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_root *root,
 897                            struct btrfs_key *key, u64 bytenr)
 898 {
 899         struct btrfs_trans_handle *trans;
 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(key->type != BTRFS_EXTENT_DATA_KEY);
 912         ret = get_reference_status(root, bytenr, 0, key->objectid,
 913                                    &min_generation, &ref_count);
 914         if (ret)
 915                 return ret;
 916
 917         if (ref_count != 1)
 918                 return 1;
 919
 920         trans = btrfs_start_transaction(root, 0);
 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         btrfs_end_transaction(trans, root);
 977         return ret;
 978 }
 979
 980 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 981                   struct extent_buffer *buf, int cache_ref)
 982 {
 983         u64 bytenr;
 984         u32 nritems;
 985         struct btrfs_key key;
 986         struct btrfs_file_extent_item *fi;
 987         int i;
 988         int level;
 989         int ret;
 990         int faili;
 991         int nr_file_extents = 0;
 992
 993         if (!root->ref_cows)
 994                 return 0;
 995
 996         level = btrfs_header_level(buf);
 997         nritems = btrfs_header_nritems(buf);
 998         for (i = 0; i < nritems; i++) {
 999                 cond_resched();
1000                 if (level == 0) {
1001                         u64 disk_bytenr;
1002                         btrfs_item_key_to_cpu(buf, &key, i);
1003                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1004                                 continue;
1005                         fi = btrfs_item_ptr(buf, i,
1006                                             struct btrfs_file_extent_item);
1007                         if (btrfs_file_extent_type(buf, fi) ==
1008                             BTRFS_FILE_EXTENT_INLINE)
1009                                 continue;
1010                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1011                         if (disk_bytenr == 0)
1012                                 continue;
1013
1014                         if (buf != root->commit_root)
1015                                 nr_file_extents++;
1016
1017                         mutex_lock(&root->fs_info->alloc_mutex);
1018                         ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr,
1019                                     btrfs_file_extent_disk_num_bytes(buf, fi),
1020                                     root->root_key.objectid, trans->transid,
1021                                     key.objectid, key.offset);
1022                         mutex_unlock(&root->fs_info->alloc_mutex);
1023                         if (ret) {
1024                                 faili = i;
1025                                 WARN_ON(1);
1026                                 goto fail;
1027                         }
1028                 } else {
1029                         bytenr = btrfs_node_blockptr(buf, i);
1030                         btrfs_node_key_to_cpu(buf, &key, i);
1031
1032                         mutex_lock(&root->fs_info->alloc_mutex);
1033                         ret = __btrfs_inc_extent_ref(trans, root, bytenr,
1034                                            btrfs_level_size(root, level - 1),
1035                                            root->root_key.objectid,
1036                                            trans->transid,
1037                                            level - 1, key.objectid);
1038                         mutex_unlock(&root->fs_info->alloc_mutex);
1039                         if (ret) {
1040                                 faili = i;
1041                                 WARN_ON(1);
1042                                 goto fail;
1043                         }
1044                 }
1045         }
1046         /* cache orignal leaf block's references */
1047         if (level == 0 && cache_ref && buf != root->commit_root) {
1048                 struct btrfs_leaf_ref *ref;
1049                 struct btrfs_extent_info *info;
1050
1051                 ref = btrfs_alloc_leaf_ref(root, nr_file_extents);
1052                 if (!ref) {
1053                         WARN_ON(1);
1054                         goto out;
1055                 }
1056
1057                 ref->root_gen = root->root_key.offset;
1058                 ref->bytenr = buf->start;
1059                 ref->owner = btrfs_header_owner(buf);
1060                 ref->generation = btrfs_header_generation(buf);
1061                 ref->nritems = nr_file_extents;
1062                 info = ref->extents;
1063
1064                 for (i = 0; nr_file_extents > 0 && i < nritems; i++) {
1065                         u64 disk_bytenr;
1066                         btrfs_item_key_to_cpu(buf, &key, i);
1067                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1068                                 continue;
1069                         fi = btrfs_item_ptr(buf, i,
1070                                             struct btrfs_file_extent_item);
1071                         if (btrfs_file_extent_type(buf, fi) ==
1072                             BTRFS_FILE_EXTENT_INLINE)
1073                                 continue;
1074                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1075                         if (disk_bytenr == 0)
1076                                 continue;
1077
1078                         info->bytenr = disk_bytenr;
1079                         info->num_bytes =
1080                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1081                         info->objectid = key.objectid;
1082                         info->offset = key.offset;
1083                         info++;
1084                 }
1085
1086                 BUG_ON(!root->ref_tree);
1087                 ret = btrfs_add_leaf_ref(root, ref);
1088                 WARN_ON(ret);
1089                 btrfs_free_leaf_ref(root, ref);
1090         }
1091 out:
1092         return 0;
1093 fail:
1094         WARN_ON(1);
1095 #if 0
1096         for (i =0; i < faili; i++) {
1097                 if (level == 0) {
1098                         u64 disk_bytenr;
1099                         btrfs_item_key_to_cpu(buf, &key, i);
1100                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1101                                 continue;
1102                         fi = btrfs_item_ptr(buf, i,
1103                                             struct btrfs_file_extent_item);
1104                         if (btrfs_file_extent_type(buf, fi) ==
1105                             BTRFS_FILE_EXTENT_INLINE)
1106                                 continue;
1107                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1108                         if (disk_bytenr == 0)
1109                                 continue;
1110                         err = btrfs_free_extent(trans, root, disk_bytenr,
1111                                     btrfs_file_extent_disk_num_bytes(buf,
1112                                                                       fi), 0);
1113                         BUG_ON(err);
1114                 } else {
1115                         bytenr = btrfs_node_blockptr(buf, i);
1116                         err = btrfs_free_extent(trans, root, bytenr,
1117                                         btrfs_level_size(root, level - 1), 0);
1118                         BUG_ON(err);
1119                 }
1120         }
1121 #endif
1122         return ret;
1123 }
1124
1125 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1126                                  struct btrfs_root *root,
1127                                  struct btrfs_path *path,
1128                                  struct btrfs_block_group_cache *cache)
1129 {
1130         int ret;
1131         int pending_ret;
1132         struct btrfs_root *extent_root = root->fs_info->extent_root;
1133         unsigned long bi;
1134         struct extent_buffer *leaf;
1135
1136         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1137         if (ret < 0)
1138                 goto fail;
1139         BUG_ON(ret);
1140
1141         leaf = path->nodes[0];
1142         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1143         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1144         btrfs_mark_buffer_dirty(leaf);
1145         btrfs_release_path(extent_root, path);
1146 fail:
1147         finish_current_insert(trans, extent_root);
1148         pending_ret = del_pending_extents(trans, extent_root);
1149         if (ret)
1150                 return ret;
1151         if (pending_ret)
1152                 return pending_ret;
1153         return 0;
1154
1155 }
1156
1157 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1158                                    struct btrfs_root *root)
1159 {
1160         struct extent_io_tree *block_group_cache;
1161         struct btrfs_block_group_cache *cache;
1162         int ret;
1163         int err = 0;
1164         int werr = 0;
1165         struct btrfs_path *path;
1166         u64 last = 0;
1167         u64 start;
1168         u64 end;
1169         u64 ptr;
1170
1171         block_group_cache = &root->fs_info->block_group_cache;
1172         path = btrfs_alloc_path();
1173         if (!path)
1174                 return -ENOMEM;
1175
1176         mutex_lock(&root->fs_info->alloc_mutex);
1177         while(1) {
1178                 ret = find_first_extent_bit(block_group_cache, last,
1179                                             &start, &end, BLOCK_GROUP_DIRTY);
1180                 if (ret)
1181                         break;
1182
1183                 last = end + 1;
1184                 ret = get_state_private(block_group_cache, start, &ptr);
1185                 if (ret)
1186                         break;
1187                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1188                 err = write_one_cache_group(trans, root,
1189                                             path, cache);
1190                 /*
1191                  * if we fail to write the cache group, we want
1192                  * to keep it marked dirty in hopes that a later
1193                  * write will work
1194                  */
1195                 if (err) {
1196                         werr = err;
1197                         continue;
1198                 }
1199                 clear_extent_bits(block_group_cache, start, end,
1200                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
1201         }
1202         btrfs_free_path(path);
1203         mutex_unlock(&root->fs_info->alloc_mutex);
1204         return werr;
1205 }
1206
1207 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1208                                                   u64 flags)
1209 {
1210         struct list_head *head = &info->space_info;
1211         struct list_head *cur;
1212         struct btrfs_space_info *found;
1213         list_for_each(cur, head) {
1214                 found = list_entry(cur, struct btrfs_space_info, list);
1215                 if (found->flags == flags)
1216                         return found;
1217         }
1218         return NULL;
1219
1220 }
1221
1222 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1223                              u64 total_bytes, u64 bytes_used,
1224                              struct btrfs_space_info **space_info)
1225 {
1226         struct btrfs_space_info *found;
1227
1228         found = __find_space_info(info, flags);
1229         if (found) {
1230                 found->total_bytes += total_bytes;
1231                 found->bytes_used += bytes_used;
1232                 found->full = 0;
1233                 *space_info = found;
1234                 return 0;
1235         }
1236         found = kmalloc(sizeof(*found), GFP_NOFS);
1237         if (!found)
1238                 return -ENOMEM;
1239
1240         list_add(&found->list, &info->space_info);
1241         found->flags = flags;
1242         found->total_bytes = total_bytes;
1243         found->bytes_used = bytes_used;
1244         found->bytes_pinned = 0;
1245         found->full = 0;
1246         found->force_alloc = 0;
1247         *space_info = found;
1248         return 0;
1249 }
1250
1251 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1252 {
1253         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1254                                    BTRFS_BLOCK_GROUP_RAID1 |
1255                                    BTRFS_BLOCK_GROUP_RAID10 |
1256                                    BTRFS_BLOCK_GROUP_DUP);
1257         if (extra_flags) {
1258                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1259                         fs_info->avail_data_alloc_bits |= extra_flags;
1260                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1261                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1262                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1263                         fs_info->avail_system_alloc_bits |= extra_flags;
1264         }
1265 }
1266
1267 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1268 {
1269         u64 num_devices = root->fs_info->fs_devices->num_devices;
1270
1271         if (num_devices == 1)
1272                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1273         if (num_devices < 4)
1274                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1275
1276         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1277             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1278                       BTRFS_BLOCK_GROUP_RAID10))) {
1279                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1280         }
1281
1282         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1283             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1284                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1285         }
1286
1287         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1288             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1289              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1290              (flags & BTRFS_BLOCK_GROUP_DUP)))
1291                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1292         return flags;
1293 }
1294
1295 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1296                           struct btrfs_root *extent_root, u64 alloc_bytes,
1297                           u64 flags, int force)
1298 {
1299         struct btrfs_space_info *space_info;
1300         u64 thresh;
1301         u64 start;
1302         u64 num_bytes;
1303         int ret;
1304
1305         flags = reduce_alloc_profile(extent_root, flags);
1306
1307         space_info = __find_space_info(extent_root->fs_info, flags);
1308         if (!space_info) {
1309                 ret = update_space_info(extent_root->fs_info, flags,
1310                                         0, 0, &space_info);
1311                 BUG_ON(ret);
1312         }
1313         BUG_ON(!space_info);
1314
1315         if (space_info->force_alloc) {
1316                 force = 1;
1317                 space_info->force_alloc = 0;
1318         }
1319         if (space_info->full)
1320                 goto out;
1321
1322         thresh = div_factor(space_info->total_bytes, 6);
1323         if (!force &&
1324            (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1325             thresh)
1326                 goto out;
1327
1328         mutex_lock(&extent_root->fs_info->chunk_mutex);
1329         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1330         if (ret == -ENOSPC) {
1331 printk("space info full %Lu\n", flags);
1332                 space_info->full = 1;
1333                 goto out_unlock;
1334         }
1335         BUG_ON(ret);
1336
1337         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1338                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1339         BUG_ON(ret);
1340 out_unlock:
1341         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1342 out:
1343         return 0;
1344 }
1345
1346 static int update_block_group(struct btrfs_trans_handle *trans,
1347                               struct btrfs_root *root,
1348                               u64 bytenr, u64 num_bytes, int alloc,
1349                               int mark_free)
1350 {
1351         struct btrfs_block_group_cache *cache;
1352         struct btrfs_fs_info *info = root->fs_info;
1353         u64 total = num_bytes;
1354         u64 old_val;
1355         u64 byte_in_group;
1356         u64 start;
1357         u64 end;
1358
1359         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1360         while(total) {
1361                 cache = btrfs_lookup_block_group(info, bytenr);
1362                 if (!cache) {
1363                         return -1;
1364                 }
1365                 byte_in_group = bytenr - cache->key.objectid;
1366                 WARN_ON(byte_in_group > cache->key.offset);
1367                 start = cache->key.objectid;
1368                 end = start + cache->key.offset - 1;
1369                 set_extent_bits(&info->block_group_cache, start, end,
1370                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1371
1372                 spin_lock(&cache->lock);
1373                 old_val = btrfs_block_group_used(&cache->item);
1374                 num_bytes = min(total, cache->key.offset - byte_in_group);
1375                 if (alloc) {
1376                         old_val += num_bytes;
1377                         cache->space_info->bytes_used += num_bytes;
1378                         btrfs_set_block_group_used(&cache->item, old_val);
1379                         spin_unlock(&cache->lock);
1380                 } else {
1381                         old_val -= num_bytes;
1382                         cache->space_info->bytes_used -= num_bytes;
1383                         btrfs_set_block_group_used(&cache->item, old_val);
1384                         spin_unlock(&cache->lock);
1385                         if (mark_free) {
1386                                 set_extent_dirty(&info->free_space_cache,
1387                                                  bytenr, bytenr + num_bytes - 1,
1388                                                  GFP_NOFS);
1389                         }
1390                 }
1391                 total -= num_bytes;
1392                 bytenr += num_bytes;
1393         }
1394         return 0;
1395 }
1396
1397 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1398 {
1399         u64 start;
1400         u64 end;
1401         int ret;
1402         ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1403                                     search_start, &start, &end,
1404                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1405                                     BLOCK_GROUP_SYSTEM);
1406         if (ret)
1407                 return 0;
1408         return start;
1409 }
1410
1411
1412 static int update_pinned_extents(struct btrfs_root *root,
1413                                 u64 bytenr, u64 num, int pin)
1414 {
1415         u64 len;
1416         struct btrfs_block_group_cache *cache;
1417         struct btrfs_fs_info *fs_info = root->fs_info;
1418
1419         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1420         if (pin) {
1421                 set_extent_dirty(&fs_info->pinned_extents,
1422                                 bytenr, bytenr + num - 1, GFP_NOFS);
1423         } else {
1424                 clear_extent_dirty(&fs_info->pinned_extents,
1425                                 bytenr, bytenr + num - 1, GFP_NOFS);
1426         }
1427         while (num > 0) {
1428                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1429                 if (!cache) {
1430                         u64 first = first_logical_byte(root, bytenr);
1431                         WARN_ON(first < bytenr);
1432                         len = min(first - bytenr, num);
1433                 } else {
1434                         len = min(num, cache->key.offset -
1435                                   (bytenr - cache->key.objectid));
1436                 }
1437                 if (pin) {
1438                         if (cache) {
1439                                 spin_lock(&cache->lock);
1440                                 cache->pinned += len;
1441                                 cache->space_info->bytes_pinned += len;
1442                                 spin_unlock(&cache->lock);
1443                         }
1444                         fs_info->total_pinned += len;
1445                 } else {
1446                         if (cache) {
1447                                 spin_lock(&cache->lock);
1448                                 cache->pinned -= len;
1449                                 cache->space_info->bytes_pinned -= len;
1450                                 spin_unlock(&cache->lock);
1451                         }
1452                         fs_info->total_pinned -= len;
1453                 }
1454                 bytenr += len;
1455                 num -= len;
1456         }
1457         return 0;
1458 }
1459
1460 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1461 {
1462         u64 last = 0;
1463         u64 start;
1464         u64 end;
1465         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1466         int ret;
1467
1468         while(1) {
1469                 ret = find_first_extent_bit(pinned_extents, last,
1470                                             &start, &end, EXTENT_DIRTY);
1471                 if (ret)
1472                         break;
1473                 set_extent_dirty(copy, start, end, GFP_NOFS);
1474                 last = end + 1;
1475         }
1476         return 0;
1477 }
1478
1479 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1480                                struct btrfs_root *root,
1481                                struct extent_io_tree *unpin)
1482 {
1483         u64 start;
1484         u64 end;
1485         int ret;
1486         struct extent_io_tree *free_space_cache;
1487         free_space_cache = &root->fs_info->free_space_cache;
1488
1489         mutex_lock(&root->fs_info->alloc_mutex);
1490         while(1) {
1491                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1492                                             EXTENT_DIRTY);
1493                 if (ret)
1494                         break;
1495                 update_pinned_extents(root, start, end + 1 - start, 0);
1496                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1497                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1498                 if (need_resched()) {
1499                         mutex_unlock(&root->fs_info->alloc_mutex);
1500                         cond_resched();
1501                         mutex_lock(&root->fs_info->alloc_mutex);
1502                 }
1503         }
1504         mutex_unlock(&root->fs_info->alloc_mutex);
1505         return 0;
1506 }
1507
1508 static int finish_current_insert(struct btrfs_trans_handle *trans,
1509                                  struct btrfs_root *extent_root)
1510 {
1511         u64 start;
1512         u64 end;
1513         struct btrfs_fs_info *info = extent_root->fs_info;
1514         struct extent_buffer *eb;
1515         struct btrfs_path *path;
1516         struct btrfs_key ins;
1517         struct btrfs_disk_key first;
1518         struct btrfs_extent_item extent_item;
1519         int ret;
1520         int level;
1521         int err = 0;
1522
1523         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1524         btrfs_set_stack_extent_refs(&extent_item, 1);
1525         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1526         path = btrfs_alloc_path();
1527
1528         while(1) {
1529                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1530                                             &end, EXTENT_LOCKED);
1531                 if (ret)
1532                         break;
1533
1534                 ins.objectid = start;
1535                 ins.offset = end + 1 - start;
1536                 err = btrfs_insert_item(trans, extent_root, &ins,
1537                                         &extent_item, sizeof(extent_item));
1538                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1539                                   GFP_NOFS);
1540
1541                 eb = btrfs_find_tree_block(extent_root, ins.objectid,
1542                                            ins.offset);
1543
1544                 if (!btrfs_buffer_uptodate(eb, trans->transid)) {
1545                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1546                         btrfs_read_buffer(eb, trans->transid);
1547                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1548                 }
1549
1550                 btrfs_tree_lock(eb);
1551                 level = btrfs_header_level(eb);
1552                 if (level == 0) {
1553                         btrfs_item_key(eb, &first, 0);
1554                 } else {
1555                         btrfs_node_key(eb, &first, 0);
1556                 }
1557                 btrfs_tree_unlock(eb);
1558                 free_extent_buffer(eb);
1559                 /*
1560                  * the first key is just a hint, so the race we've created
1561                  * against reading it is fine
1562                  */
1563                 err = btrfs_insert_extent_backref(trans, extent_root, path,
1564                                           start, extent_root->root_key.objectid,
1565                                           0, level,
1566                                           btrfs_disk_key_objectid(&first));
1567                 BUG_ON(err);
1568                 if (need_resched()) {
1569                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1570                         cond_resched();
1571                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1572                 }
1573         }
1574         btrfs_free_path(path);
1575         return 0;
1576 }
1577
1578 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1579                           int pending)
1580 {
1581         int err = 0;
1582
1583         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1584         if (!pending) {
1585                 struct extent_buffer *buf;
1586                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1587                 if (buf) {
1588                         if (btrfs_buffer_uptodate(buf, 0) &&
1589                             btrfs_try_tree_lock(buf)) {
1590                                 u64 transid =
1591                                     root->fs_info->running_transaction->transid;
1592                                 u64 header_transid =
1593                                         btrfs_header_generation(buf);
1594                                 if (header_transid == transid &&
1595                                     !btrfs_header_flag(buf,
1596                                                BTRFS_HEADER_FLAG_WRITTEN)) {
1597                                         clean_tree_block(NULL, root, buf);
1598                                         btrfs_tree_unlock(buf);
1599                                         free_extent_buffer(buf);
1600                                         return 1;
1601                                 }
1602                                 btrfs_tree_unlock(buf);
1603                         }
1604                         free_extent_buffer(buf);
1605                 }
1606                 update_pinned_extents(root, bytenr, num_bytes, 1);
1607         } else {
1608                 set_extent_bits(&root->fs_info->pending_del,
1609                                 bytenr, bytenr + num_bytes - 1,
1610                                 EXTENT_LOCKED, GFP_NOFS);
1611         }
1612         BUG_ON(err < 0);
1613         return 0;
1614 }
1615
1616 /*
1617  * remove an extent from the root, returns 0 on success
1618  */
1619 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1620                          *root, u64 bytenr, u64 num_bytes,
1621                          u64 root_objectid, u64 ref_generation,
1622                          u64 owner_objectid, u64 owner_offset, int pin,
1623                          int mark_free)
1624 {
1625         struct btrfs_path *path;
1626         struct btrfs_key key;
1627         struct btrfs_fs_info *info = root->fs_info;
1628         struct btrfs_root *extent_root = info->extent_root;
1629         struct extent_buffer *leaf;
1630         int ret;
1631         int extent_slot = 0;
1632         int found_extent = 0;
1633         int num_to_del = 1;
1634         struct btrfs_extent_item *ei;
1635         u32 refs;
1636
1637         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1638         key.objectid = bytenr;
1639         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1640         key.offset = num_bytes;
1641         path = btrfs_alloc_path();
1642         if (!path)
1643                 return -ENOMEM;
1644
1645         path->reada = 1;
1646         ret = lookup_extent_backref(trans, extent_root, path,
1647                                     bytenr, root_objectid,
1648                                     ref_generation,
1649                                     owner_objectid, owner_offset, 1);
1650         if (ret == 0) {
1651                 struct btrfs_key found_key;
1652                 extent_slot = path->slots[0];
1653                 while(extent_slot > 0) {
1654                         extent_slot--;
1655                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1656                                               extent_slot);
1657                         if (found_key.objectid != bytenr)
1658                                 break;
1659                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1660                             found_key.offset == num_bytes) {
1661                                 found_extent = 1;
1662                                 break;
1663                         }
1664                         if (path->slots[0] - extent_slot > 5)
1665                                 break;
1666                 }
1667                 if (!found_extent)
1668                         ret = btrfs_del_item(trans, extent_root, path);
1669         } else {
1670                 btrfs_print_leaf(extent_root, path->nodes[0]);
1671                 WARN_ON(1);
1672                 printk("Unable to find ref byte nr %Lu root %Lu "
1673                        " gen %Lu owner %Lu offset %Lu\n", bytenr,
1674                        root_objectid, ref_generation, owner_objectid,
1675                        owner_offset);
1676         }
1677         if (!found_extent) {
1678                 btrfs_release_path(extent_root, path);
1679                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1680                 if (ret < 0)
1681                         return ret;
1682                 BUG_ON(ret);
1683                 extent_slot = path->slots[0];
1684         }
1685
1686         leaf = path->nodes[0];
1687         ei = btrfs_item_ptr(leaf, extent_slot,
1688                             struct btrfs_extent_item);
1689         refs = btrfs_extent_refs(leaf, ei);
1690         BUG_ON(refs == 0);
1691         refs -= 1;
1692         btrfs_set_extent_refs(leaf, ei, refs);
1693
1694         btrfs_mark_buffer_dirty(leaf);
1695
1696         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1697                 /* if the back ref and the extent are next to each other
1698                  * they get deleted below in one shot
1699                  */
1700                 path->slots[0] = extent_slot;
1701                 num_to_del = 2;
1702         } else if (found_extent) {
1703                 /* otherwise delete the extent back ref */
1704                 ret = btrfs_del_item(trans, extent_root, path);
1705                 BUG_ON(ret);
1706                 /* if refs are 0, we need to setup the path for deletion */
1707                 if (refs == 0) {
1708                         btrfs_release_path(extent_root, path);
1709                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1710                                                 -1, 1);
1711                         if (ret < 0)
1712                                 return ret;
1713                         BUG_ON(ret);
1714                 }
1715         }
1716
1717         if (refs == 0) {
1718                 u64 super_used;
1719                 u64 root_used;
1720
1721                 if (pin) {
1722                         ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1723                         if (ret > 0)
1724                                 mark_free = 1;
1725                         BUG_ON(ret < 0);
1726                 }
1727
1728                 /* block accounting for super block */
1729                 spin_lock_irq(&info->delalloc_lock);
1730                 super_used = btrfs_super_bytes_used(&info->super_copy);
1731                 btrfs_set_super_bytes_used(&info->super_copy,
1732                                            super_used - num_bytes);
1733                 spin_unlock_irq(&info->delalloc_lock);
1734
1735                 /* block accounting for root item */
1736                 root_used = btrfs_root_used(&root->root_item);
1737                 btrfs_set_root_used(&root->root_item,
1738                                            root_used - num_bytes);
1739                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1740                                       num_to_del);
1741                 if (ret) {
1742                         return ret;
1743                 }
1744                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1745                                          mark_free);
1746                 BUG_ON(ret);
1747         }
1748         btrfs_free_path(path);
1749         finish_current_insert(trans, extent_root);
1750         return ret;
1751 }
1752
1753 /*
1754  * find all the blocks marked as pending in the radix tree and remove
1755  * them from the extent map
1756  */
1757 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1758                                btrfs_root *extent_root)
1759 {
1760         int ret;
1761         int err = 0;
1762         u64 start;
1763         u64 end;
1764         struct extent_io_tree *pending_del;
1765         struct extent_io_tree *pinned_extents;
1766
1767         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1768         pending_del = &extent_root->fs_info->pending_del;
1769         pinned_extents = &extent_root->fs_info->pinned_extents;
1770
1771         while(1) {
1772                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1773                                             EXTENT_LOCKED);
1774                 if (ret)
1775                         break;
1776                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1777                                   GFP_NOFS);
1778                 if (!test_range_bit(&extent_root->fs_info->extent_ins,
1779                                     start, end, EXTENT_LOCKED, 0)) {
1780                         update_pinned_extents(extent_root, start,
1781                                               end + 1 - start, 1);
1782                         ret = __free_extent(trans, extent_root,
1783                                              start, end + 1 - start,
1784                                              extent_root->root_key.objectid,
1785                                              0, 0, 0, 0, 0);
1786                 } else {
1787                         clear_extent_bits(&extent_root->fs_info->extent_ins,
1788                                           start, end, EXTENT_LOCKED, GFP_NOFS);
1789                 }
1790                 if (ret)
1791                         err = ret;
1792
1793                 if (need_resched()) {
1794                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1795                         cond_resched();
1796                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1797                 }
1798         }
1799         return err;
1800 }
1801
1802 /*
1803  * remove an extent from the root, returns 0 on success
1804  */
1805 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1806                                struct btrfs_root *root, u64 bytenr,
1807                                u64 num_bytes, u64 root_objectid,
1808                                u64 ref_generation, u64 owner_objectid,
1809                                u64 owner_offset, int pin)
1810 {
1811         struct btrfs_root *extent_root = root->fs_info->extent_root;
1812         int pending_ret;
1813         int ret;
1814
1815         WARN_ON(num_bytes < root->sectorsize);
1816         if (!root->ref_cows)
1817                 ref_generation = 0;
1818
1819         if (root == extent_root) {
1820                 pin_down_bytes(root, bytenr, num_bytes, 1);
1821                 return 0;
1822         }
1823         ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1824                             ref_generation, owner_objectid, owner_offset,
1825                             pin, pin == 0);
1826
1827         finish_current_insert(trans, root->fs_info->extent_root);
1828         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1829         return ret ? ret : pending_ret;
1830 }
1831
1832 int btrfs_free_extent(struct btrfs_trans_handle *trans,
1833                       struct btrfs_root *root, u64 bytenr,
1834                       u64 num_bytes, u64 root_objectid,
1835                       u64 ref_generation, u64 owner_objectid,
1836                       u64 owner_offset, int pin)
1837 {
1838         int ret;
1839
1840         maybe_lock_mutex(root);
1841         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1842                                   root_objectid, ref_generation,
1843                                   owner_objectid, owner_offset, pin);
1844         maybe_unlock_mutex(root);
1845         return ret;
1846 }
1847
1848 static u64 stripe_align(struct btrfs_root *root, u64 val)
1849 {
1850         u64 mask = ((u64)root->stripesize - 1);
1851         u64 ret = (val + mask) & ~mask;
1852         return ret;
1853 }
1854
1855 /*
1856  * walks the btree of allocated extents and find a hole of a given size.
1857  * The key ins is changed to record the hole:
1858  * ins->objectid == block start
1859  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1860  * ins->offset == number of blocks
1861  * Any available blocks before search_start are skipped.
1862  */
1863 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1864                                      struct btrfs_root *orig_root,
1865                                      u64 num_bytes, u64 empty_size,
1866                                      u64 search_start, u64 search_end,
1867                                      u64 hint_byte, struct btrfs_key *ins,
1868                                      u64 exclude_start, u64 exclude_nr,
1869                                      int data)
1870 {
1871         int ret;
1872         u64 orig_search_start;
1873         struct btrfs_root * root = orig_root->fs_info->extent_root;
1874         struct btrfs_fs_info *info = root->fs_info;
1875         u64 total_needed = num_bytes;
1876         u64 *last_ptr = NULL;
1877         struct btrfs_block_group_cache *block_group;
1878         int full_scan = 0;
1879         int wrapped = 0;
1880         int chunk_alloc_done = 0;
1881         int empty_cluster = 2 * 1024 * 1024;
1882         int allowed_chunk_alloc = 0;
1883
1884         WARN_ON(num_bytes < root->sectorsize);
1885         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1886
1887         if (orig_root->ref_cows || empty_size)
1888                 allowed_chunk_alloc = 1;
1889
1890         if (data & BTRFS_BLOCK_GROUP_METADATA) {
1891                 last_ptr = &root->fs_info->last_alloc;
1892                 empty_cluster = 256 * 1024;
1893         }
1894
1895         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1896                 last_ptr = &root->fs_info->last_data_alloc;
1897         }
1898
1899         if (last_ptr) {
1900                 if (*last_ptr)
1901                         hint_byte = *last_ptr;
1902                 else {
1903                         empty_size += empty_cluster;
1904                 }
1905         }
1906
1907         search_start = max(search_start, first_logical_byte(root, 0));
1908         orig_search_start = search_start;
1909
1910         if (search_end == (u64)-1)
1911                 search_end = btrfs_super_total_bytes(&info->super_copy);
1912
1913         if (hint_byte) {
1914                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
1915                 if (!block_group)
1916                         hint_byte = search_start;
1917                 block_group = btrfs_find_block_group(root, block_group,
1918                                                      hint_byte, data, 1);
1919                 if (last_ptr && *last_ptr == 0 && block_group)
1920                         hint_byte = block_group->key.objectid;
1921         } else {
1922                 block_group = btrfs_find_block_group(root,
1923                                                      trans->block_group,
1924                                                      search_start, data, 1);
1925         }
1926         search_start = max(search_start, hint_byte);
1927
1928         total_needed += empty_size;
1929
1930 check_failed:
1931         if (!block_group) {
1932                 block_group = btrfs_lookup_first_block_group(info,
1933                                                              search_start);
1934                 if (!block_group)
1935                         block_group = btrfs_lookup_first_block_group(info,
1936                                                        orig_search_start);
1937         }
1938         if (full_scan && !chunk_alloc_done) {
1939                 if (allowed_chunk_alloc) {
1940                         do_chunk_alloc(trans, root,
1941                                      num_bytes + 2 * 1024 * 1024, data, 1);
1942                         allowed_chunk_alloc = 0;
1943                 } else if (block_group && block_group_bits(block_group, data)) {
1944                         block_group->space_info->force_alloc = 1;
1945                 }
1946                 chunk_alloc_done = 1;
1947         }
1948         ret = find_search_start(root, &block_group, &search_start,
1949                                 total_needed, data);
1950         if (ret == -ENOSPC && last_ptr && *last_ptr) {
1951                 *last_ptr = 0;
1952                 block_group = btrfs_lookup_first_block_group(info,
1953                                                              orig_search_start);
1954                 search_start = orig_search_start;
1955                 ret = find_search_start(root, &block_group, &search_start,
1956                                         total_needed, data);
1957         }
1958         if (ret == -ENOSPC)
1959                 goto enospc;
1960         if (ret)
1961                 goto error;
1962
1963         if (last_ptr && *last_ptr && search_start != *last_ptr) {
1964                 *last_ptr = 0;
1965                 if (!empty_size) {
1966                         empty_size += empty_cluster;
1967                         total_needed += empty_size;
1968                 }
1969                 block_group = btrfs_lookup_first_block_group(info,
1970                                                        orig_search_start);
1971                 search_start = orig_search_start;
1972                 ret = find_search_start(root, &block_group,
1973                                         &search_start, total_needed, data);
1974                 if (ret == -ENOSPC)
1975                         goto enospc;
1976                 if (ret)
1977                         goto error;
1978         }
1979
1980         search_start = stripe_align(root, search_start);
1981         ins->objectid = search_start;
1982         ins->offset = num_bytes;
1983
1984         if (ins->objectid + num_bytes >= search_end)
1985                 goto enospc;
1986
1987         if (ins->objectid + num_bytes >
1988             block_group->key.objectid + block_group->key.offset) {
1989                 search_start = block_group->key.objectid +
1990                         block_group->key.offset;
1991                 goto new_group;
1992         }
1993
1994         if (test_range_bit(&info->extent_ins, ins->objectid,
1995                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1996                 search_start = ins->objectid + num_bytes;
1997                 goto new_group;
1998         }
1999
2000         if (test_range_bit(&info->pinned_extents, ins->objectid,
2001                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2002                 search_start = ins->objectid + num_bytes;
2003                 goto new_group;
2004         }
2005
2006         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2007             ins->objectid < exclude_start + exclude_nr)) {
2008                 search_start = exclude_start + exclude_nr;
2009                 goto new_group;
2010         }
2011
2012         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2013                 block_group = btrfs_lookup_block_group(info, ins->objectid);
2014                 if (block_group)
2015                         trans->block_group = block_group;
2016         }
2017         ins->offset = num_bytes;
2018         if (last_ptr) {
2019                 *last_ptr = ins->objectid + ins->offset;
2020                 if (*last_ptr ==
2021                     btrfs_super_total_bytes(&root->fs_info->super_copy)) {
2022                         *last_ptr = 0;
2023                 }
2024         }
2025         return 0;
2026
2027 new_group:
2028         if (search_start + num_bytes >= search_end) {
2029 enospc:
2030                 search_start = orig_search_start;
2031                 if (full_scan) {
2032                         ret = -ENOSPC;
2033                         goto error;
2034                 }
2035                 if (wrapped) {
2036                         if (!full_scan)
2037                                 total_needed -= empty_size;
2038                         full_scan = 1;
2039                 } else
2040                         wrapped = 1;
2041         }
2042         block_group = btrfs_lookup_first_block_group(info, search_start);
2043         cond_resched();
2044         block_group = btrfs_find_block_group(root, block_group,
2045                                              search_start, data, 0);
2046         goto check_failed;
2047
2048 error:
2049         return ret;
2050 }
2051
2052 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2053                                   struct btrfs_root *root,
2054                                   u64 num_bytes, u64 min_alloc_size,
2055                                   u64 empty_size, u64 hint_byte,
2056                                   u64 search_end, struct btrfs_key *ins,
2057                                   u64 data)
2058 {
2059         int ret;
2060         u64 search_start = 0;
2061         u64 alloc_profile;
2062         struct btrfs_fs_info *info = root->fs_info;
2063
2064         if (data) {
2065                 alloc_profile = info->avail_data_alloc_bits &
2066                                 info->data_alloc_profile;
2067                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2068         } else if (root == root->fs_info->chunk_root) {
2069                 alloc_profile = info->avail_system_alloc_bits &
2070                                 info->system_alloc_profile;
2071                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2072         } else {
2073                 alloc_profile = info->avail_metadata_alloc_bits &
2074                                 info->metadata_alloc_profile;
2075                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2076         }
2077 again:
2078         data = reduce_alloc_profile(root, data);
2079         /*
2080          * the only place that sets empty_size is btrfs_realloc_node, which
2081          * is not called recursively on allocations
2082          */
2083         if (empty_size || root->ref_cows) {
2084                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2085                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2086                                      2 * 1024 * 1024,
2087                                      BTRFS_BLOCK_GROUP_METADATA |
2088                                      (info->metadata_alloc_profile &
2089                                       info->avail_metadata_alloc_bits), 0);
2090                         BUG_ON(ret);
2091                 }
2092                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2093                                      num_bytes + 2 * 1024 * 1024, data, 0);
2094                 BUG_ON(ret);
2095         }
2096
2097         WARN_ON(num_bytes < root->sectorsize);
2098         ret = find_free_extent(trans, root, num_bytes, empty_size,
2099                                search_start, search_end, hint_byte, ins,
2100                                trans->alloc_exclude_start,
2101                                trans->alloc_exclude_nr, data);
2102
2103         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2104                 num_bytes = num_bytes >> 1;
2105                 num_bytes = max(num_bytes, min_alloc_size);
2106                 do_chunk_alloc(trans, root->fs_info->extent_root,
2107                                num_bytes, data, 1);
2108                 goto again;
2109         }
2110         if (ret) {
2111                 printk("allocation failed flags %Lu\n", data);
2112                 BUG();
2113         }
2114         clear_extent_dirty(&root->fs_info->free_space_cache,
2115                            ins->objectid, ins->objectid + ins->offset - 1,
2116                            GFP_NOFS);
2117         return 0;
2118 }
2119
2120 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2121 {
2122         maybe_lock_mutex(root);
2123         set_extent_dirty(&root->fs_info->free_space_cache,
2124                          start, start + len - 1, GFP_NOFS);
2125         maybe_unlock_mutex(root);
2126         return 0;
2127 }
2128
2129 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2130                                   struct btrfs_root *root,
2131                                   u64 num_bytes, u64 min_alloc_size,
2132                                   u64 empty_size, u64 hint_byte,
2133                                   u64 search_end, struct btrfs_key *ins,
2134                                   u64 data)
2135 {
2136         int ret;
2137         maybe_lock_mutex(root);
2138         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2139                                      empty_size, hint_byte, search_end, ins,
2140                                      data);
2141         maybe_unlock_mutex(root);
2142         return ret;
2143 }
2144
2145 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2146                                          struct btrfs_root *root,
2147                                          u64 root_objectid, u64 ref_generation,
2148                                          u64 owner, u64 owner_offset,
2149                                          struct btrfs_key *ins)
2150 {
2151         int ret;
2152         int pending_ret;
2153         u64 super_used;
2154         u64 root_used;
2155         u64 num_bytes = ins->offset;
2156         u32 sizes[2];
2157         struct btrfs_fs_info *info = root->fs_info;
2158         struct btrfs_root *extent_root = info->extent_root;
2159         struct btrfs_extent_item *extent_item;
2160         struct btrfs_extent_ref *ref;
2161         struct btrfs_path *path;
2162         struct btrfs_key keys[2];
2163
2164         /* block accounting for super block */
2165         spin_lock_irq(&info->delalloc_lock);
2166         super_used = btrfs_super_bytes_used(&info->super_copy);
2167         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2168         spin_unlock_irq(&info->delalloc_lock);
2169
2170         /* block accounting for root item */
2171         root_used = btrfs_root_used(&root->root_item);
2172         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2173
2174         if (root == extent_root) {
2175                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2176                                 ins->objectid + ins->offset - 1,
2177                                 EXTENT_LOCKED, GFP_NOFS);
2178                 goto update_block;
2179         }
2180
2181         memcpy(&keys[0], ins, sizeof(*ins));
2182         keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
2183                                          owner, owner_offset);
2184         keys[1].objectid = ins->objectid;
2185         keys[1].type = BTRFS_EXTENT_REF_KEY;
2186         sizes[0] = sizeof(*extent_item);
2187         sizes[1] = sizeof(*ref);
2188
2189         path = btrfs_alloc_path();
2190         BUG_ON(!path);
2191
2192         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2193                                        sizes, 2);
2194
2195         BUG_ON(ret);
2196         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2197                                      struct btrfs_extent_item);
2198         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2199         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2200                              struct btrfs_extent_ref);
2201
2202         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2203         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2204         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2205         btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2206
2207         btrfs_mark_buffer_dirty(path->nodes[0]);
2208
2209         trans->alloc_exclude_start = 0;
2210         trans->alloc_exclude_nr = 0;
2211         btrfs_free_path(path);
2212         finish_current_insert(trans, extent_root);
2213         pending_ret = del_pending_extents(trans, extent_root);
2214
2215         if (ret)
2216                 goto out;
2217         if (pending_ret) {
2218                 ret = pending_ret;
2219                 goto out;
2220         }
2221
2222 update_block:
2223         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2224         if (ret) {
2225                 printk("update block group failed for %Lu %Lu\n",
2226                        ins->objectid, ins->offset);
2227                 BUG();
2228         }
2229 out:
2230         return ret;
2231 }
2232
2233 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2234                                 struct btrfs_root *root,
2235                                 u64 root_objectid, u64 ref_generation,
2236                                 u64 owner, u64 owner_offset,
2237                                 struct btrfs_key *ins)
2238 {
2239         int ret;
2240         maybe_lock_mutex(root);
2241         ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2242                                             ref_generation, owner,
2243                                             owner_offset, ins);
2244         maybe_unlock_mutex(root);
2245         return ret;
2246 }
2247 /*
2248  * finds a free extent and does all the dirty work required for allocation
2249  * returns the key for the extent through ins, and a tree buffer for
2250  * the first block of the extent through buf.
2251  *
2252  * returns 0 if everything worked, non-zero otherwise.
2253  */
2254 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2255                        struct btrfs_root *root,
2256                        u64 num_bytes, u64 min_alloc_size,
2257                        u64 root_objectid, u64 ref_generation,
2258                        u64 owner, u64 owner_offset,
2259                        u64 empty_size, u64 hint_byte,
2260                        u64 search_end, struct btrfs_key *ins, u64 data)
2261 {
2262         int ret;
2263
2264         maybe_lock_mutex(root);
2265
2266         ret = __btrfs_reserve_extent(trans, root, num_bytes,
2267                                      min_alloc_size, empty_size, hint_byte,
2268                                      search_end, ins, data);
2269         BUG_ON(ret);
2270         ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2271                                             ref_generation, owner,
2272                                             owner_offset, ins);
2273         BUG_ON(ret);
2274
2275         maybe_unlock_mutex(root);
2276         return ret;
2277 }
2278
2279 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2280                                             struct btrfs_root *root,
2281                                             u64 bytenr, u32 blocksize)
2282 {
2283         struct extent_buffer *buf;
2284
2285         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2286         if (!buf)
2287                 return ERR_PTR(-ENOMEM);
2288         btrfs_set_header_generation(buf, trans->transid);
2289         btrfs_tree_lock(buf);
2290         clean_tree_block(trans, root, buf);
2291         btrfs_set_buffer_uptodate(buf);
2292         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2293                          buf->start + buf->len - 1, GFP_NOFS);
2294         trans->blocks_used++;
2295         return buf;
2296 }
2297
2298 /*
2299  * helper function to allocate a block for a given tree
2300  * returns the tree buffer or NULL.
2301  */
2302 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2303                                              struct btrfs_root *root,
2304                                              u32 blocksize,
2305                                              u64 root_objectid,
2306                                              u64 ref_generation,
2307                                              u64 first_objectid,
2308                                              int level,
2309                                              u64 hint,
2310                                              u64 empty_size)
2311 {
2312         struct btrfs_key ins;
2313         int ret;
2314         struct extent_buffer *buf;
2315
2316         ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2317                                  root_objectid, ref_generation,
2318                                  level, first_objectid, empty_size, hint,
2319                                  (u64)-1, &ins, 0);
2320         if (ret) {
2321                 BUG_ON(ret > 0);
2322                 return ERR_PTR(ret);
2323         }
2324
2325         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2326         return buf;
2327 }
2328
2329 static int noinline drop_leaf_ref_no_cache(struct btrfs_trans_handle *trans,
2330                                            struct btrfs_root *root,
2331                                            struct extent_buffer *leaf)
2332 {
2333         u64 leaf_owner;
2334         u64 leaf_generation;
2335         struct btrfs_key key;
2336         struct btrfs_file_extent_item *fi;
2337         int i;
2338         int nritems;
2339         int ret;
2340
2341         BUG_ON(!btrfs_is_leaf(leaf));
2342         nritems = btrfs_header_nritems(leaf);
2343         leaf_owner = btrfs_header_owner(leaf);
2344         leaf_generation = btrfs_header_generation(leaf);
2345
2346         for (i = 0; i < nritems; i++) {
2347                 u64 disk_bytenr;
2348                 cond_resched();
2349
2350                 btrfs_item_key_to_cpu(leaf, &key, i);
2351                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2352                         continue;
2353                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2354                 if (btrfs_file_extent_type(leaf, fi) ==
2355                     BTRFS_FILE_EXTENT_INLINE)
2356                         continue;
2357                 /*
2358                  * FIXME make sure to insert a trans record that
2359                  * repeats the snapshot del on crash
2360                  */
2361                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2362                 if (disk_bytenr == 0)
2363                         continue;
2364
2365                 mutex_lock(&root->fs_info->alloc_mutex);
2366                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2367                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2368                                 leaf_owner, leaf_generation,
2369                                 key.objectid, key.offset, 0);
2370                 mutex_unlock(&root->fs_info->alloc_mutex);
2371
2372                 atomic_inc(&root->fs_info->throttle_gen);
2373                 wake_up(&root->fs_info->transaction_throttle);
2374                 cond_resched();
2375
2376                 BUG_ON(ret);
2377         }
2378         return 0;
2379 }
2380
2381 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2382                                          struct btrfs_root *root,
2383                                          struct btrfs_leaf_ref *ref)
2384 {
2385         int i;
2386         int ret;
2387         struct btrfs_extent_info *info = ref->extents;
2388
2389         for (i = 0; i < ref->nritems; i++) {
2390                 mutex_lock(&root->fs_info->alloc_mutex);
2391                 ret = __btrfs_free_extent(trans, root,
2392                                         info->bytenr, info->num_bytes,
2393                                         ref->owner, ref->generation,
2394                                         info->objectid, info->offset, 0);
2395                 mutex_unlock(&root->fs_info->alloc_mutex);
2396
2397                 atomic_inc(&root->fs_info->throttle_gen);
2398                 wake_up(&root->fs_info->transaction_throttle);
2399                 cond_resched();
2400
2401                 BUG_ON(ret);
2402                 info++;
2403         }
2404
2405         return 0;
2406 }
2407
2408 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2409                               u32 *refs)
2410 {
2411         int ret;
2412
2413         ret = lookup_extent_ref(NULL, root, start, len, refs);
2414         BUG_ON(ret);
2415
2416 #if 0 // some debugging code in case we see problems here
2417         /* if the refs count is one, it won't get increased again.  But
2418          * if the ref count is > 1, someone may be decreasing it at
2419          * the same time we are.
2420          */
2421         if (*refs != 1) {
2422                 struct extent_buffer *eb = NULL;
2423                 eb = btrfs_find_create_tree_block(root, start, len);
2424                 if (eb)
2425                         btrfs_tree_lock(eb);
2426
2427                 mutex_lock(&root->fs_info->alloc_mutex);
2428                 ret = lookup_extent_ref(NULL, root, start, len, refs);
2429                 BUG_ON(ret);
2430                 mutex_unlock(&root->fs_info->alloc_mutex);
2431
2432                 if (eb) {
2433                         btrfs_tree_unlock(eb);
2434                         free_extent_buffer(eb);
2435                 }
2436                 if (*refs == 1) {
2437                         printk("block %llu went down to one during drop_snap\n",
2438                                (unsigned long long)start);
2439                 }
2440
2441         }
2442 #endif
2443
2444         cond_resched();
2445         return ret;
2446 }
2447
2448 /*
2449  * helper function for drop_snapshot, this walks down the tree dropping ref
2450  * counts as it goes.
2451  */
2452 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2453                                    struct btrfs_root *root,
2454                                    struct btrfs_path *path, int *level)
2455 {
2456         u64 root_owner;
2457         u64 root_gen;
2458         u64 bytenr;
2459         u64 ptr_gen;
2460         struct extent_buffer *next;
2461         struct extent_buffer *cur;
2462         struct extent_buffer *parent;
2463         struct btrfs_leaf_ref *ref;
2464         u32 blocksize;
2465         int ret;
2466         u32 refs;
2467
2468         WARN_ON(*level < 0);
2469         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2470         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2471                                 path->nodes[*level]->len, &refs);
2472         BUG_ON(ret);
2473         if (refs > 1)
2474                 goto out;
2475
2476         /*
2477          * walk down to the last node level and free all the leaves
2478          */
2479         while(*level >= 0) {
2480                 WARN_ON(*level < 0);
2481                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2482                 cur = path->nodes[*level];
2483
2484                 if (btrfs_header_level(cur) != *level)
2485                         WARN_ON(1);
2486
2487                 if (path->slots[*level] >=
2488                     btrfs_header_nritems(cur))
2489                         break;
2490                 if (*level == 0) {
2491                         ret = drop_leaf_ref_no_cache(trans, root, cur);
2492                         BUG_ON(ret);
2493                         break;
2494                 }
2495                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2496                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2497                 blocksize = btrfs_level_size(root, *level - 1);
2498
2499                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2500                 BUG_ON(ret);
2501                 if (refs != 1) {
2502                         parent = path->nodes[*level];
2503                         root_owner = btrfs_header_owner(parent);
2504                         root_gen = btrfs_header_generation(parent);
2505                         path->slots[*level]++;
2506
2507                         mutex_lock(&root->fs_info->alloc_mutex);
2508                         ret = __btrfs_free_extent(trans, root, bytenr,
2509                                                 blocksize, root_owner,
2510                                                 root_gen, 0, 0, 1);
2511                         BUG_ON(ret);
2512                         mutex_unlock(&root->fs_info->alloc_mutex);
2513
2514                         atomic_inc(&root->fs_info->throttle_gen);
2515                         wake_up(&root->fs_info->transaction_throttle);
2516                         cond_resched();
2517
2518                         continue;
2519                 }
2520                 /*
2521                  * at this point, we have a single ref, and since the
2522                  * only place referencing this extent is a dead root
2523                  * the reference count should never go higher.
2524                  * So, we don't need to check it again
2525                  */
2526                 if (*level == 1) {
2527                         struct btrfs_key key;
2528                         btrfs_node_key_to_cpu(cur, &key, path->slots[*level]);
2529                         ref = btrfs_lookup_leaf_ref(root, bytenr);
2530                         if (ref) {
2531                                 ret = drop_leaf_ref(trans, root, ref);
2532                                 BUG_ON(ret);
2533                                 btrfs_remove_leaf_ref(root, ref);
2534                                 btrfs_free_leaf_ref(root, ref);
2535                                 *level = 0;
2536                                 break;
2537                         }
2538                         if (printk_ratelimit())
2539                                 printk("leaf ref miss for bytenr %llu\n",
2540                                        (unsigned long long)bytenr);
2541                 }
2542                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2543                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2544                         free_extent_buffer(next);
2545
2546                         next = read_tree_block(root, bytenr, blocksize,
2547                                                ptr_gen);
2548                         cond_resched();
2549 #if 0
2550                         /*
2551                          * this is a debugging check and can go away
2552                          * the ref should never go all the way down to 1
2553                          * at this point
2554                          */
2555                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2556                                                 &refs);
2557                         BUG_ON(ret);
2558                         WARN_ON(refs != 1);
2559 #endif
2560                 }
2561                 WARN_ON(*level <= 0);
2562                 if (path->nodes[*level-1])
2563                         free_extent_buffer(path->nodes[*level-1]);
2564                 path->nodes[*level-1] = next;
2565                 *level = btrfs_header_level(next);
2566                 path->slots[*level] = 0;
2567                 cond_resched();
2568         }
2569 out:
2570         WARN_ON(*level < 0);
2571         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2572
2573         if (path->nodes[*level] == root->node) {
2574                 parent = path->nodes[*level];
2575                 bytenr = path->nodes[*level]->start;
2576         } else {
2577                 parent = path->nodes[*level + 1];
2578                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
2579         }
2580
2581         blocksize = btrfs_level_size(root, *level);
2582         root_owner = btrfs_header_owner(parent);
2583         root_gen = btrfs_header_generation(parent);
2584
2585         mutex_lock(&root->fs_info->alloc_mutex);
2586         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
2587                                   root_owner, root_gen, 0, 0, 1);
2588         free_extent_buffer(path->nodes[*level]);
2589         path->nodes[*level] = NULL;
2590         *level += 1;
2591         BUG_ON(ret);
2592         mutex_unlock(&root->fs_info->alloc_mutex);
2593
2594         cond_resched();
2595         return 0;
2596 }
2597
2598 /*
2599  * helper for dropping snapshots.  This walks back up the tree in the path
2600  * to find the first node higher up where we haven't yet gone through
2601  * all the slots
2602  */
2603 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2604                                  struct btrfs_root *root,
2605                                  struct btrfs_path *path, int *level)
2606 {
2607         u64 root_owner;
2608         u64 root_gen;
2609         struct btrfs_root_item *root_item = &root->root_item;
2610         int i;
2611         int slot;
2612         int ret;
2613
2614         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2615                 slot = path->slots[i];
2616                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2617                         struct extent_buffer *node;
2618                         struct btrfs_disk_key disk_key;
2619                         node = path->nodes[i];
2620                         path->slots[i]++;
2621                         *level = i;
2622                         WARN_ON(*level == 0);
2623                         btrfs_node_key(node, &disk_key, path->slots[i]);
2624                         memcpy(&root_item->drop_progress,
2625                                &disk_key, sizeof(disk_key));
2626                         root_item->drop_level = i;
2627                         return 0;
2628                 } else {
2629                         if (path->nodes[*level] == root->node) {
2630                                 root_owner = root->root_key.objectid;
2631                                 root_gen =
2632                                    btrfs_header_generation(path->nodes[*level]);
2633                         } else {
2634                                 struct extent_buffer *node;
2635                                 node = path->nodes[*level + 1];
2636                                 root_owner = btrfs_header_owner(node);
2637                                 root_gen = btrfs_header_generation(node);
2638                         }
2639                         ret = btrfs_free_extent(trans, root,
2640                                                 path->nodes[*level]->start,
2641                                                 path->nodes[*level]->len,
2642                                                 root_owner, root_gen, 0, 0, 1);
2643                         BUG_ON(ret);
2644                         free_extent_buffer(path->nodes[*level]);
2645                         path->nodes[*level] = NULL;
2646                         *level = i + 1;
2647                 }
2648         }
2649         return 1;
2650 }
2651
2652 /*
2653  * drop the reference count on the tree rooted at 'snap'.  This traverses
2654  * the tree freeing any blocks that have a ref count of zero after being
2655  * decremented.
2656  */
2657 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2658                         *root)
2659 {
2660         int ret = 0;
2661         int wret;
2662         int level;
2663         struct btrfs_path *path;
2664         int i;
2665         int orig_level;
2666         struct btrfs_root_item *root_item = &root->root_item;
2667
2668         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2669         path = btrfs_alloc_path();
2670         BUG_ON(!path);
2671
2672         level = btrfs_header_level(root->node);
2673         orig_level = level;
2674         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2675                 path->nodes[level] = root->node;
2676                 extent_buffer_get(root->node);
2677                 path->slots[level] = 0;
2678         } else {
2679                 struct btrfs_key key;
2680                 struct btrfs_disk_key found_key;
2681                 struct extent_buffer *node;
2682
2683                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2684                 level = root_item->drop_level;
2685                 path->lowest_level = level;
2686                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2687                 if (wret < 0) {
2688                         ret = wret;
2689                         goto out;
2690                 }
2691                 node = path->nodes[level];
2692                 btrfs_node_key(node, &found_key, path->slots[level]);
2693                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2694                                sizeof(found_key)));
2695                 /*
2696                  * unlock our path, this is safe because only this
2697                  * function is allowed to delete this snapshot
2698                  */
2699                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2700                         if (path->nodes[i] && path->locks[i]) {
2701                                 path->locks[i] = 0;
2702                                 btrfs_tree_unlock(path->nodes[i]);
2703                         }
2704                 }
2705         }
2706         while(1) {
2707                 wret = walk_down_tree(trans, root, path, &level);
2708                 if (wret > 0)
2709                         break;
2710                 if (wret < 0)
2711                         ret = wret;
2712
2713                 wret = walk_up_tree(trans, root, path, &level);
2714                 if (wret > 0)
2715                         break;
2716                 if (wret < 0)
2717                         ret = wret;
2718                 if (trans->transaction->in_commit) {
2719                         ret = -EAGAIN;
2720                         break;
2721                 }
2722                 atomic_inc(&root->fs_info->throttle_gen);
2723                 wake_up(&root->fs_info->transaction_throttle);
2724         }
2725         for (i = 0; i <= orig_level; i++) {
2726                 if (path->nodes[i]) {
2727                         free_extent_buffer(path->nodes[i]);
2728                         path->nodes[i] = NULL;
2729                 }
2730         }
2731 out:
2732         btrfs_free_path(path);
2733         return ret;
2734 }
2735
2736 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2737 {
2738         u64 start;
2739         u64 end;
2740         u64 ptr;
2741         int ret;
2742
2743         mutex_lock(&info->alloc_mutex);
2744         while(1) {
2745                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2746                                             &start, &end, (unsigned int)-1);
2747                 if (ret)
2748                         break;
2749                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2750                 if (!ret)
2751                         kfree((void *)(unsigned long)ptr);
2752                 clear_extent_bits(&info->block_group_cache, start,
2753                                   end, (unsigned int)-1, GFP_NOFS);
2754         }
2755         while(1) {
2756                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2757                                             &start, &end, EXTENT_DIRTY);
2758                 if (ret)
2759                         break;
2760                 clear_extent_dirty(&info->free_space_cache, start,
2761                                    end, GFP_NOFS);
2762         }
2763         mutex_unlock(&info->alloc_mutex);
2764         return 0;
2765 }
2766
2767 static unsigned long calc_ra(unsigned long start, unsigned long last,
2768                              unsigned long nr)
2769 {
2770         return min(last, start + nr - 1);
2771 }
2772
2773 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2774                                          u64 len)
2775 {
2776         u64 page_start;
2777         u64 page_end;
2778         unsigned long last_index;
2779         unsigned long i;
2780         struct page *page;
2781         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2782         struct file_ra_state *ra;
2783         unsigned long total_read = 0;
2784         unsigned long ra_pages;
2785         struct btrfs_ordered_extent *ordered;
2786         struct btrfs_trans_handle *trans;
2787
2788         ra = kzalloc(sizeof(*ra), GFP_NOFS);
2789
2790         mutex_lock(&inode->i_mutex);
2791         i = start >> PAGE_CACHE_SHIFT;
2792         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2793
2794         ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2795
2796         file_ra_state_init(ra, inode->i_mapping);
2797
2798         for (; i <= last_index; i++) {
2799                 if (total_read % ra_pages == 0) {
2800                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2801                                        calc_ra(i, last_index, ra_pages));
2802                 }
2803                 total_read++;
2804 again:
2805                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
2806                         goto truncate_racing;
2807                 page = grab_cache_page(inode->i_mapping, i);
2808                 if (!page) {
2809                         goto out_unlock;
2810                 }
2811                 if (!PageUptodate(page)) {
2812                         btrfs_readpage(NULL, page);
2813                         lock_page(page);
2814                         if (!PageUptodate(page)) {
2815                                 unlock_page(page);
2816                                 page_cache_release(page);
2817                                 goto out_unlock;
2818                         }
2819                 }
2820                 wait_on_page_writeback(page);
2821
2822                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2823                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2824                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2825
2826                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2827                 if (ordered) {
2828                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2829                         unlock_page(page);
2830                         page_cache_release(page);
2831                         btrfs_start_ordered_extent(inode, ordered, 1);
2832                         btrfs_put_ordered_extent(ordered);
2833                         goto again;
2834                 }
2835                 set_page_extent_mapped(page);
2836
2837                 /*
2838                  * make sure page_mkwrite is called for this page if userland
2839                  * wants to change it from mmap
2840                  */
2841                 clear_page_dirty_for_io(page);
2842
2843                 btrfs_set_extent_delalloc(inode, page_start, page_end);
2844                 set_page_dirty(page);
2845
2846                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2847                 unlock_page(page);
2848                 page_cache_release(page);
2849         }
2850
2851 out_unlock:
2852         /* we have to start the IO in order to get the ordered extents
2853          * instantiated.  This allows the relocation to code to wait
2854          * for all the ordered extents to hit the disk.
2855          *
2856          * Otherwise, it would constantly loop over the same extents
2857          * because the old ones don't get deleted  until the IO is
2858          * started
2859          */
2860         btrfs_fdatawrite_range(inode->i_mapping, start, start + len - 1,
2861                                WB_SYNC_NONE);
2862         kfree(ra);
2863         trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2864         if (trans) {
2865                 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2866                 mark_inode_dirty(inode);
2867         }
2868         mutex_unlock(&inode->i_mutex);
2869         return 0;
2870
2871 truncate_racing:
2872         vmtruncate(inode, inode->i_size);
2873         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2874                                            total_read);
2875         goto out_unlock;
2876 }
2877
2878 /*
2879  * The back references tell us which tree holds a ref on a block,
2880  * but it is possible for the tree root field in the reference to
2881  * reflect the original root before a snapshot was made.  In this
2882  * case we should search through all the children of a given root
2883  * to find potential holders of references on a block.
2884  *
2885  * Instead, we do something a little less fancy and just search
2886  * all the roots for a given key/block combination.
2887  */
2888 static int find_root_for_ref(struct btrfs_root *root,
2889                              struct btrfs_path *path,
2890                              struct btrfs_key *key0,
2891                              int level,
2892                              int file_key,
2893                              struct btrfs_root **found_root,
2894                              u64 bytenr)
2895 {
2896         struct btrfs_key root_location;
2897         struct btrfs_root *cur_root = *found_root;
2898         struct btrfs_file_extent_item *file_extent;
2899         u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2900         u64 found_bytenr;
2901         int ret;
2902
2903         root_location.offset = (u64)-1;
2904         root_location.type = BTRFS_ROOT_ITEM_KEY;
2905         path->lowest_level = level;
2906         path->reada = 0;
2907         while(1) {
2908                 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2909                 found_bytenr = 0;
2910                 if (ret == 0 && file_key) {
2911                         struct extent_buffer *leaf = path->nodes[0];
2912                         file_extent = btrfs_item_ptr(leaf, path->slots[0],
2913                                              struct btrfs_file_extent_item);
2914                         if (btrfs_file_extent_type(leaf, file_extent) ==
2915                             BTRFS_FILE_EXTENT_REG) {
2916                                 found_bytenr =
2917                                         btrfs_file_extent_disk_bytenr(leaf,
2918                                                                file_extent);
2919                        }
2920                 } else if (!file_key) {
2921                         if (path->nodes[level])
2922                                 found_bytenr = path->nodes[level]->start;
2923                 }
2924
2925                 btrfs_release_path(cur_root, path);
2926
2927                 if (found_bytenr == bytenr) {
2928                         *found_root = cur_root;
2929                         ret = 0;
2930                         goto out;
2931                 }
2932                 ret = btrfs_search_root(root->fs_info->tree_root,
2933                                         root_search_start, &root_search_start);
2934                 if (ret)
2935                         break;
2936
2937                 root_location.objectid = root_search_start;
2938                 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2939                                                       &root_location);
2940                 if (!cur_root) {
2941                         ret = 1;
2942                         break;
2943                 }
2944         }
2945 out:
2946         path->lowest_level = 0;
2947         return ret;
2948 }
2949
2950 /*
2951  * note, this releases the path
2952  */
2953 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2954                                   struct btrfs_path *path,
2955                                   struct btrfs_key *extent_key,
2956                                   u64 *last_file_objectid,
2957                                   u64 *last_file_offset,
2958                                   u64 *last_file_root,
2959                                   u64 last_extent)
2960 {
2961         struct inode *inode;
2962         struct btrfs_root *found_root;
2963         struct btrfs_key root_location;
2964         struct btrfs_key found_key;
2965         struct btrfs_extent_ref *ref;
2966         u64 ref_root;
2967         u64 ref_gen;
2968         u64 ref_objectid;
2969         u64 ref_offset;
2970         int ret;
2971         int level;
2972
2973         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
2974
2975         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2976                              struct btrfs_extent_ref);
2977         ref_root = btrfs_ref_root(path->nodes[0], ref);
2978         ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2979         ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2980         ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2981         btrfs_release_path(extent_root, path);
2982
2983         root_location.objectid = ref_root;
2984         if (ref_gen == 0)
2985                 root_location.offset = 0;
2986         else
2987                 root_location.offset = (u64)-1;
2988         root_location.type = BTRFS_ROOT_ITEM_KEY;
2989
2990         found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2991                                                 &root_location);
2992         BUG_ON(!found_root);
2993         mutex_unlock(&extent_root->fs_info->alloc_mutex);
2994
2995         if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2996                 found_key.objectid = ref_objectid;
2997                 found_key.type = BTRFS_EXTENT_DATA_KEY;
2998                 found_key.offset = ref_offset;
2999                 level = 0;
3000
3001                 if (last_extent == extent_key->objectid &&
3002                     *last_file_objectid == ref_objectid &&
3003                     *last_file_offset == ref_offset &&
3004                     *last_file_root == ref_root)
3005                         goto out;
3006
3007                 ret = find_root_for_ref(extent_root, path, &found_key,
3008                                         level, 1, &found_root,
3009                                         extent_key->objectid);
3010
3011                 if (ret)
3012                         goto out;
3013
3014                 if (last_extent == extent_key->objectid &&
3015                     *last_file_objectid == ref_objectid &&
3016                     *last_file_offset == ref_offset &&
3017                     *last_file_root == ref_root)
3018                         goto out;
3019
3020                 inode = btrfs_iget_locked(extent_root->fs_info->sb,
3021                                           ref_objectid, found_root);
3022                 if (inode->i_state & I_NEW) {
3023                         /* the inode and parent dir are two different roots */
3024                         BTRFS_I(inode)->root = found_root;
3025                         BTRFS_I(inode)->location.objectid = ref_objectid;
3026                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
3027                         BTRFS_I(inode)->location.offset = 0;
3028                         btrfs_read_locked_inode(inode);
3029                         unlock_new_inode(inode);
3030
3031                 }
3032                 /* this can happen if the reference is not against
3033                  * the latest version of the tree root
3034                  */
3035                 if (is_bad_inode(inode))
3036                         goto out;
3037
3038                 *last_file_objectid = inode->i_ino;
3039                 *last_file_root = found_root->root_key.objectid;
3040                 *last_file_offset = ref_offset;
3041
3042                 relocate_inode_pages(inode, ref_offset, extent_key->offset);
3043                 iput(inode);
3044         } else {
3045                 struct btrfs_trans_handle *trans;
3046                 struct extent_buffer *eb;
3047                 int needs_lock = 0;
3048
3049                 eb = read_tree_block(found_root, extent_key->objectid,
3050                                      extent_key->offset, 0);
3051                 btrfs_tree_lock(eb);
3052                 level = btrfs_header_level(eb);
3053
3054                 if (level == 0)
3055                         btrfs_item_key_to_cpu(eb, &found_key, 0);
3056                 else
3057                         btrfs_node_key_to_cpu(eb, &found_key, 0);
3058
3059                 btrfs_tree_unlock(eb);
3060                 free_extent_buffer(eb);
3061
3062                 ret = find_root_for_ref(extent_root, path, &found_key,
3063                                         level, 0, &found_root,
3064                                         extent_key->objectid);
3065
3066                 if (ret)
3067                         goto out;
3068
3069                 /*
3070                  * right here almost anything could happen to our key,
3071                  * but that's ok.  The cow below will either relocate it
3072                  * or someone else will have relocated it.  Either way,
3073                  * it is in a different spot than it was before and
3074                  * we're happy.
3075                  */
3076
3077                 trans = btrfs_start_transaction(found_root, 1);
3078
3079                 if (found_root == extent_root->fs_info->extent_root ||
3080                     found_root == extent_root->fs_info->chunk_root ||
3081                     found_root == extent_root->fs_info->dev_root) {
3082                         needs_lock = 1;
3083                         mutex_lock(&extent_root->fs_info->alloc_mutex);
3084                 }
3085
3086                 path->lowest_level = level;
3087                 path->reada = 2;
3088                 ret = btrfs_search_slot(trans, found_root, &found_key, path,
3089                                         0, 1);
3090                 path->lowest_level = 0;
3091                 btrfs_release_path(found_root, path);
3092
3093                 if (found_root == found_root->fs_info->extent_root)
3094                         btrfs_extent_post_op(trans, found_root);
3095                 if (needs_lock)
3096                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
3097
3098                 btrfs_end_transaction(trans, found_root);
3099
3100         }
3101 out:
3102         mutex_lock(&extent_root->fs_info->alloc_mutex);
3103         return 0;
3104 }
3105
3106 static int noinline del_extent_zero(struct btrfs_root *extent_root,
3107                                     struct btrfs_path *path,
3108                                     struct btrfs_key *extent_key)
3109 {
3110         int ret;
3111         struct btrfs_trans_handle *trans;
3112
3113         trans = btrfs_start_transaction(extent_root, 1);
3114         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
3115         if (ret > 0) {
3116                 ret = -EIO;
3117                 goto out;
3118         }
3119         if (ret < 0)
3120                 goto out;
3121         ret = btrfs_del_item(trans, extent_root, path);
3122 out:
3123         btrfs_end_transaction(trans, extent_root);
3124         return ret;
3125 }
3126
3127 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
3128                                         struct btrfs_path *path,
3129                                         struct btrfs_key *extent_key)
3130 {
3131         struct btrfs_key key;
3132         struct btrfs_key found_key;
3133         struct extent_buffer *leaf;
3134         u64 last_file_objectid = 0;
3135         u64 last_file_root = 0;
3136         u64 last_file_offset = (u64)-1;
3137         u64 last_extent = 0;
3138         u32 nritems;
3139         u32 item_size;
3140         int ret = 0;
3141
3142         if (extent_key->objectid == 0) {
3143                 ret = del_extent_zero(extent_root, path, extent_key);
3144                 goto out;
3145         }
3146         key.objectid = extent_key->objectid;
3147         key.type = BTRFS_EXTENT_REF_KEY;
3148         key.offset = 0;
3149
3150         while(1) {
3151                 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3152
3153                 if (ret < 0)
3154                         goto out;
3155
3156                 ret = 0;
3157                 leaf = path->nodes[0];
3158                 nritems = btrfs_header_nritems(leaf);
3159                 if (path->slots[0] == nritems) {
3160                         ret = btrfs_next_leaf(extent_root, path);
3161                         if (ret > 0) {
3162                                 ret = 0;
3163                                 goto out;
3164                         }
3165                         if (ret < 0)
3166                                 goto out;
3167                         leaf = path->nodes[0];
3168                 }
3169
3170                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3171                 if (found_key.objectid != extent_key->objectid) {
3172                         break;
3173                 }
3174
3175                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
3176                         break;
3177                 }
3178
3179                 key.offset = found_key.offset + 1;
3180                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3181
3182                 ret = relocate_one_reference(extent_root, path, extent_key,
3183                                              &last_file_objectid,
3184                                              &last_file_offset,
3185                                              &last_file_root, last_extent);
3186                 if (ret)
3187                         goto out;
3188                 last_extent = extent_key->objectid;
3189         }
3190         ret = 0;
3191 out:
3192         btrfs_release_path(extent_root, path);
3193         return ret;
3194 }
3195
3196 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
3197 {
3198         u64 num_devices;
3199         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
3200                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
3201
3202         num_devices = root->fs_info->fs_devices->num_devices;
3203         if (num_devices == 1) {
3204                 stripped |= BTRFS_BLOCK_GROUP_DUP;
3205                 stripped = flags & ~stripped;
3206
3207                 /* turn raid0 into single device chunks */
3208                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
3209                         return stripped;
3210
3211                 /* turn mirroring into duplication */
3212                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3213                              BTRFS_BLOCK_GROUP_RAID10))
3214                         return stripped | BTRFS_BLOCK_GROUP_DUP;
3215                 return flags;
3216         } else {
3217                 /* they already had raid on here, just return */
3218                 if (flags & stripped)
3219                         return flags;
3220
3221                 stripped |= BTRFS_BLOCK_GROUP_DUP;
3222                 stripped = flags & ~stripped;
3223
3224                 /* switch duplicated blocks with raid1 */
3225                 if (flags & BTRFS_BLOCK_GROUP_DUP)
3226                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
3227
3228                 /* turn single device chunks into raid0 */
3229                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
3230         }
3231         return flags;
3232 }
3233
3234 int __alloc_chunk_for_shrink(struct btrfs_root *root,
3235                      struct btrfs_block_group_cache *shrink_block_group,
3236                      int force)
3237 {
3238         struct btrfs_trans_handle *trans;
3239         u64 new_alloc_flags;
3240         u64 calc;
3241
3242         spin_lock(&shrink_block_group->lock);
3243         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
3244                 spin_unlock(&shrink_block_group->lock);
3245                 mutex_unlock(&root->fs_info->alloc_mutex);
3246
3247                 trans = btrfs_start_transaction(root, 1);
3248                 mutex_lock(&root->fs_info->alloc_mutex);
3249                 spin_lock(&shrink_block_group->lock);
3250
3251                 new_alloc_flags = update_block_group_flags(root,
3252                                                    shrink_block_group->flags);
3253                 if (new_alloc_flags != shrink_block_group->flags) {
3254                         calc =
3255                              btrfs_block_group_used(&shrink_block_group->item);
3256                 } else {
3257                         calc = shrink_block_group->key.offset;
3258                 }
3259                 spin_unlock(&shrink_block_group->lock);
3260
3261                 do_chunk_alloc(trans, root->fs_info->extent_root,
3262                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
3263
3264                 mutex_unlock(&root->fs_info->alloc_mutex);
3265                 btrfs_end_transaction(trans, root);
3266                 mutex_lock(&root->fs_info->alloc_mutex);
3267         } else
3268                 spin_unlock(&shrink_block_group->lock);
3269         return 0;
3270 }
3271
3272 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
3273 {
3274         struct btrfs_trans_handle *trans;
3275         struct btrfs_root *tree_root = root->fs_info->tree_root;
3276         struct btrfs_path *path;
3277         u64 cur_byte;
3278         u64 total_found;
3279         u64 shrink_last_byte;
3280         struct btrfs_block_group_cache *shrink_block_group;
3281         struct btrfs_fs_info *info = root->fs_info;
3282         struct btrfs_key key;
3283         struct btrfs_key found_key;
3284         struct extent_buffer *leaf;
3285         u32 nritems;
3286         int ret;
3287         int progress;
3288
3289         mutex_lock(&root->fs_info->alloc_mutex);
3290         shrink_block_group = btrfs_lookup_block_group(root->fs_info,
3291                                                       shrink_start);
3292         BUG_ON(!shrink_block_group);
3293
3294         shrink_last_byte = shrink_block_group->key.objectid +
3295                 shrink_block_group->key.offset;
3296
3297         shrink_block_group->space_info->total_bytes -=
3298                 shrink_block_group->key.offset;
3299         path = btrfs_alloc_path();
3300         root = root->fs_info->extent_root;
3301         path->reada = 2;
3302
3303         printk("btrfs relocating block group %llu flags %llu\n",
3304                (unsigned long long)shrink_start,
3305                (unsigned long long)shrink_block_group->flags);
3306
3307         __alloc_chunk_for_shrink(root, shrink_block_group, 1);
3308
3309 again:
3310
3311         shrink_block_group->ro = 1;
3312
3313         total_found = 0;
3314         progress = 0;
3315         key.objectid = shrink_start;
3316         key.offset = 0;
3317         key.type = 0;
3318         cur_byte = key.objectid;
3319
3320         mutex_unlock(&root->fs_info->alloc_mutex);
3321
3322         btrfs_start_delalloc_inodes(root);
3323         btrfs_wait_ordered_extents(tree_root);
3324
3325         mutex_lock(&root->fs_info->alloc_mutex);
3326
3327         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3328         if (ret < 0)
3329                 goto out;
3330
3331         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
3332         if (ret < 0)
3333                 goto out;
3334
3335         if (ret == 0) {
3336                 leaf = path->nodes[0];
3337                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3338                 if (found_key.objectid + found_key.offset > shrink_start &&
3339                     found_key.objectid < shrink_last_byte) {
3340                         cur_byte = found_key.objectid;
3341                         key.objectid = cur_byte;
3342                 }
3343         }
3344         btrfs_release_path(root, path);
3345
3346         while(1) {
3347                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3348                 if (ret < 0)
3349                         goto out;
3350
3351 next:
3352                 leaf = path->nodes[0];
3353                 nritems = btrfs_header_nritems(leaf);
3354                 if (path->slots[0] >= nritems) {
3355                         ret = btrfs_next_leaf(root, path);
3356                         if (ret < 0)
3357                                 goto out;
3358                         if (ret == 1) {
3359                                 ret = 0;
3360                                 break;
3361                         }
3362                         leaf = path->nodes[0];
3363                         nritems = btrfs_header_nritems(leaf);
3364                 }
3365
3366                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3367
3368                 if (found_key.objectid >= shrink_last_byte)
3369                         break;
3370
3371                 if (progress && need_resched()) {
3372                         memcpy(&key, &found_key, sizeof(key));
3373                         cond_resched();
3374                         btrfs_release_path(root, path);
3375                         btrfs_search_slot(NULL, root, &key, path, 0, 0);
3376                         progress = 0;
3377                         goto next;
3378                 }
3379                 progress = 1;
3380
3381                 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3382                     found_key.objectid + found_key.offset <= cur_byte) {
3383                         memcpy(&key, &found_key, sizeof(key));
3384                         key.offset++;
3385                         path->slots[0]++;
3386                         goto next;
3387                 }
3388
3389                 total_found++;
3390                 cur_byte = found_key.objectid + found_key.offset;
3391                 key.objectid = cur_byte;
3392                 btrfs_release_path(root, path);
3393                 ret = relocate_one_extent(root, path, &found_key);
3394                 __alloc_chunk_for_shrink(root, shrink_block_group, 0);
3395         }
3396
3397         btrfs_release_path(root, path);
3398
3399         if (total_found > 0) {
3400                 printk("btrfs relocate found %llu last extent was %llu\n",
3401                        (unsigned long long)total_found,
3402                        (unsigned long long)found_key.objectid);
3403                 mutex_unlock(&root->fs_info->alloc_mutex);
3404                 trans = btrfs_start_transaction(tree_root, 1);
3405                 btrfs_commit_transaction(trans, tree_root);
3406
3407                 btrfs_clean_old_snapshots(tree_root);
3408
3409                 btrfs_start_delalloc_inodes(root);
3410                 btrfs_wait_ordered_extents(tree_root);
3411
3412                 trans = btrfs_start_transaction(tree_root, 1);
3413                 btrfs_commit_transaction(trans, tree_root);
3414                 mutex_lock(&root->fs_info->alloc_mutex);
3415                 goto again;
3416         }
3417
3418         /*
3419          * we've freed all the extents, now remove the block
3420          * group item from the tree
3421          */
3422         mutex_unlock(&root->fs_info->alloc_mutex);
3423
3424         trans = btrfs_start_transaction(root, 1);
3425
3426         mutex_lock(&root->fs_info->alloc_mutex);
3427         memcpy(&key, &shrink_block_group->key, sizeof(key));
3428
3429         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3430         if (ret > 0)
3431                 ret = -EIO;
3432         if (ret < 0) {
3433                 btrfs_end_transaction(trans, root);
3434                 goto out;
3435         }
3436
3437         clear_extent_bits(&info->block_group_cache, key.objectid,
3438                           key.objectid + key.offset - 1,
3439                           (unsigned int)-1, GFP_NOFS);
3440
3441
3442         clear_extent_bits(&info->free_space_cache,
3443                            key.objectid, key.objectid + key.offset - 1,
3444                            (unsigned int)-1, GFP_NOFS);
3445
3446         /*
3447         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3448         kfree(shrink_block_group);
3449         */
3450
3451         btrfs_del_item(trans, root, path);
3452         btrfs_release_path(root, path);
3453         mutex_unlock(&root->fs_info->alloc_mutex);
3454         btrfs_commit_transaction(trans, root);
3455
3456         mutex_lock(&root->fs_info->alloc_mutex);
3457
3458         /* the code to unpin extents might set a few bits in the free
3459          * space cache for this range again
3460          */
3461         clear_extent_bits(&info->free_space_cache,
3462                            key.objectid, key.objectid + key.offset - 1,
3463                            (unsigned int)-1, GFP_NOFS);
3464 out:
3465         btrfs_free_path(path);
3466         mutex_unlock(&root->fs_info->alloc_mutex);
3467         return ret;
3468 }
3469
3470 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3471                            struct btrfs_key *key)
3472 {
3473         int ret = 0;
3474         struct btrfs_key found_key;
3475         struct extent_buffer *leaf;
3476         int slot;
3477
3478         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3479         if (ret < 0)
3480                 goto out;
3481
3482         while(1) {
3483                 slot = path->slots[0];
3484                 leaf = path->nodes[0];
3485                 if (slot >= btrfs_header_nritems(leaf)) {
3486                         ret = btrfs_next_leaf(root, path);
3487                         if (ret == 0)
3488                                 continue;
3489                         if (ret < 0)
3490                                 goto out;
3491                         break;
3492                 }
3493                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3494
3495                 if (found_key.objectid >= key->objectid &&
3496                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3497                         ret = 0;
3498                         goto out;
3499                 }
3500                 path->slots[0]++;
3501         }
3502         ret = -ENOENT;
3503 out:
3504         return ret;
3505 }
3506
3507 int btrfs_read_block_groups(struct btrfs_root *root)
3508 {
3509         struct btrfs_path *path;
3510         int ret;
3511         int bit;
3512         struct btrfs_block_group_cache *cache;
3513         struct btrfs_fs_info *info = root->fs_info;
3514         struct btrfs_space_info *space_info;
3515         struct extent_io_tree *block_group_cache;
3516         struct btrfs_key key;
3517         struct btrfs_key found_key;
3518         struct extent_buffer *leaf;
3519
3520         block_group_cache = &info->block_group_cache;
3521         root = info->extent_root;
3522         key.objectid = 0;
3523         key.offset = 0;
3524         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3525         path = btrfs_alloc_path();
3526         if (!path)
3527                 return -ENOMEM;
3528
3529         mutex_lock(&root->fs_info->alloc_mutex);
3530         while(1) {
3531                 ret = find_first_block_group(root, path, &key);
3532                 if (ret > 0) {
3533                         ret = 0;
3534                         goto error;
3535                 }
3536                 if (ret != 0)
3537                         goto error;
3538
3539                 leaf = path->nodes[0];
3540                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3541                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3542                 if (!cache) {
3543                         ret = -ENOMEM;
3544                         break;
3545                 }
3546
3547                 spin_lock_init(&cache->lock);
3548                 read_extent_buffer(leaf, &cache->item,
3549                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3550                                    sizeof(cache->item));
3551                 memcpy(&cache->key, &found_key, sizeof(found_key));
3552
3553                 key.objectid = found_key.objectid + found_key.offset;
3554                 btrfs_release_path(root, path);
3555                 cache->flags = btrfs_block_group_flags(&cache->item);
3556                 bit = 0;
3557                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3558                         bit = BLOCK_GROUP_DATA;
3559                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3560                         bit = BLOCK_GROUP_SYSTEM;
3561                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3562                         bit = BLOCK_GROUP_METADATA;
3563                 }
3564                 set_avail_alloc_bits(info, cache->flags);
3565
3566                 ret = update_space_info(info, cache->flags, found_key.offset,
3567                                         btrfs_block_group_used(&cache->item),
3568                                         &space_info);
3569                 BUG_ON(ret);
3570                 cache->space_info = space_info;
3571
3572                 /* use EXTENT_LOCKED to prevent merging */
3573                 set_extent_bits(block_group_cache, found_key.objectid,
3574                                 found_key.objectid + found_key.offset - 1,
3575                                 EXTENT_LOCKED, GFP_NOFS);
3576                 set_state_private(block_group_cache, found_key.objectid,
3577                                   (unsigned long)cache);
3578                 set_extent_bits(block_group_cache, found_key.objectid,
3579                                 found_key.objectid + found_key.offset - 1,
3580                                 bit | EXTENT_LOCKED, GFP_NOFS);
3581                 if (key.objectid >=
3582                     btrfs_super_total_bytes(&info->super_copy))
3583                         break;
3584         }
3585         ret = 0;
3586 error:
3587         btrfs_free_path(path);
3588         mutex_unlock(&root->fs_info->alloc_mutex);
3589         return ret;
3590 }
3591
3592 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3593                            struct btrfs_root *root, u64 bytes_used,
3594                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3595                            u64 size)
3596 {
3597         int ret;
3598         int bit = 0;
3599         struct btrfs_root *extent_root;
3600         struct btrfs_block_group_cache *cache;
3601         struct extent_io_tree *block_group_cache;
3602
3603         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
3604         extent_root = root->fs_info->extent_root;
3605         block_group_cache = &root->fs_info->block_group_cache;
3606
3607         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3608         BUG_ON(!cache);
3609         cache->key.objectid = chunk_offset;
3610         cache->key.offset = size;
3611         spin_lock_init(&cache->lock);
3612         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3613
3614         btrfs_set_block_group_used(&cache->item, bytes_used);
3615         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3616         cache->flags = type;
3617         btrfs_set_block_group_flags(&cache->item, type);
3618
3619         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3620                                 &cache->space_info);
3621         BUG_ON(ret);
3622
3623         bit = block_group_state_bits(type);
3624         set_extent_bits(block_group_cache, chunk_offset,
3625                         chunk_offset + size - 1,
3626                         EXTENT_LOCKED, GFP_NOFS);
3627         set_state_private(block_group_cache, chunk_offset,
3628                           (unsigned long)cache);
3629         set_extent_bits(block_group_cache, chunk_offset,
3630                         chunk_offset + size - 1,
3631                         bit | EXTENT_LOCKED, GFP_NOFS);
3632
3633         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3634                                 sizeof(cache->item));
3635         BUG_ON(ret);
3636
3637         finish_current_insert(trans, extent_root);
3638         ret = del_pending_extents(trans, extent_root);
3639         BUG_ON(ret);
3640         set_avail_alloc_bits(extent_root->fs_info, type);
3641
3642         return 0;
3643 }