, returns 0 on success
1617  */
1618 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1619                          *root, u64 bytenr, u64 num_bytes,
1620                          u64 root_objectid, u64 ref_generation,
1621                          u64 owner_objectid, u64 owner_offset, int pin,
1622                          int mark_free)
1623 {
1624         struct btrfs_path *path;
1625         struct btrfs_key key;
1626         struct btrfs_fs_info *info = root->fs_info;
1627         struct btrfs_root *extent_root = info->extent_root;
1628         struct extent_buffer *leaf;
1629         int ret;
1630         int extent_slot = 0;
1631         int found_extent = 0;
1632         int num_to_del = 1;
1633         struct btrfs_extent_item *ei;
1634         u32 refs;
1635
1636         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1637         key.objectid = bytenr;
1638         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1639         key.offset = num_bytes;
1640         path = btrfs_alloc_path();
1641         if (!path)
1642                 return -ENOMEM;
1643
1644         path->reada = 1;
1645         ret = lookup_extent_backref(trans, extent_root, path,
1646                                     bytenr, root_objectid,
1647                                     ref_generation,
1648                                     owner_objectid, owner_offset, 1);
1649         if (ret == 0) {
1650                 struct btrfs_key found_key;
1651                 extent_slot = path->slots[0];
1652                 while(extent_slot > 0) {
1653                         extent_slot--;
1654                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1655                                               extent_slot);
1656                         if (found_key.objectid != bytenr)
1657                                 break;
1658                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1659                             found_key.offset == num_bytes) {
1660                                 found_extent = 1;
1661                                 break;
1662                         }
1663                         if (path->slots[0] - extent_slot > 5)
1664                                 break;
1665                 }
1666                 if (!found_extent)
1667                         ret = btrfs_del_item(trans, extent_root, path);
1668         } else {
1669                 btrfs_print_leaf(extent_root, path->nodes[0]);
1670                 WARN_ON(1);
1671                 printk("Unable to find ref byte nr %Lu root %Lu "
1672                        " gen %Lu owner %Lu offset %Lu\n", bytenr,
1673                        root_objectid, ref_generation, owner_objectid,
1674                        owner_offset);
1675         }
1676         if (!found_extent) {
1677                 btrfs_release_path(extent_root, path);
1678                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1679                 if (ret < 0)
1680                         return ret;
1681                 BUG_ON(ret);
1682                 extent_slot = path->slots[0];
1683         }
1684
1685         leaf = path->nodes[0];
1686         ei = btrfs_item_ptr(leaf, extent_slot,
1687                             struct btrfs_extent_item);
1688         refs = btrfs_extent_refs(leaf, ei);
1689         BUG_ON(refs == 0);
1690         refs -= 1;
1691         btrfs_set_extent_refs(leaf, ei, refs);
1692
1693         btrfs_mark_buffer_dirty(leaf);
1694
1695         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1696                 /* if the back ref and the extent are next to each other
1697                  * they get deleted below in one shot
1698                  */
1699                 path->slots[0] = extent_slot;
1700                 num_to_del = 2;
1701         } else if (found_extent) {
1702                 /* otherwise delete the extent back ref */
1703                 ret = btrfs_del_item(trans, extent_root, path);
1704                 BUG_ON(ret);
1705                 /* if refs are 0, we need to setup the path for deletion */
1706                 if (refs == 0) {
1707                         btrfs_release_path(extent_root, path);
1708                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1709                                                 -1, 1);
1710                         if (ret < 0)
1711                                 return ret;
1712                         BUG_ON(ret);
1713                 }
1714         }
1715
1716         if (refs == 0) {
1717                 u64 super_used;
1718                 u64 root_used;
1719
1720                 if (pin) {
1721                         ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1722                         if (ret > 0)
1723                                 mark_free = 1;
1724                         BUG_ON(ret < 0);
1725                 }
1726
1727                 /* block accounting for super block */
1728                 spin_lock_irq(&info->delalloc_lock);
1729                 super_used = btrfs_super_bytes_used(&info->super_copy);
1730                 btrfs_set_super_bytes_used(&info->super_copy,
1731                                            super_used - num_bytes);
1732                 spin_unlock_irq(&info->delalloc_lock);
1733
1734                 /* block accounting for root item */
1735                 root_used = btrfs_root_used(&root->root_item);
1736                 btrfs_set_root_used(&root->root_item,
1737                                            root_used - num_bytes);
1738                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1739                                       num_to_del);
1740                 if (ret) {
1741                         return ret;
1742                 }
1743                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1744                                          mark_free);
1745                 BUG_ON(ret);
1746         }
1747         btrfs_free_path(path);
1748         finish_current_insert(trans, extent_root);
1749         return ret;
1750 }
1751
1752 /*
1753  * find all the blocks marked as pending in the radix tree and remove
1754  * them from the extent map
1755  */
1756 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1757                                btrfs_root *extent_root)
1758 {
1759         int ret;
1760         int err = 0;
1761         u64 start;
1762         u64 end;
1763         struct extent_io_tree *pending_del;
1764         struct extent_io_tree *pinned_extents;
1765
1766         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1767         pending_del = &extent_root->fs_info->pending_del;
1768         pinned_extents = &extent_root->fs_info->pinned_extents;
1769
1770         while(1) {
1771                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1772                                             EXTENT_LOCKED);
1773                 if (ret)
1774                         break;
1775                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1776                                   GFP_NOFS);
1777                 if (!test_range_bit(&extent_root->fs_info->extent_ins,
1778                                     start, end, EXTENT_LOCKED, 0)) {
1779                         update_pinned_extents(extent_root, start,
1780                                               end + 1 - start, 1);
1781                         ret = __free_extent(trans, extent_root,
1782                                              start, end + 1 - start,
1783                                              extent_root->root_key.objectid,
1784                                              0, 0, 0, 0, 0);
1785                 } else {
1786                         clear_extent_bits(&extent_root->fs_info->extent_ins,
1787                                           start, end, EXTENT_LOCKED, GFP_NOFS);
1788                 }
1789                 if (ret)
1790                         err = ret;
1791
1792                 if (need_resched()) {
1793                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1794                         cond_resched();
1795                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1796                 }
1797         }
1798         return err;
1799 }
1800
1801 /*
1802  * remove an extent from the root, returns 0 on success
1803  */
1804 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1805                                struct btrfs_root *root, u64 bytenr,
1806                                u64 num_bytes, u64 root_objectid,
1807                                u64 ref_generation, u64 owner_objectid,
1808                                u64 owner_offset, int pin)
1809 {
1810         struct btrfs_root *extent_root = root->fs_info->extent_root;
1811         int pending_ret;
1812         int ret;
1813
1814         WARN_ON(num_bytes < root->sectorsize);
1815         if (!root->ref_cows)
1816                 ref_generation = 0;
1817
1818         if (root == extent_root) {
1819                 pin_down_bytes(root, bytenr, num_bytes, 1);
1820                 return 0;
1821         }
1822         ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1823                             ref_generation, owner_objectid, owner_offset,
1824                             pin, pin == 0);
1825
1826         finish_current_insert(trans, root->fs_info->extent_root);
1827         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1828         return ret ? ret : pending_ret;
1829 }
1830
1831 int btrfs_free_extent(struct btrfs_trans_handle *trans,
1832                       struct btrfs_root *root, u64 bytenr,
1833                       u64 num_bytes, u64 root_objectid,
1834                       u64 ref_generation, u64 owner_objectid,
1835                       u64 owner_offset, int pin)
1836 {
1837         int ret;
1838
1839         maybe_lock_mutex(root);
1840         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1841                                   root_objectid, ref_generation,
1842                                   owner_objectid, owner_offset, pin);
1843         maybe_unlock_mutex(root);
1844         return ret;
1845 }
1846
1847 static u64 stripe_align(struct btrfs_root *root, u64 val)
1848 {
1849         u64 mask = ((u64)root->stripesize - 1);
1850         u64 ret = (val + mask) & ~mask;
1851         return ret;
1852 }
1853
1854 /*
1855  * walks the btree of allocated extents and find a hole of a given size.
1856  * The key ins is changed to record the hole:
1857  * ins->objectid == block start
1858  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1859  * ins->offset == number of blocks
1860  * Any available blocks before search_start are skipped.
1861  */
1862 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1863                                      struct btrfs_root *orig_root,
1864                                      u64 num_bytes, u64 empty_size,
1865                                      u64 search_start, u64 search_end,
1866                                      u64 hint_byte, struct btrfs_key *ins,
1867                                      u64 exclude_start, u64 exclude_nr,
1868                                      int data)
1869 {
1870         int ret;
1871         u64 orig_search_start;
1872         struct btrfs_root * root = orig_root->fs_info->extent_root;
1873         struct btrfs_fs_info *info = root->fs_info;
1874         u64 total_needed = num_bytes;
1875         u64 *last_ptr = NULL;
1876         struct btrfs_block_group_cache *block_group;
1877         int full_scan = 0;
1878         int wrapped = 0;
1879         int chunk_alloc_done = 0;
1880         int empty_cluster = 2 * 1024 * 1024;
1881         int allowed_chunk_alloc = 0;
1882
1883         WARN_ON(num_bytes < root->sectorsize);
1884         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1885
1886         if (orig_root->ref_cows || empty_size)
1887                 allowed_chunk_alloc = 1;
1888
1889         if (data & BTRFS_BLOCK_GROUP_METADATA) {
1890                 last_ptr = &root->fs_info->last_alloc;
1891                 empty_cluster = 256 * 1024;
1892         }
1893
1894         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1895                 last_ptr = &root->fs_info->last_data_alloc;
1896         }
1897
1898         if (last_ptr) {
1899                 if (*last_ptr)
1900                         hint_byte = *last_ptr;
1901                 else {
1902                         empty_size += empty_cluster;
1903                 }
1904         }
1905
1906         search_start = max(search_start, first_logical_byte(root, 0));
1907         orig_search_start = search_start;
1908
1909         if (search_end == (u64)-1)
1910                 search_end = btrfs_super_total_bytes(&info->super_copy);
1911
1912         if (hint_byte) {
1913                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
1914                 if (!block_group)
1915                         hint_byte = search_start;
1916                 block_group = btrfs_find_block_group(root, block_group,
1917                                                      hint_byte, data, 1);
1918                 if (last_ptr && *last_ptr == 0 && block_group)
1919                         hint_byte = block_group->key.objectid;
1920         } else {
1921                 block_group = btrfs_find_block_group(root,
1922                                                      trans->block_group,
1923                                                      search_start, data, 1);
1924         }
1925         search_start = max(search_start, hint_byte);
1926
1927         total_needed += empty_size;
1928
1929 check_failed:
1930         if (!block_group) {
1931                 block_group = btrfs_lookup_first_block_group(info,
1932                                                              search_start);
1933                 if (!block_group)
1934                         block_group = btrfs_lookup_first_block_group(info,
1935                                                        orig_search_start);
1936         }
1937         if (full_scan && !chunk_alloc_done) {
1938                 if (allowed_chunk_alloc) {
1939                         do_chunk_alloc(trans, root,
1940                                      num_bytes + 2 * 1024 * 1024, data, 1);
1941                         allowed_chunk_alloc = 0;
1942                 } else if (block_group && block_group_bits(block_group, data)) {
1943                         block_group->space_info->force_alloc = 1;
1944                 }
1945                 chunk_alloc_done = 1;
1946         }
1947         ret = find_search_start(root, &block_group, &search_start,
1948                                 total_needed, data);
1949         if (ret == -ENOSPC && last_ptr && *last_ptr) {
1950                 *last_ptr = 0;
1951                 block_group = btrfs_lookup_first_block_group(info,
1952                                                              orig_search_start);
1953                 search_start = orig_search_start;
1954                 ret = find_search_start(root, &block_group, &search_start,
1955                                         total_needed, data);
1956         }
1957         if (ret == -ENOSPC)
1958                 goto enospc;
1959         if (ret)
1960                 goto error;
1961
1962         if (last_ptr && *last_ptr && search_start != *last_ptr) {
1963                 *last_ptr = 0;
1964                 if (!empty_size) {
1965                         empty_size += empty_cluster;
1966                         total_needed += empty_size;
1967                 }
1968                 block_group = btrfs_lookup_first_block_group(info,
1969                                                        orig_search_start);
1970                 search_start = orig_search_start;
1971                 ret = find_search_start(root, &block_group,
1972                                         &search_start, total_needed, data);
1973                 if (ret == -ENOSPC)
1974                         goto enospc;
1975                 if (ret)
1976                         goto error;
1977         }
1978
1979         search_start = stripe_align(root, search_start);
1980         ins->objectid = search_start;
1981         ins->offset = num_bytes;
1982
1983         if (ins->objectid + num_bytes >= search_end)
1984                 goto enospc;
1985
1986         if (ins->objectid + num_bytes >
1987             block_group->key.objectid + block_group->key.offset) {
1988                 search_start = block_group->key.objectid +
1989                         block_group->key.offset;
1990                 goto new_group;
1991         }
1992
1993         if (test_range_bit(&info->extent_ins, ins->objectid,
1994                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1995                 search_start = ins->objectid + num_bytes;
1996                 goto new_group;
1997         }
1998
1999         if (test_range_bit(&info->pinned_extents, ins->objectid,
2000                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2001                 search_start = ins->objectid + num_bytes;
2002                 goto new_group;
2003         }
2004
2005         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2006             ins->objectid < exclude_start + exclude_nr)) {
2007                 search_start = exclude_start + exclude_nr;
2008                 goto new_group;
2009         }
2010
2011         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2012                 block_group = btrfs_lookup_block_group(info, ins->objectid);
2013                 if (block_group)
2014                         trans->block_group = block_group;
2015         }
2016         ins->offset = num_bytes;
2017         if (last_ptr) {
2018                 *last_ptr = ins->objectid + ins->offset;
2019                 if (*last_ptr ==
2020                     btrfs_super_total_bytes(&root->fs_info->super_copy)) {
2021                         *last_ptr = 0;
2022                 }
2023         }
2024         return 0;
2025
2026 new_group:
2027         if (search_start + num_bytes >= search_end) {
2028 enospc:
2029                 search_start = orig_search_start;
2030                 if (full_scan) {
2031                         ret = -ENOSPC;
2032                         goto error;
2033                 }
2034                 if (wrapped) {
2035                         if (!full_scan)
2036                                 total_needed -= empty_size;
2037                         full_scan = 1;
2038                 } else
2039                         wrapped = 1;
2040         }
2041         block_group = btrfs_lookup_first_block_group(info, search_start);
2042         cond_resched();
2043         block_group = btrfs_find_block_group(root, block_group,
2044                                              search_start, data, 0);
2045         goto check_failed;
2046
2047 error:
2048         return ret;
2049 }
2050
2051 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2052                                   struct btrfs_root *root,
2053                                   u64 num_bytes, u64 min_alloc_size,
2054                                   u64 empty_size, u64 hint_byte,
2055                                   u64 search_end, struct btrfs_key *ins,
2056                                   u64 data)
2057 {
2058         int ret;
2059         u64 search_start = 0;
2060         u64 alloc_profile;
2061         struct btrfs_fs_info *info = root->fs_info;
2062
2063         if (data) {
2064                 alloc_profile = info->avail_data_alloc_bits &
2065                                 info->data_alloc_profile;
2066                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2067         } else if (root == root->fs_info->chunk_root) {
2068                 alloc_profile = info->avail_system_alloc_bits &
2069                                 info->system_alloc_profile;
2070                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2071         } else {
2072                 alloc_profile = info->avail_metadata_alloc_bits &
2073                                 info->metadata_alloc_profile;
2074                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2075         }
2076 again:
2077         data = reduce_alloc_profile(root, data);
2078         /*
2079          * the only place that sets empty_size is btrfs_realloc_node, which
2080          * is not called recursively on allocations
2081          */
2082         if (empty_size || root->ref_cows) {
2083                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2084                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2085                                      2 * 1024 * 1024,
2086                                      BTRFS_BLOCK_GROUP_METADATA |
2087                                      (info->metadata_alloc_profile &
2088                                       info->avail_metadata_alloc_bits), 0);
2089                         BUG_ON(ret);
2090                 }
2091                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2092                                      num_bytes + 2 * 1024 * 1024, data, 0);
2093                 BUG_ON(ret);
2094         }
2095
2096         WARN_ON(num_bytes < root->sectorsize);
2097         ret = find_free_extent(trans, root, num_bytes, empty_size,
2098                                search_start, search_end, hint_byte, ins,
2099                                trans->alloc_exclude_start,
2100                                trans->alloc_exclude_nr, data);
2101
2102         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2103                 num_bytes = num_bytes >> 1;
2104                 num_bytes = max(num_bytes, min_alloc_size);
2105                 do_chunk_alloc(trans, root->fs_info->extent_root,
2106                                num_bytes, data, 1);
2107                 goto again;
2108         }
2109         if (ret) {
2110                 printk("allocation failed flags %Lu\n", data);
2111                 BUG();
2112         }
2113         clear_extent_dirty(&root->fs_info->free_space_cache,
2114                            ins->objectid, ins->objectid + ins->offset - 1,
2115                            GFP_NOFS);
2116         return 0;
2117 }
2118
2119 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2120 {
2121         maybe_lock_mutex(root);
2122         set_extent_dirty(&root->fs_info->free_space_cache,
2123                          start, start + len - 1, GFP_NOFS);
2124         maybe_unlock_mutex(root);
2125         return 0;
2126 }
2127
2128 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2129                                   struct btrfs_root *root,
2130                                   u64 num_bytes, u64 min_alloc_size,
2131                                   u64 empty_size, u64 hint_byte,
2132                                   u64 search_end, struct btrfs_key *ins,
2133                                   u64 data)
2134 {
2135         int ret;
2136         maybe_lock_mutex(root);
2137         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2138                                      empty_size, hint_byte, search_end, ins,
2139                                      data);
2140         maybe_unlock_mutex(root);
2141         return ret;
2142 }
2143
2144 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2145                                          struct btrfs_root *root,
2146                                          u64 root_objectid, u64 ref_generation,
2147                                          u64 owner, u64 owner_offset,
2148                                          struct btrfs_key *ins)
2149 {
2150         int ret;
2151         int pending_ret;
2152         u64 super_used;
2153         u64 root_used;
2154         u64 num_bytes = ins->offset;
2155         u32 sizes[2];
2156         struct btrfs_fs_info *info = root->fs_info;
2157         struct btrfs_root *extent_root = info->extent_root;
2158         struct btrfs_extent_item *extent_item;
2159         struct btrfs_extent_ref *ref;
2160         struct btrfs_path *path;
2161         struct btrfs_key keys[2];
2162
2163         /* block accounting for super block */
2164         spin_lock_irq(&info->delalloc_lock);
2165         super_used = btrfs_super_bytes_used(&info->super_copy);
2166         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2167         spin_unlock_irq(&info->delalloc_lock);
2168
2169         /* block accounting for root item */
2170         root_used = btrfs_root_used(&root->root_item);
2171         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2172
2173         if (root == extent_root) {
2174                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2175                                 ins->objectid + ins->offset - 1,
2176                                 EXTENT_LOCKED, GFP_NOFS);
2177                 goto update_block;
2178         }
2179
2180         memcpy(&keys[0], ins, sizeof(*ins));
2181         keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
2182                                          owner, owner_offset);
2183         keys[1].objectid = ins->objectid;
2184         keys[1].type = BTRFS_EXTENT_REF_KEY;
2185         sizes[0] = sizeof(*extent_item);
2186         sizes[1] = sizeof(*ref);
2187
2188         path = btrfs_alloc_path();
2189         BUG_ON(!path);
2190
2191         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2192                                        sizes, 2);
2193
2194         BUG_ON(ret);
2195         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2196                                      struct btrfs_extent_item);
2197         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2198         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2199                              struct btrfs_extent_ref);
2200
2201         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2202         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2203         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2204         btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2205
2206         btrfs_mark_buffer_dirty(path->nodes[0]);
2207
2208         trans->alloc_exclude_start = 0;
2209         trans->alloc_exclude_nr = 0;
2210         btrfs_free_path(path);
2211         finish_current_insert(trans, extent_root);
2212         pending_ret = del_pending_extents(trans, extent_root);
2213
2214         if (ret)
2215                 goto out;
2216         if (pending_ret) {
2217                 ret = pending_ret;
2218                 goto out;
2219         }
2220
2221 update_block:
2222         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2223         if (ret) {
2224                 printk("update block group failed for %Lu %Lu\n",
2225                        ins->objectid, ins->offset);
2226                 BUG();
2227         }
2228 out:
2229         return ret;
2230 }
2231
2232 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2233                                 struct btrfs_root *root,
2234                                 u64 root_objectid, u64 ref_generation,
2235                                 u64 owner, u64 owner_offset,
2236                                 struct btrfs_key *ins)
2237 {
2238         int ret;
2239         maybe_lock_mutex(root);
2240         ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2241                                             ref_generation, owner,
2242                                             owner_offset, ins);
2243         maybe_unlock_mutex(root);
2244         return ret;
2245 }
2246 /*
2247  * finds a free extent and does all the dirty work required for allocation
2248  * returns the key for the extent through ins, and a tree buffer for
2249  * the first block of the extent through buf.
2250  *
2251  * returns 0 if everything worked, non-zero otherwise.
2252  */
2253 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2254                        struct btrfs_root *root,
2255                        u64 num_bytes, u64 min_alloc_size,
2256                        u64 root_objectid, u64 ref_generation,
2257                        u64 owner, u64 owner_offset,
2258                        u64 empty_size, u64 hint_byte,
2259                        u64 search_end, struct btrfs_key *ins, u64 data)
2260 {
2261         int ret;
2262
2263         maybe_lock_mutex(root);
2264
2265         ret = __btrfs_reserve_extent(trans, root, num_bytes,
2266                                      min_alloc_size, empty_size, hint_byte,
2267                                      search_end, ins, data);
2268         BUG_ON(ret);
2269         ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2270                                             ref_generation, owner,
2271                                             owner_offset, ins);
2272         BUG_ON(ret);
2273
2274         maybe_unlock_mutex(root);
2275         return ret;
2276 }
2277
2278 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2279                                             struct btrfs_root *root,
2280                                             u64 bytenr, u32 blocksize)
2281 {
2282         struct extent_buffer *buf;
2283
2284         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2285         if (!buf)
2286                 return ERR_PTR(-ENOMEM);
2287         btrfs_set_header_generation(buf, trans->transid);
2288         btrfs_tree_lock(buf);
2289         clean_tree_block(trans, root, buf);
2290         btrfs_set_buffer_uptodate(buf);
2291         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2292                          buf->start + buf->len - 1, GFP_NOFS);
2293         trans->blocks_used++;
2294         return buf;
2295 }
2296
2297 /*
2298  * helper function to allocate a block for a given tree
2299  * returns the tree buffer or NULL.
2300  */
2301 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2302                                              struct btrfs_root *root,
2303                                              u32 blocksize,
2304                                              u64 root_objectid,
2305                                              u64 ref_generation,
2306                                              u64 first_objectid,
2307                                              int level,
2308                                              u64 hint,
2309                                              u64 empty_size)
2310 {
2311         struct btrfs_key ins;
2312         int ret;
2313         struct extent_buffer *buf;
2314
2315         ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2316                                  root_objectid, ref_generation,
2317                                  level, first_objectid, empty_size, hint,
2318                                  (u64)-1, &ins, 0);
2319         if (ret) {
2320                 BUG_ON(ret > 0);
2321                 return ERR_PTR(ret);
2322         }
2323
2324         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2325         return buf;
2326 }
2327
2328 static int noinline drop_leaf_ref_no_cache(struct btrfs_trans_handle *trans,
2329                                            struct btrfs_root *root,
2330                                            struct extent_buffer *leaf)
2331 {
2332         u64 leaf_owner;
2333         u64 leaf_generation;
2334         struct btrfs_key key;
2335         struct btrfs_file_extent_item *fi;
2336         int i;
2337         int nritems;
2338         int ret;
2339
2340         BUG_ON(!btrfs_is_leaf(leaf));
2341         nritems = btrfs_header_nritems(leaf);
2342         leaf_owner = btrfs_header_owner(leaf);
2343         leaf_generation = btrfs_header_generation(leaf);
2344
2345         for (i = 0; i < nritems; i++) {
2346                 u64 disk_bytenr;
2347                 cond_resched();
2348
2349                 btrfs_item_key_to_cpu(leaf, &key, i);
2350                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2351                         continue;
2352                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2353                 if (btrfs_file_extent_type(leaf, fi) ==
2354                     BTRFS_FILE_EXTENT_INLINE)
2355                         continue;
2356                 /*
2357                  * FIXME make sure to insert a trans record that
2358                  * repeats the snapshot del on crash
2359                  */
2360                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2361                 if (disk_bytenr == 0)
2362                         continue;
2363
2364                 mutex_lock(&root->fs_info->alloc_mutex);
2365                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2366                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2367                                 leaf_owner, leaf_generation,
2368                                 key.objectid, key.offset, 0);
2369                 mutex_unlock(&root->fs_info->alloc_mutex);
2370
2371                 atomic_inc(&root->fs_info->throttle_gen);
2372                 wake_up(&root->fs_info->transaction_throttle);
2373                 cond_resched();
2374
2375                 BUG_ON(ret);
2376         }
2377         return 0;
2378 }
2379
2380 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2381                                          struct btrfs_root *root,
2382                                          struct btrfs_leaf_ref *ref)
2383 {
2384         int i;
2385         int ret;
2386         struct btrfs_extent_info *info = ref->extents;
2387
2388         for (i = 0; i < ref->nritems; i++) {
2389                 mutex_lock(&root->fs_info->alloc_mutex);
2390                 ret = __btrfs_free_extent(trans, root,
2391                                         info->bytenr, info->num_bytes,
2392                                         ref->owner, ref->generation,
2393                                         info->objectid, info->offset, 0);
2394                 mutex_unlock(&root->fs_info->alloc_mutex);
2395
2396                 atomic_inc(&root->fs_info->throttle_gen);
2397                 wake_up(&root->fs_info->transaction_throttle);
2398                 cond_resched();
2399
2400                 BUG_ON(ret);
2401                 info++;
2402         }
2403
2404         return 0;
2405 }
2406
2407 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2408                               u32 *refs)
2409 {
2410         int ret;
2411
2412         ret = lookup_extent_ref(NULL, root, start, len, refs);
2413         BUG_ON(ret);
2414
2415 #if 0 // some debugging code in case we see problems here
2416         /* if the refs count is one, it won't get increased again.  But
2417          * if the ref count is > 1, someone may be decreasing it at
2418          * the same time we are.
2419          */
2420         if (*refs != 1) {
2421                 struct extent_buffer *eb = NULL;
2422                 eb = btrfs_find_create_tree_block(root, start, len);
2423                 if (eb)
2424                         btrfs_tree_lock(eb);
2425
2426                 mutex_lock(&root->fs_info->alloc_mutex);
2427                 ret = lookup_extent_ref(NULL, root, start, len, refs);
2428                 BUG_ON(ret);
2429                 mutex_unlock(&root->fs_info->alloc_mutex);
2430
2431                 if (eb) {
2432                         btrfs_tree_unlock(eb);
2433                         free_extent_buffer(eb);
2434                 }
2435                 if (*refs == 1) {
2436                         printk("block %llu went down to one during drop_snap\n",
2437                                (unsigned long long)start);
2438                 }
2439
2440         }
2441 #endif
2442
2443         cond_resched();
2444         return ret;
2445 }
2446
2447 /*
2448  * helper function for drop_snapshot, this walks down the tree dropping ref
2449  * counts as it goes.
2450  */
2451 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2452                                    struct btrfs_root *root,
2453                                    struct btrfs_path *path, int *level)
2454 {
2455         u64 root_owner;
2456         u64 root_gen;
2457         u64 bytenr;
2458         u64 ptr_gen;
2459         struct extent_buffer *next;
2460         struct extent_buffer *cur;
2461         struct extent_buffer *parent;
2462         struct btrfs_leaf_ref *ref;
2463         u32 blocksize;
2464         int ret;
2465         u32 refs;
2466
2467         WARN_ON(*level < 0);
2468         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2469         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2470                                 path->nodes[*level]->len, &refs);
2471         BUG_ON(ret);
2472         if (refs > 1)
2473                 goto out;
2474
2475         /*
2476          * walk down to the last node level and free all the leaves
2477          */
2478         while(*level >= 0) {
2479                 WARN_ON(*level < 0);
2480                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2481                 cur = path->nodes[*level];
2482
2483                 if (btrfs_header_level(cur) != *level)
2484                         WARN_ON(1);
2485
2486                 if (path->slots[*level] >=
2487                     btrfs_header_nritems(cur))
2488                         break;
2489                 if (*level == 0) {
2490                         ret = drop_leaf_ref_no_cache(trans, root, cur);
2491                         BUG_ON(ret);
2492                         break;
2493                 }
2494                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2495                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2496                 blocksize = btrfs_level_size(root, *level - 1);
2497
2498                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2499                 BUG_ON(ret);
2500                 if (refs != 1) {
2501                         parent = path->nodes[*level];
2502                         root_owner = btrfs_header_owner(parent);
2503                         root_gen = btrfs_header_generation(parent);
2504                         path->slots[*level]++;
2505
2506                         mutex_lock(&root->fs_info->alloc_mutex);
2507                         ret = __btrfs_free_extent(trans, root, bytenr,
2508                                                 blocksize, root_owner,
2509                                                 root_gen, 0, 0, 1);
2510                         BUG_ON(ret);
2511                         mutex_unlock(&root->fs_info->alloc_mutex);
2512
2513                         atomic_inc(&root->fs_info->throttle_gen);
2514                         wake_up(&root->fs_info->transaction_throttle);
2515                         cond_resched();
2516
2517                         continue;
2518                 }
2519                 /*
2520                  * at this point, we have a single ref, and since the
2521                  * only place referencing this extent is a dead root
2522                  * the reference count should never go higher.
2523                  * So, we don't need to check it again
2524                  */
2525                 if (*level == 1) {
2526                         struct btrfs_key key;
2527                         btrfs_node_key_to_cpu(cur, &key, path->slots[*level]);
2528                         ref = btrfs_lookup_leaf_ref(root, bytenr);
2529                         if (ref) {
2530                                 ret = drop_leaf_ref(trans, root, ref);
2531                                 BUG_ON(ret);
2532                                 btrfs_remove_leaf_ref(root, ref);
2533                                 btrfs_free_leaf_ref(root, ref);
2534                                 *level = 0;
2535                                 break;
2536                         }
2537                         if (printk_ratelimit())
2538                                 printk("leaf ref miss for bytenr %llu\n",
2539                                        (unsigned long long)bytenr);
2540                 }
2541                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2542                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2543                         free_extent_buffer(next);
2544
2545                         next = read_tree_block(root, bytenr, blocksize,
2546                                                ptr_gen);
2547                         cond_resched();
2548 #if 0
2549                         /*
2550                          * this is a debugging check and can go away
2551                          * the ref should never go all the way down to 1
2552                          * at this point
2553                          */
2554                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2555                                                 &refs);
2556                         BUG_ON(ret);
2557                         WARN_ON(refs != 1);
2558 #endif
2559                 }
2560                 WARN_ON(*level <= 0);
2561                 if (path->nodes[*level-1])
2562                         free_extent_buffer(path->nodes[*level-1]);
2563                 path->nodes[*level-1] = next;
2564                 *level = btrfs_header_level(next);
2565                 path->slots[*level] = 0;
2566                 cond_resched();
2567         }
2568 out:
2569         WARN_ON(*level < 0);
2570         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2571
2572         if (path->nodes[*level] == root->node) {
2573                 parent = path->nodes[*level];
2574                 bytenr = path->nodes[*level]->start;
2575         } else {
2576                 parent = path->nodes[*level + 1];
2577                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
2578         }
2579
2580         blocksize = btrfs_level_size(root, *level);
2581         root_owner = btrfs_header_owner(parent);
2582         root_gen = btrfs_header_generation(parent);
2583
2584         mutex_lock(&root->fs_info->alloc_mutex);
2585         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
2586                                   root_owner, root_gen, 0, 0, 1);
2587         free_extent_buffer(path->nodes[*level]);
2588         path->nodes[*level] = NULL;
2589         *level += 1;
2590         BUG_ON(ret);
2591         mutex_unlock(&root->fs_info->alloc_mutex);
2592
2593         cond_resched();
2594         return 0;
2595 }
2596
2597 /*
2598  * helper for dropping snapshots.  This walks back up the tree in the path
2599  * to find the first node higher up where we haven't yet gone through
2600  * all the slots
2601  */
2602 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2603                                  struct btrfs_root *root,
2604                                  struct btrfs_path *path, int *level)
2605 {
2606         u64 root_owner;
2607         u64 root_gen;
2608         struct btrfs_root_item *root_item = &root->root_item;
2609         int i;
2610         int slot;
2611         int ret;
2612
2613         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2614                 slot = path->slots[i];
2615                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2616                         struct extent_buffer *node;
2617                         struct btrfs_disk_key disk_key;
2618                         node = path->nodes[i];
2619                         path->slots[i]++;
2620                         *level = i;
2621                         WARN_ON(*level == 0);
2622                         btrfs_node_key(node, &disk_key, path->slots[i]);
2623                         memcpy(&root_item->drop_progress,
2624                                &disk_key, sizeof(disk_key));
2625                         root_item->drop_level = i;
2626                         return 0;
2627                 } else {
2628                         if (path->nodes[*level] == root->node) {
2629                                 root_owner = root->root_key.objectid;
2630                                 root_gen =
2631                                    btrfs_header_generation(path->nodes[*level]);
2632                         } else {
2633                                 struct extent_buffer *node;
2634                                 node = path->nodes[*level + 1];
2635                                 root_owner = btrfs_header_owner(node);
2636                                 root_gen = btrfs_header_generation(node);
2637                         }
2638                         ret = btrfs_free_extent(trans, root,
2639                                                 path->nodes[*level]->start,
2640                                                 path->nodes[*level]->len,
2641                                                 root_owner, root_gen, 0, 0, 1);
2642                         BUG_ON(ret);
2643                         free_extent_buffer(path->nodes[*level]);
2644                         path->nodes[*level] = NULL;
2645                         *level = i + 1;
2646                 }
2647         }
2648         return 1;
2649 }
2650
2651 /*
2652  * drop the reference count on the tree rooted at 'snap'.  This traverses
2653  * the tree freeing any blocks that have a ref count of zero after being
2654  * decremented.
2655  */
2656 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2657                         *root)
2658 {
2659         int ret = 0;
2660         int wret;
2661         int level;
2662         struct btrfs_path *path;
2663         int i;
2664         int orig_level;
2665         struct btrfs_root_item *root_item = &root->root_item;
2666
2667         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2668         path = btrfs_alloc_path();
2669         BUG_ON(!path);
2670
2671         level = btrfs_header_level(root->node);
2672         orig_level = level;
2673         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2674                 path->nodes[level] = root->node;
2675                 extent_buffer_get(root->node);
2676                 path->slots[level] = 0;
2677         } else {
2678                 struct btrfs_key key;
2679                 struct btrfs_disk_key found_key;
2680                 struct extent_buffer *node;
2681
2682                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2683                 level = root_item->drop_level;
2684                 path->lowest_level = level;
2685                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2686                 if (wret < 0) {
2687                         ret = wret;
2688                         goto out;
2689                 }
2690                 node = path->nodes[level];
2691                 btrfs_node_key(node, &found_key, path->slots[level]);
2692                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2693                                sizeof(found_key)));
2694                 /*
2695                  * unlock our path, this is safe because only this
2696                  * function is allowed to delete this snapshot
2697                  */
2698                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2699                         if (path->nodes[i] && path->locks[i]) {
2700                                 path->locks[i] = 0;
2701                                 btrfs_tree_unlock(path->nodes[i]);
2702                         }
2703                 }
2704         }
2705         while(1) {
2706                 wret = walk_down_tree(trans, root, path, &level);
2707                 if (wret > 0)
2708                         break;
2709                 if (wret < 0)
2710                         ret = wret;
2711
2712                 wret = walk_up_tree(trans, root, path, &level);
2713                 if (wret > 0)
2714                         break;
2715                 if (wret < 0)
2716                         ret = wret;
2717                 if (trans->transaction->in_commit) {
2718                         ret = -EAGAIN;
2719                         break;
2720                 }
2721                 atomic_inc(&root->fs_info->throttle_gen);
2722                 wake_up(&root->fs_info->transaction_throttle);
2723         }
2724         for (i = 0; i <= orig_level; i++) {
2725                 if (path->nodes[i]) {
2726                         free_extent_buffer(path->nodes[i]);
2727                         path->nodes[i] = NULL;
2728                 }
2729         }
2730 out:
2731         btrfs_free_path(path);
2732         return ret;
2733 }
2734
2735 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2736 {
2737         u64 start;
2738         u64 end;
2739         u64 ptr;
2740         int ret;
2741
2742         mutex_lock(&info->alloc_mutex);
2743         while(1) {
2744                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2745                                             &start, &end, (unsigned int)-1);
2746                 if (ret)
2747                         break;
2748                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2749                 if (!ret)
2750                         kfree((void *)(unsigned long)ptr);
2751                 clear_extent_bits(&info->block_group_cache, start,
2752                                   end, (unsigned int)-1, GFP_NOFS);
2753         }
2754         while(1) {
2755                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2756                                             &start, &end, EXTENT_DIRTY);
2757                 if (ret)
2758                         break;
2759                 clear_extent_dirty(&info->free_space_cache, start,
2760                                    end, GFP_NOFS);
2761         }
2762         mutex_unlock(&info->alloc_mutex);
2763         return 0;
2764 }
2765
2766 static unsigned long calc_ra(unsigned long start, unsigned long last,
2767                              unsigned long nr)
2768 {
2769         return min(last, start + nr - 1);
2770 }
2771
2772 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2773                                          u64 len)
2774 {
2775         u64 page_start;
2776         u64 page_end;
2777         unsigned long last_index;
2778         unsigned long i;
2779         struct page *page;
2780         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2781         struct file_ra_state *ra;
2782         unsigned long total_read = 0;
2783         unsigned long ra_pages;
2784         struct btrfs_ordered_extent *ordered;
2785         struct btrfs_trans_handle *trans;
2786
2787         ra = kzalloc(sizeof(*ra), GFP_NOFS);
2788
2789         mutex_lock(&inode->i_mutex);
2790         i = start >> PAGE_CACHE_SHIFT;
2791         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2792
2793         ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2794
2795         file_ra_state_init(ra, inode->i_mapping);
2796
2797         for (; i <= last_index; i++) {
2798                 if (total_read % ra_pages == 0) {
2799                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2800                                        calc_ra(i, last_index, ra_pages));
2801                 }
2802                 total_read++;
2803 again:
2804                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
2805                         goto truncate_racing;
2806                 page = grab_cache_page(inode->i_mapping, i);
2807                 if (!page) {
2808                         goto out_unlock;
2809                 }
2810                 if (!PageUptodate(page)) {
2811                         btrfs_readpage(NULL, page);
2812                         lock_page(page);
2813                         if (!PageUptodate(page)) {
2814                                 unlock_page(page);
2815                                 page_cache_release(page);
2816                                 goto out_unlock;
2817                         }
2818                 }
2819                 wait_on_page_writeback(page);
2820
2821                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2822                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2823                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2824
2825                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2826                 if (ordered) {
2827                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2828                         unlock_page(page);
2829                         page_cache_release(page);
2830                         btrfs_start_ordered_extent(inode, ordered, 1);
2831                         btrfs_put_ordered_extent(ordered);
2832                         goto again;
2833                 }
2834                 set_page_extent_mapped(page);
2835
2836                 /*
2837                  * make sure page_mkwrite is called for this page if userland
2838                  * wants to change it from mmap
2839                  */
2840                 clear_page_dirty_for_io(page);
2841
2842                 btrfs_set_extent_delalloc(inode, page_start, page_end);
2843                 set_page_dirty(page);
2844
2845                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2846                 unlock_page(page);
2847                 page_cache_release(page);
2848         }
2849
2850 out_unlock:
2851         /* we have to start the IO in order to get the ordered extents
2852          * instantiated.  This allows the relocation to code to wait
2853          * for all the ordered extents to hit the disk.
2854          *
2855          * Otherwise, it would constantly loop over the same extents
2856          * because the old ones don't get deleted  until the IO is
2857          * started
2858          */
2859         btrfs_fdatawrite_range(inode->i_mapping, start, start + len - 1,
2860                                WB_SYNC_NONE);
2861         kfree(ra);
2862         trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2863         if (trans) {
2864                 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2865                 mark_inode_dirty(inode);
2866         }
2867         mutex_unlock(&inode->i_mutex);
2868         return 0;
2869
2870 truncate_racing:
2871         vmtruncate(inode, inode->i_size);
2872         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2873                                            total_read);
2874         goto out_unlock;
2875 }
2876
2877 /*
2878  * The back references tell us which tree holds a ref on a block,
2879  * but it is possible for the tree root field in the reference to
2880  * reflect the original root before a snapshot was made.  In this
2881  * case we should search through all the children of a given root
2882  * to find potential holders of references on a block.
2883  *
2884  * Instead, we do something a little less fancy and just search
2885  * all the roots for a given key/block combination.
2886  */
2887 static int find_root_for_ref(struct btrfs_root *root,
2888                              struct btrfs_path *path,
2889                              struct btrfs_key *key0,
2890                              int level,
2891                              int file_key,
2892                              struct btrfs_root **found_root,
2893                              u64 bytenr)
2894 {
2895         struct btrfs_key root_location;
2896         struct btrfs_root *cur_root = *found_root;
2897         struct btrfs_file_extent_item *file_extent;
2898         u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2899         u64 found_bytenr;
2900         int ret;
2901
2902         root_location.offset = (u64)-1;
2903         root_location.type = BTRFS_ROOT_ITEM_KEY;
2904         path->lowest_level = level;
2905         path->reada = 0;
2906         while(1) {
2907                 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2908                 found_bytenr = 0;
2909                 if (ret == 0 && file_key) {
2910                         struct extent_buffer *leaf = path->nodes[0];
2911                         file_extent = btrfs_item_ptr(leaf, path->slots[0],
2912                                              struct btrfs_file_extent_item);
2913                         if (btrfs_file_extent_type(leaf, file_extent) ==
2914                             BTRFS_FILE_EXTENT_REG) {
2915                                 found_bytenr =
2916                                         btrfs_file_extent_disk_bytenr(leaf,
2917                                                                file_extent);
2918                        }
2919                 } else if (!file_key) {
2920                         if (path->nodes[level])
2921                                 found_bytenr = path->nodes[level]->start;
2922                 }
2923
2924                 btrfs_release_path(cur_root, path);
2925
2926                 if (found_bytenr == bytenr) {
2927                         *found_root = cur_root;
2928                         ret = 0;
2929                         goto out;
2930                 }
2931                 ret = btrfs_search_root(root->fs_info->tree_root,
2932                                         root_search_start, &root_search_start);
2933                 if (ret)
2934                         break;
2935
2936                 root_location.objectid = root_search_start;
2937                 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2938                                                       &root_location);
2939                 if (!cur_root) {
2940                         ret = 1;
2941                         break;
2942                 }
2943         }
2944 out:
2945         path->lowest_level = 0;
2946         return ret;
2947 }
2948
2949 /*
2950  * note, this releases the path
2951  */
2952 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2953                                   struct btrfs_path *path,
2954                                   struct btrfs_key *extent_key,
2955                                   u64 *last_file_objectid,
2956                                   u64 *last_file_offset,
2957                                   u64 *last_file_root,
2958                                   u64 last_extent)
2959 {
2960         struct inode *inode;
2961         struct btrfs_root *found_root;
2962         struct btrfs_key root_location;
2963         struct btrfs_key found_key;
2964         struct btrfs_extent_ref *ref;
2965         u64 ref_root;
2966         u64 ref_gen;
2967         u64 ref_objectid;
2968         u64 ref_offset;
2969         int ret;
2970         int level;
2971
2972         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
2973
2974         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2975                              struct btrfs_extent_ref);
2976         ref_root = btrfs_ref_root(path->nodes[0], ref);
2977         ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2978         ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2979         ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2980         btrfs_release_path(extent_root, path);
2981
2982         root_location.objectid = ref_root;
2983         if (ref_gen == 0)
2984                 root_location.offset = 0;
2985         else
2986                 root_location.offset = (u64)-1;
2987         root_location.type = BTRFS_ROOT_ITEM_KEY;
2988
2989         found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2990                                                 &root_location);
2991         BUG_ON(!found_root);
2992         mutex_unlock(&extent_root->fs_info->alloc_mutex);
2993
2994         if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2995                 found_key.objectid = ref_objectid;
2996                 found_key.type = BTRFS_EXTENT_DATA_KEY;
2997                 found_key.offset = ref_offset;
2998                 level = 0;
2999
3000                 if (last_extent == extent_key->objectid &&
3001                     *last_file_objectid == ref_objectid &&
3002                     *last_file_offset == ref_offset &&
3003                     *last_file_root == ref_root)
3004                         goto out;
3005
3006                 ret = find_root_for_ref(extent_root, path, &found_key,
3007                                         level, 1, &found_root,
3008                                         extent_key->objectid);
3009
3010                 if (ret)
3011                         goto out;
3012
3013                 if (last_extent == extent_key->objectid &&
3014                     *last_file_objectid == ref_objectid &&
3015                     *last_file_offset == ref_offset &&
3016                     *last_file_root == ref_root)
3017                         goto out;
3018
3019                 inode = btrfs_iget_locked(extent_root->fs_info->sb,
3020                                           ref_objectid, found_root);
3021                 if (inode->i_state & I_NEW) {
3022                         /* the inode and parent dir are two different roots */
3023                         BTRFS_I(inode)->root = found_root;
3024                         BTRFS_I(inode)->location.objectid = ref_objectid;
3025                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
3026                         BTRFS_I(inode)->location.offset = 0;
3027                         btrfs_read_locked_inode(inode);
3028                         unlock_new_inode(inode);
3029
3030                 }
3031                 /* this can happen if the reference is not against
3032                  * the latest version of the tree root
3033                  */
3034                 if (is_bad_inode(inode))
3035                         goto out;
3036
3037                 *last_file_objectid = inode->i_ino;
3038                 *last_file_root = found_root->root_key.objectid;
3039                 *last_file_offset = ref_offset;
3040
3041                 relocate_inode_pages(inode, ref_offset, extent_key->offset);
3042                 iput(inode);
3043         } else {
3044                 struct btrfs_trans_handle *trans;
3045                 struct extent_buffer *eb;
3046                 int needs_lock = 0;
3047
3048                 eb = read_tree_block(found_root, extent_key->objectid,
3049                                      extent_key->offset, 0);
3050                 btrfs_tree_lock(eb);
3051                 level = btrfs_header_level(eb);
3052
3053                 if (level == 0)
3054                         btrfs_item_key_to_cpu(eb, &found_key, 0);
3055                 else
3056                         btrfs_node_key_to_cpu(eb, &found_key, 0);
3057
3058                 btrfs_tree_unlock(eb);
3059                 free_extent_buffer(eb);
3060
3061                 ret = find_root_for_ref(extent_root, path, &found_key,
3062                                         level, 0, &found_root,
3063                                         extent_key->objectid);
3064
3065                 if (ret)
3066                         goto out;
3067
3068                 /*
3069                  * right here almost anything could happen to our key,
3070                  * but that's ok.  The cow below will either relocate it
3071                  * or someone else will have relocated it.  Either way,
3072                  * it is in a different spot than it was before and
3073                  * we're happy.
3074                  */
3075
3076                 trans = btrfs_start_transaction(found_root, 1);
3077
3078                 if (found_root == extent_root->fs_info->extent_root ||
3079                     found_root == extent_root->fs_info->chunk_root ||
3080                     found_root == extent_root->fs_info->dev_root) {
3081                         needs_lock = 1;
3082                         mutex_lock(&extent_root->fs_info->alloc_mutex);
3083                 }
3084
3085                 path->lowest_level = level;
3086                 path->reada = 2;
3087                 ret = btrfs_search_slot(trans, found_root, &found_key, path,
3088                                         0, 1);
3089                 path->lowest_level = 0;
3090                 btrfs_release_path(found_root, path);
3091
3092                 if (found_root == found_root->fs_info->extent_root)
3093                         btrfs_extent_post_op(trans, found_root);
3094                 if (needs_lock)
3095                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
3096
3097                 btrfs_end_transaction(trans, found_root);
3098
3099         }
3100 out:
3101         mutex_lock(&extent_root->fs_info->alloc_mutex);
3102         return 0;
3103 }
3104
3105 static int noinline del_extent_zero(struct btrfs_root *extent_root,
3106                                     struct btrfs_path *path,
3107                                     struct btrfs_key *extent_key)
3108 {
3109         int ret;
3110         struct btrfs_trans_handle *trans;
3111
3112         trans = btrfs_start_transaction(extent_root, 1);
3113         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
3114         if (ret > 0) {
3115                 ret = -EIO;
3116                 goto out;
3117         }
3118         if (ret < 0)
3119                 goto out;
3120         ret = btrfs_del_item(trans, extent_root, path);
3121 out:
3122         btrfs_end_transaction(trans, extent_root);
3123         return ret;
3124 }
3125
3126 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
3127                                         struct btrfs_path *path,
3128                                         struct btrfs_key *extent_key)
3129 {
3130         struct btrfs_key key;
3131         struct btrfs_key found_key;
3132         struct extent_buffer *leaf;
3133         u64 last_file_objectid = 0;
3134         u64 last_file_root = 0;
3135         u64 last_file_offset = (u64)-1;
3136         u64 last_extent = 0;
3137         u32 nritems;
3138         u32 item_size;
3139         int ret = 0;
3140
3141         if (extent_key->objectid == 0) {
3142                 ret = del_extent_zero(extent_root, path, extent_key);
3143                 goto out;
3144         }
3145         key.objectid = extent_key->objectid;
3146         key.type = BTRFS_EXTENT_REF_KEY;
3147         key.offset = 0;
3148
3149         while(1) {
3150                 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3151
3152                 if (ret < 0)
3153                         goto out;
3154
3155                 ret = 0;
3156                 leaf = path->nodes[0];
3157                 nritems = btrfs_header_nritems(leaf);
3158                 if (path->slots[0] == nritems) {
3159                         ret = btrfs_next_leaf(extent_root, path);
3160                         if (ret > 0) {
3161                                 ret = 0;
3162                                 goto out;
3163                         }
3164                         if (ret < 0)
3165                                 goto out;
3166                         leaf = path->nodes[0];
3167                 }
3168
3169                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3170                 if (found_key.objectid != extent_key->objectid) {
3171                         break;
3172                 }
3173
3174                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
3175                         break;
3176                 }
3177
3178                 key.offset = found_key.offset + 1;
3179                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3180
3181                 ret = relocate_one_reference(extent_root, path, extent_key,
3182                                              &last_file_objectid,
3183                                              &last_file_offset,
3184                                              &last_file_root, last_extent);
3185                 if (ret)
3186                         goto out;
3187                 last_extent = extent_key->objectid;
3188         }
3189         ret = 0;
3190 out:
3191         btrfs_release_path(extent_root, path);
3192         return ret;
3193 }
3194
3195 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
3196 {
3197         u64 num_devices;
3198         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
3199                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
3200
3201         num_devices = root->fs_info->fs_devices->num_devices;
3202         if (num_devices == 1) {
3203                 stripped |= BTRFS_BLOCK_GROUP_DUP;
3204                 stripped = flags & ~stripped;
3205
3206                 /* turn raid0 into single device chunks */
3207                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
3208                         return stripped;
3209
3210                 /* turn mirroring into duplication */
3211                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3212                              BTRFS_BLOCK_GROUP_RAID10))
3213                         return stripped | BTRFS_BLOCK_GROUP_DUP;
3214                 return flags;
3215         } else {
3216                 /* they already had raid on here, just return */
3217                 if (flags & stripped)
3218                         return flags;
3219
3220                 stripped |= BTRFS_BLOCK_GROUP_DUP;
3221                 stripped = flags & ~stripped;
3222
3223                 /* switch duplicated blocks with raid1 */
3224                 if (flags & BTRFS_BLOCK_GROUP_DUP)
3225                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
3226
3227                 /* turn single device chunks into raid0 */
3228                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
3229         }
3230         return flags;
3231 }
3232
3233 int __alloc_chunk_for_shrink(struct btrfs_root *root,
3234                      struct btrfs_block_group_cache *shrink_block_group,
3235                      int force)
3236 {
3237         struct btrfs_trans_handle *trans;
3238         u64 new_alloc_flags;
3239         u64 calc;
3240
3241         spin_lock(&shrink_block_group->lock);
3242         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
3243                 spin_unlock(&shrink_block_group->lock);
3244                 mutex_unlock(&root->fs_info->alloc_mutex);
3245
3246                 trans = btrfs_start_transaction(root, 1);
3247                 mutex_lock(&root->fs_info->alloc_mutex);
3248                 spin_lock(&shrink_block_group->lock);
3249
3250                 new_alloc_flags = update_block_group_flags(root,
3251                                                    shrink_block_group->flags);
3252                 if (new_alloc_flags != shrink_block_group->flags) {
3253                         calc =
3254                              btrfs_block_group_used(&shrink_block_group->item);
3255                 } else {
3256                         calc = shrink_block_group->key.offset;
3257                 }
3258                 spin_unlock(&shrink_block_group->lock);
3259
3260                 do_chunk_alloc(trans, root->fs_info->extent_root,
3261                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
3262
3263                 mutex_unlock(&root->fs_info->alloc_mutex);
3264                 btrfs_end_transaction(trans, root);
3265                 mutex_lock(&root->fs_info->alloc_mutex);
3266         } else
3267                 spin_unlock(&shrink_block_group->lock);
3268         return 0;
3269 }
3270
3271 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
3272 {
3273         struct btrfs_trans_handle *trans;
3274         struct btrfs_root *tree_root = root->fs_info->tree_root;
3275         struct btrfs_path *path;
3276         u64 cur_byte;
3277         u64 total_found;
3278         u64 shrink_last_byte;
3279         struct btrfs_block_group_cache *shrink_block_group;
3280         struct btrfs_fs_info *info = root->fs_info;
3281         struct btrfs_key key;
3282         struct btrfs_key found_key;
3283         struct extent_buffer *leaf;
3284         u32 nritems;
3285         int ret;
3286         int progress;
3287
3288         mutex_lock(&root->fs_info->alloc_mutex);
3289         shrink_block_group = btrfs_lookup_block_group(root->fs_info,
3290                                                       shrink_start);
3291         BUG_ON(!shrink_block_group);
3292
3293         shrink_last_byte = shrink_block_group->key.objectid +
3294                 shrink_block_group->key.offset;
3295
3296         shrink_block_group->space_info->total_bytes -=
3297                 shrink_block_group->key.offset;
3298         path = btrfs_alloc_path();
3299         root = root->fs_info->extent_root;
3300         path->reada = 2;
3301
3302         printk("btrfs relocating block group %llu flags %llu\n",
3303                (unsigned long long)shrink_start,
3304                (unsigned long long)shrink_block_group->flags);
3305
3306         __alloc_chunk_for_shrink(root, shrink_block_group, 1);
3307
3308 again:
3309
3310         shrink_block_group->ro = 1;
3311
3312         total_found = 0;
3313         progress = 0;
3314         key.objectid = shrink_start;
3315         key.offset = 0;
3316         key.type = 0;
3317         cur_byte = key.objectid;
3318
3319         mutex_unlock(&root->fs_info->alloc_mutex);
3320
3321         btrfs_start_delalloc_inodes(root);
3322         btrfs_wait_ordered_extents(tree_root, 0);
3323
3324         mutex_lock(&root->fs_info->alloc_mutex);
3325
3326         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3327         if (ret < 0)
3328                 goto out;
3329
3330         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
3331         if (ret < 0)
3332                 goto out;
3333
3334         if (ret == 0) {
3335                 leaf = path->nodes[0];
3336                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3337                 if (found_key.objectid + found_key.offset > shrink_start &&
3338                     found_key.objectid < shrink_last_byte) {
3339                         cur_byte = found_key.objectid;
3340                         key.objectid = cur_byte;
3341                 }
3342         }
3343         btrfs_release_path(root, path);
3344
3345         while(1) {
3346                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3347                 if (ret < 0)
3348                         goto out;
3349
3350 next:
3351                 leaf = path->nodes[0];
3352                 nritems = btrfs_header_nritems(leaf);
3353                 if (path->slots[0] >= nritems) {
3354                         ret = btrfs_next_leaf(root, path);
3355                         if (ret < 0)
3356                                 goto out;
3357                         if (ret == 1) {
3358                                 ret = 0;
3359                                 break;
3360                         }
3361                         leaf = path->nodes[0];
3362                         nritems = btrfs_header_nritems(leaf);
3363                 }
3364
3365                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3366
3367                 if (found_key.objectid >= shrink_last_byte)
3368                         break;
3369
3370                 if (progress && need_resched()) {
3371                         memcpy(&key, &found_key, sizeof(key));
3372                         cond_resched();
3373                         btrfs_release_path(root, path);
3374                         btrfs_search_slot(NULL, root, &key, path, 0, 0);
3375                         progress = 0;
3376                         goto next;
3377                 }
3378                 progress = 1;
3379
3380                 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3381                     found_key.objectid + found_key.offset <= cur_byte) {
3382                         memcpy(&key, &found_key, sizeof(key));
3383                         key.offset++;
3384                         path->slots[0]++;
3385                         goto next;
3386                 }
3387
3388                 total_found++;
3389                 cur_byte = found_key.objectid + found_key.offset;
3390                 key.objectid = cur_byte;
3391                 btrfs_release_path(root, path);
3392                 ret = relocate_one_extent(root, path, &found_key);
3393                 __alloc_chunk_for_shrink(root, shrink_block_group, 0);
3394         }
3395
3396         btrfs_release_path(root, path);
3397
3398         if (total_found > 0) {
3399                 printk("btrfs relocate found %llu last extent was %llu\n",
3400                        (unsigned long long)total_found,
3401                        (unsigned long long)found_key.objectid);
3402                 mutex_unlock(&root->fs_info->alloc_mutex);
3403                 trans = btrfs_start_transaction(tree_root, 1);
3404                 btrfs_commit_transaction(trans, tree_root);
3405
3406                 btrfs_clean_old_snapshots(tree_root);
3407
3408                 btrfs_start_delalloc_inodes(root);
3409                 btrfs_wait_ordered_extents(tree_root, 0);
3410
3411                 trans = btrfs_start_transaction(tree_root, 1);
3412                 btrfs_commit_transaction(trans, tree_root);
3413                 mutex_lock(&root->fs_info->alloc_mutex);
3414                 goto again;
3415         }
3416
3417         /*
3418          * we've freed all the extents, now remove the block
3419          * group item from the tree
3420          */
3421         mutex_unlock(&root->fs_info->alloc_mutex);
3422
3423         trans = btrfs_start_transaction(root, 1);
3424
3425         mutex_lock(&root->fs_info->alloc_mutex);
3426         memcpy(&key, &shrink_block_group->key, sizeof(key));
3427
3428         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3429         if (ret > 0)
3430                 ret = -EIO;
3431         if (ret < 0) {
3432                 btrfs_end_transaction(trans, root);
3433                 goto out;
3434         }
3435
3436         clear_extent_bits(&info->block_group_cache, key.objectid,
3437                           key.objectid + key.offset - 1,
3438                           (unsigned int)-1, GFP_NOFS);
3439
3440
3441         clear_extent_bits(&info->free_space_cache,
3442                            key.objectid, key.objectid + key.offset - 1,
3443                            (unsigned int)-1, GFP_NOFS);
3444
3445         /*
3446         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3447         kfree(shrink_block_group);
3448         */
3449
3450         btrfs_del_item(trans, root, path);
3451         btrfs_release_path(root, path);
3452         mutex_unlock(&root->fs_info->alloc_mutex);
3453         btrfs_commit_transaction(trans, root);
3454
3455         mutex_lock(&root->fs_info->alloc_mutex);
3456
3457         /* the code to unpin extents might set a few bits in the free
3458          * space cache for this range again
3459          */
3460         clear_extent_bits(&info->free_space_cache,
3461                            key.objectid, key.objectid + key.offset - 1,
3462                            (unsigned int)-1, GFP_NOFS);
3463 out:
3464         btrfs_free_path(path);
3465         mutex_unlock(&root->fs_info->alloc_mutex);
3466         return ret;
3467 }
3468
3469 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3470                            struct btrfs_key *key)
3471 {
3472         int ret = 0;
3473         struct btrfs_key found_key;
3474         struct extent_buffer *leaf;
3475         int slot;
3476
3477         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3478         if (ret < 0)
3479                 goto out;
3480
3481         while(1) {
3482                 slot = path->slots[0];
3483                 leaf = path->nodes[0];
3484                 if (slot >= btrfs_header_nritems(leaf)) {
3485                         ret = btrfs_next_leaf(root, path);
3486                         if (ret == 0)
3487                                 continue;
3488                         if (ret < 0)
3489                                 goto out;
3490                         break;
3491                 }
3492                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3493
3494                 if (found_key.objectid >= key->objectid &&
3495                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3496                         ret = 0;
3497                         goto out;
3498                 }
3499                 path->slots[0]++;
3500         }
3501         ret = -ENOENT;
3502 out:
3503         return ret;
3504 }
3505
3506 int btrfs_read_block_groups(struct btrfs_root *root)
3507 {
3508         struct btrfs_path *path;
3509         int ret;
3510         int bit;
3511         struct btrfs_block_group_cache *cache;
3512         struct btrfs_fs_info *info = root->fs_info;
3513         struct btrfs_space_info *space_info;
3514         struct extent_io_tree *block_group_cache;
3515         struct btrfs_key key;
3516         struct btrfs_key found_key;
3517         struct extent_buffer *leaf;
3518
3519         block_group_cache = &info->block_group_cache;
3520         root = info->extent_root;
3521         key.objectid = 0;
3522         key.offset = 0;
3523         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3524         path = btrfs_alloc_path();
3525         if (!path)
3526                 return -ENOMEM;
3527
3528         mutex_lock(&root->fs_info->alloc_mutex);
3529         while(1) {
3530                 ret = find_first_block_group(root, path, &key);
3531                 if (ret > 0) {
3532                         ret = 0;
3533                         goto error;
3534                 }
3535                 if (ret != 0)
3536                         goto error;
3537
3538                 leaf = path->nodes[0];
3539                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3540                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3541                 if (!cache) {
3542                         ret = -ENOMEM;
3543                         break;
3544                 }
3545
3546                 spin_lock_init(&cache->lock);
3547                 read_extent_buffer(leaf, &cache->item,
3548                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3549                                    sizeof(cache->item));
3550                 memcpy(&cache->key, &found_key, sizeof(found_key));
3551
3552                 key.objectid = found_key.objectid + found_key.offset;
3553                 btrfs_release_path(root, path);
3554                 cache->flags = btrfs_block_group_flags(&cache->item);
3555                 bit = 0;
3556                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3557                         bit = BLOCK_GROUP_DATA;
3558                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3559                         bit = BLOCK_GROUP_SYSTEM;
3560                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3561                         bit = BLOCK_GROUP_METADATA;
3562                 }
3563                 set_avail_alloc_bits(info, cache->flags);
3564
3565                 ret = update_space_info(info, cache->flags, found_key.offset,
3566                                         btrfs_block_group_used(&cache->item),
3567                                         &space_info);
3568                 BUG_ON(ret);
3569                 cache->space_info = space_info;
3570
3571                 /* use EXTENT_LOCKED to prevent merging */
3572                 set_extent_bits(block_group_cache, found_key.objectid,
3573                                 found_key.objectid + found_key.offset - 1,
3574                                 EXTENT_LOCKED, GFP_NOFS);
3575                 set_state_private(block_group_cache, found_key.objectid,
3576                                   (unsigned long)cache);
3577                 set_extent_bits(block_group_cache, found_key.objectid,
3578                                 found_key.objectid + found_key.offset - 1,
3579                                 bit | EXTENT_LOCKED, GFP_NOFS);
3580                 if (key.objectid >=
3581                     btrfs_super_total_bytes(&info->super_copy))
3582                         break;
3583         }
3584         ret = 0;
3585 error:
3586         btrfs_free_path(path);
3587         mutex_unlock(&root->fs_info->alloc_mutex);
3588         return ret;
3589 }
3590
3591 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3592                            struct btrfs_root *root, u64 bytes_used,
3593                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3594                            u64 size)
3595 {
3596         int ret;
3597         int bit = 0;
3598         struct btrfs_root *extent_root;
3599         struct btrfs_block_group_cache *cache;
3600         struct extent_io_tree *block_group_cache;
3601
3602         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
3603         extent_root = root->fs_info->extent_root;
3604         block_group_cache = &root->fs_info->block_group_cache;
3605
3606         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3607         BUG_ON(!cache);
3608         cache->key.objectid = chunk_offset;
3609         cache->key.offset = size;
3610         spin_lock_init(&cache->lock);
3611         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3612
3613         btrfs_set_block_group_used(&cache->item, bytes_used);
3614         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3615         cache->flags = type;
3616         btrfs_set_block_group_flags(&cache->item, type);
3617
3618         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3619                                 &cache->space_info);
3620         BUG_ON(ret);
3621
3622         bit = block_group_state_bits(type);
3623         set_extent_bits(block_group_cache, chunk_offset,
3624                         chunk_offset + size - 1,
3625                         EXTENT_LOCKED, GFP_NOFS);
3626         set_state_private(block_group_cache, chunk_offset,
3627                           (unsigned long)cache);
3628         set_extent_bits(block_group_cache, chunk_offset,
3629                         chunk_offset + size - 1,
3630                         bit | EXTENT_LOCKED, GFP_NOFS);
3631
3632         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3633                                 sizeof(cache->item));
3634         BUG_ON(ret);
3635
3636         finish_current_insert(trans, extent_root);
3637         ret = del_pending_extents(trans, extent_root);
3638         BUG_ON(ret);
3639         set_avail_alloc_bits(extent_root->fs_info, type);
3640
3641         return 0;
3642 }