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

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include "hash.h"
  23 #include "crc32c.h"
  24 #include "ctree.h"
  25 #include "disk-io.h"
  26 #include "print-tree.h"
  27 #include "transaction.h"
  28 #include "volumes.h"
  29 #include "locking.h"
  30 #include "ref-cache.h"
  31
  32 #define BLOCK_GROUP_DATA     EXTENT_WRITEBACK
  33 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
  34 #define BLOCK_GROUP_SYSTEM   EXTENT_NEW
  35
  36 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
  37
  38 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  39                                  btrfs_root *extent_root);
  40 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  41                                btrfs_root *extent_root);
  42 static struct btrfs_block_group_cache *
  43 __btrfs_find_block_group(struct btrfs_root *root,
  44                          struct btrfs_block_group_cache *hint,
  45                          u64 search_start, int data, int owner);
  46
  47 void maybe_lock_mutex(struct btrfs_root *root)
  48 {
  49         if (root != root->fs_info->extent_root &&
  50             root != root->fs_info->chunk_root &&
  51             root != root->fs_info->dev_root) {
  52                 mutex_lock(&root->fs_info->alloc_mutex);
  53         }
  54 }
  55
  56 void maybe_unlock_mutex(struct btrfs_root *root)
  57 {
  58         if (root != root->fs_info->extent_root &&
  59             root != root->fs_info->chunk_root &&
  60             root != root->fs_info->dev_root) {
  61                 mutex_unlock(&root->fs_info->alloc_mutex);
  62         }
  63 }
  64
  65 static int cache_block_group(struct btrfs_root *root,
  66                              struct btrfs_block_group_cache *block_group)
  67 {
  68         struct btrfs_path *path;
  69         int ret;
  70         struct btrfs_key key;
  71         struct extent_buffer *leaf;
  72         struct extent_io_tree *free_space_cache;
  73         int slot;
  74         u64 last = 0;
  75         u64 hole_size;
  76         u64 first_free;
  77         int found = 0;
  78
  79         if (!block_group)
  80                 return 0;
  81
  82         root = root->fs_info->extent_root;
  83         free_space_cache = &root->fs_info->free_space_cache;
  84
  85         if (block_group->cached)
  86                 return 0;
  87
  88         path = btrfs_alloc_path();
  89         if (!path)
  90                 return -ENOMEM;
  91
  92         path->reada = 2;
  93         /*
  94          * we get into deadlocks with paths held by callers of this function.
  95          * since the alloc_mutex is protecting things right now, just
  96          * skip the locking here
  97          */
  98         path->skip_locking = 1;
  99         first_free = block_group->key.objectid;
 100         key.objectid = block_group->key.objectid;
 101         key.offset = 0;
 102         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 103         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 104         if (ret < 0)
 105                 return ret;
 106         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
 107         if (ret < 0)
 108                 return ret;
 109         if (ret == 0) {
 110                 leaf = path->nodes[0];
 111                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 112                 if (key.objectid + key.offset > first_free)
 113                         first_free = key.objectid + key.offset;
 114         }
 115         while(1) {
 116                 leaf = path->nodes[0];
 117                 slot = path->slots[0];
 118                 if (slot >= btrfs_header_nritems(leaf)) {
 119                         ret = btrfs_next_leaf(root, path);
 120                         if (ret < 0)
 121                                 goto err;
 122                         if (ret == 0) {
 123                                 continue;
 124                         } else {
 125                                 break;
 126                         }
 127                 }
 128                 btrfs_item_key_to_cpu(leaf, &key, slot);
 129                 if (key.objectid < block_group->key.objectid) {
 130                         goto next;
 131                 }
 132                 if (key.objectid >= block_group->key.objectid +
 133                     block_group->key.offset) {
 134                         break;
 135                 }
 136
 137                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 138                         if (!found) {
 139                                 last = first_free;
 140                                 found = 1;
 141                         }
 142                         if (key.objectid > last) {
 143                                 hole_size = key.objectid - last;
 144                                 set_extent_dirty(free_space_cache, last,
 145                                                  last + hole_size - 1,
 146                                                  GFP_NOFS);
 147                         }
 148                         last = key.objectid + key.offset;
 149                 }
 150 next:
 151                 path->slots[0]++;
 152         }
 153
 154         if (!found)
 155                 last = first_free;
 156         if (block_group->key.objectid +
 157             block_group->key.offset > last) {
 158                 hole_size = block_group->key.objectid +
 159                         block_group->key.offset - last;
 160                 set_extent_dirty(free_space_cache, last,
 161                                  last + hole_size - 1, GFP_NOFS);
 162         }
 163         block_group->cached = 1;
 164 err:
 165         btrfs_free_path(path);
 166         return 0;
 167 }
 168
 169 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 170                                                        btrfs_fs_info *info,
 171                                                          u64 bytenr)
 172 {
 173         struct extent_io_tree *block_group_cache;
 174         struct btrfs_block_group_cache *block_group = NULL;
 175         u64 ptr;
 176         u64 start;
 177         u64 end;
 178         int ret;
 179
 180         bytenr = max_t(u64, bytenr,
 181                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 182         block_group_cache = &info->block_group_cache;
 183         ret = find_first_extent_bit(block_group_cache,
 184                                     bytenr, &start, &end,
 185                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 186                                     BLOCK_GROUP_SYSTEM);
 187         if (ret) {
 188                 return NULL;
 189         }
 190         ret = get_state_private(block_group_cache, start, &ptr);
 191         if (ret)
 192                 return NULL;
 193
 194         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 195         return block_group;
 196 }
 197
 198 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 199                                                          btrfs_fs_info *info,
 200                                                          u64 bytenr)
 201 {
 202         struct extent_io_tree *block_group_cache;
 203         struct btrfs_block_group_cache *block_group = NULL;
 204         u64 ptr;
 205         u64 start;
 206         u64 end;
 207         int ret;
 208
 209         bytenr = max_t(u64, bytenr,
 210                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 211         block_group_cache = &info->block_group_cache;
 212         ret = find_first_extent_bit(block_group_cache,
 213                                     bytenr, &start, &end,
 214                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
 215                                     BLOCK_GROUP_SYSTEM);
 216         if (ret) {
 217                 return NULL;
 218         }
 219         ret = get_state_private(block_group_cache, start, &ptr);
 220         if (ret)
 221                 return NULL;
 222
 223         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 224         if (block_group->key.objectid <= bytenr && bytenr <
 225             block_group->key.objectid + block_group->key.offset)
 226                 return block_group;
 227         return NULL;
 228 }
 229
 230 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 231 {
 232         return (cache->flags & bits) == bits;
 233 }
 234
 235 static int noinline find_search_start(struct btrfs_root *root,
 236                               struct btrfs_block_group_cache **cache_ret,
 237                               u64 *start_ret, u64 num, int data)
 238 {
 239         int ret;
 240         struct btrfs_block_group_cache *cache = *cache_ret;
 241         struct extent_io_tree *free_space_cache;
 242         struct extent_state *state;
 243         u64 last;
 244         u64 start = 0;
 245         u64 cache_miss = 0;
 246         u64 total_fs_bytes;
 247         u64 search_start = *start_ret;
 248         int wrapped = 0;
 249
 250         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
 251         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 252         free_space_cache = &root->fs_info->free_space_cache;
 253
 254         if (!cache)
 255                 goto out;
 256
 257 again:
 258         ret = cache_block_group(root, cache);
 259         if (ret) {
 260                 goto out;
 261         }
 262
 263         last = max(search_start, cache->key.objectid);
 264         if (!block_group_bits(cache, data) || cache->ro)
 265                 goto new_group;
 266
 267         spin_lock_irq(&free_space_cache->lock);
 268         state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
 269         while(1) {
 270                 if (!state) {
 271                         if (!cache_miss)
 272                                 cache_miss = last;
 273                         spin_unlock_irq(&free_space_cache->lock);
 274                         goto new_group;
 275                 }
 276
 277                 start = max(last, state->start);
 278                 last = state->end + 1;
 279                 if (last - start < num) {
 280                         do {
 281                                 state = extent_state_next(state);
 282                         } while(state && !(state->state & EXTENT_DIRTY));
 283                         continue;
 284                 }
 285                 spin_unlock_irq(&free_space_cache->lock);
 286                 if (cache->ro) {
 287                         goto new_group;
 288                 }
 289                 if (start + num > cache->key.objectid + cache->key.offset)
 290                         goto new_group;
 291                 if (!block_group_bits(cache, data)) {
 292                         printk("block group bits don't match %Lu %d\n", cache->flags, data);
 293                 }
 294                 *start_ret = start;
 295                 return 0;
 296         }
 297 out:
 298         cache = btrfs_lookup_block_group(root->fs_info, search_start);
 299         if (!cache) {
 300                 printk("Unable to find block group for %Lu\n", search_start);
 301                 WARN_ON(1);
 302         }
 303         return -ENOSPC;
 304
 305 new_group:
 306         last = cache->key.objectid + cache->key.offset;
 307 wrapped:
 308         cache = btrfs_lookup_first_block_group(root->fs_info, last);
 309         if (!cache || cache->key.objectid >= total_fs_bytes) {
 310 no_cache:
 311                 if (!wrapped) {
 312                         wrapped = 1;
 313                         last = search_start;
 314                         goto wrapped;
 315                 }
 316                 goto out;
 317         }
 318         if (cache_miss && !cache->cached) {
 319                 cache_block_group(root, cache);
 320                 last = cache_miss;
 321                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 322         }
 323         cache_miss = 0;
 324         cache = btrfs_find_block_group(root, cache, last, data, 0);
 325         if (!cache)
 326                 goto no_cache;
 327         *cache_ret = cache;
 328         goto again;
 329 }
 330
 331 static u64 div_factor(u64 num, int factor)
 332 {
 333         if (factor == 10)
 334                 return num;
 335         num *= factor;
 336         do_div(num, 10);
 337         return num;
 338 }
 339
 340 static int block_group_state_bits(u64 flags)
 341 {
 342         int bits = 0;
 343         if (flags & BTRFS_BLOCK_GROUP_DATA)
 344                 bits |= BLOCK_GROUP_DATA;
 345         if (flags & BTRFS_BLOCK_GROUP_METADATA)
 346                 bits |= BLOCK_GROUP_METADATA;
 347         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
 348                 bits |= BLOCK_GROUP_SYSTEM;
 349         return bits;
 350 }
 351
 352 static struct btrfs_block_group_cache *
 353 __btrfs_find_block_group(struct btrfs_root *root,
 354                          struct btrfs_block_group_cache *hint,
 355                          u64 search_start, int data, int owner)
 356 {
 357         struct btrfs_block_group_cache *cache;
 358         struct extent_io_tree *block_group_cache;
 359         struct btrfs_block_group_cache *found_group = NULL;
 360         struct btrfs_fs_info *info = root->fs_info;
 361         u64 used;
 362         u64 last = 0;
 363         u64 start;
 364         u64 end;
 365         u64 free_check;
 366         u64 ptr;
 367         int bit;
 368         int ret;
 369         int full_search = 0;
 370         int factor = 10;
 371         int wrapped = 0;
 372
 373         block_group_cache = &info->block_group_cache;
 374
 375         if (data & BTRFS_BLOCK_GROUP_METADATA)
 376                 factor = 9;
 377
 378         bit = block_group_state_bits(data);
 379
 380         if (search_start) {
 381                 struct btrfs_block_group_cache *shint;
 382                 shint = btrfs_lookup_first_block_group(info, search_start);
 383                 if (shint && block_group_bits(shint, data) && !shint->ro) {
 384                         spin_lock(&shint->lock);
 385                         used = btrfs_block_group_used(&shint->item);
 386                         if (used + shint->pinned <
 387                             div_factor(shint->key.offset, factor)) {
 388                                 spin_unlock(&shint->lock);
 389                                 return shint;
 390                         }
 391                         spin_unlock(&shint->lock);
 392                 }
 393         }
 394         if (hint && !hint->ro && block_group_bits(hint, data)) {
 395                 spin_lock(&hint->lock);
 396                 used = btrfs_block_group_used(&hint->item);
 397                 if (used + hint->pinned <
 398                     div_factor(hint->key.offset, factor)) {
 399                         spin_unlock(&hint->lock);
 400                         return hint;
 401                 }
 402                 spin_unlock(&hint->lock);
 403                 last = hint->key.objectid + hint->key.offset;
 404         } else {
 405                 if (hint)
 406                         last = max(hint->key.objectid, search_start);
 407                 else
 408                         last = search_start;
 409         }
 410 again:
 411         while(1) {
 412                 ret = find_first_extent_bit(block_group_cache, last,
 413                                             &start, &end, bit);
 414                 if (ret)
 415                         break;
 416
 417                 ret = get_state_private(block_group_cache, start, &ptr);
 418                 if (ret) {
 419                         last = end + 1;
 420                         continue;
 421                 }
 422
 423                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
 424                 spin_lock(&cache->lock);
 425                 last = cache->key.objectid + cache->key.offset;
 426                 used = btrfs_block_group_used(&cache->item);
 427
 428                 if (!cache->ro && block_group_bits(cache, data)) {
 429                         free_check = div_factor(cache->key.offset, factor);
 430                         if (used + cache->pinned < free_check) {
 431                                 found_group = cache;
 432                                 spin_unlock(&cache->lock);
 433                                 goto found;
 434                         }
 435                 }
 436                 spin_unlock(&cache->lock);
 437                 cond_resched();
 438         }
 439         if (!wrapped) {
 440                 last = search_start;
 441                 wrapped = 1;
 442                 goto again;
 443         }
 444         if (!full_search && factor < 10) {
 445                 last = search_start;
 446                 full_search = 1;
 447                 factor = 10;
 448                 goto again;
 449         }
 450 found:
 451         return found_group;
 452 }
 453
 454 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 455                                                  struct btrfs_block_group_cache
 456                                                  *hint, u64 search_start,
 457                                                  int data, int owner)
 458 {
 459
 460         struct btrfs_block_group_cache *ret;
 461         ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
 462         return ret;
 463 }
 464 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
 465                            u64 owner, u64 owner_offset)
 466 {
 467         u32 high_crc = ~(u32)0;
 468         u32 low_crc = ~(u32)0;
 469         __le64 lenum;
 470         lenum = cpu_to_le64(root_objectid);
 471         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
 472         lenum = cpu_to_le64(ref_generation);
 473         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 474         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
 475                 lenum = cpu_to_le64(owner);
 476                 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 477                 lenum = cpu_to_le64(owner_offset);
 478                 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 479         }
 480         return ((u64)high_crc << 32) | (u64)low_crc;
 481 }
 482
 483 static int match_extent_ref(struct extent_buffer *leaf,
 484                             struct btrfs_extent_ref *disk_ref,
 485                             struct btrfs_extent_ref *cpu_ref)
 486 {
 487         int ret;
 488         int len;
 489
 490         if (cpu_ref->objectid)
 491                 len = sizeof(*cpu_ref);
 492         else
 493                 len = 2 * sizeof(u64);
 494         ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
 495                                    len);
 496         return ret == 0;
 497 }
 498
 499 /* simple helper to search for an existing extent at a given offset */
 500 int btrfs_lookup_extent(struct btrfs_root *root, struct btrfs_path *path,
 501                         u64 start, u64 len)
 502 {
 503         int ret;
 504         struct btrfs_key key;
 505
 506         maybe_lock_mutex(root);
 507         key.objectid = start;
 508         key.offset = len;
 509         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 510         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 511                                 0, 0);
 512         maybe_unlock_mutex(root);
 513         return ret;
 514 }
 515
 516 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 517                                           struct btrfs_root *root,
 518                                           struct btrfs_path *path, u64 bytenr,
 519                                           u64 root_objectid,
 520                                           u64 ref_generation, u64 owner,
 521                                           u64 owner_offset, int del)
 522 {
 523         u64 hash;
 524         struct btrfs_key key;
 525         struct btrfs_key found_key;
 526         struct btrfs_extent_ref ref;
 527         struct extent_buffer *leaf;
 528         struct btrfs_extent_ref *disk_ref;
 529         int ret;
 530         int ret2;
 531
 532         btrfs_set_stack_ref_root(&ref, root_objectid);
 533         btrfs_set_stack_ref_generation(&ref, ref_generation);
 534         btrfs_set_stack_ref_objectid(&ref, owner);
 535         btrfs_set_stack_ref_offset(&ref, owner_offset);
 536
 537         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 538                                owner_offset);
 539         key.offset = hash;
 540         key.objectid = bytenr;
 541         key.type = BTRFS_EXTENT_REF_KEY;
 542
 543         while (1) {
 544                 ret = btrfs_search_slot(trans, root, &key, path,
 545                                         del ? -1 : 0, del);
 546                 if (ret < 0)
 547                         goto out;
 548                 leaf = path->nodes[0];
 549                 if (ret != 0) {
 550                         u32 nritems = btrfs_header_nritems(leaf);
 551                         if (path->slots[0] >= nritems) {
 552                                 ret2 = btrfs_next_leaf(root, path);
 553                                 if (ret2)
 554                                         goto out;
 555                                 leaf = path->nodes[0];
 556                         }
 557                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 558                         if (found_key.objectid != bytenr ||
 559                             found_key.type != BTRFS_EXTENT_REF_KEY)
 560                                 goto out;
 561                         key.offset = found_key.offset;
 562                         if (del) {
 563                                 btrfs_release_path(root, path);
 564                                 continue;
 565                         }
 566                 }
 567                 disk_ref = btrfs_item_ptr(path->nodes[0],
 568                                           path->slots[0],
 569                                           struct btrfs_extent_ref);
 570                 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
 571                         ret = 0;
 572                         goto out;
 573                 }
 574                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 575                 key.offset = found_key.offset + 1;
 576                 btrfs_release_path(root, path);
 577         }
 578 out:
 579         return ret;
 580 }
 581
 582 /*
 583  * Back reference rules.  Back refs have three main goals:
 584  *
 585  * 1) differentiate between all holders of references to an extent so that
 586  *    when a reference is dropped we can make sure it was a valid reference
 587  *    before freeing the extent.
 588  *
 589  * 2) Provide enough information to quickly find the holders of an extent
 590  *    if we notice a given block is corrupted or bad.
 591  *
 592  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 593  *    maintenance.  This is actually the same as #2, but with a slightly
 594  *    different use case.
 595  *
 596  * File extents can be referenced by:
 597  *
 598  * - multiple snapshots, subvolumes, or different generations in one subvol
 599  * - different files inside a single subvolume (in theory, not implemented yet)
 600  * - different offsets inside a file (bookend extents in file.c)
 601  *
 602  * The extent ref structure has fields for:
 603  *
 604  * - Objectid of the subvolume root
 605  * - Generation number of the tree holding the reference
 606  * - objectid of the file holding the reference
 607  * - offset in the file corresponding to the key holding the reference
 608  *
 609  * When a file extent is allocated the fields are filled in:
 610  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 611  *
 612  * When a leaf is cow'd new references are added for every file extent found
 613  * in the leaf.  It looks the same as the create case, but trans->transid
 614  * will be different when the block is cow'd.
 615  *
 616  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 617  *
 618  * When a file extent is removed either during snapshot deletion or file
 619  * truncation, the corresponding back reference is found
 620  * by searching for:
 621  *
 622  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 623  *      inode objectid, offset in file)
 624  *
 625  * Btree extents can be referenced by:
 626  *
 627  * - Different subvolumes
 628  * - Different generations of the same subvolume
 629  *
 630  * Storing sufficient information for a full reverse mapping of a btree
 631  * block would require storing the lowest key of the block in the backref,
 632  * and it would require updating that lowest key either before write out or
 633  * every time it changed.  Instead, the objectid of the lowest key is stored
 634  * along with the level of the tree block.  This provides a hint
 635  * about where in the btree the block can be found.  Searches through the
 636  * btree only need to look for a pointer to that block, so they stop one
 637  * level higher than the level recorded in the backref.
 638  *
 639  * Some btrees do not do reference counting on their extents.  These
 640  * include the extent tree and the tree of tree roots.  Backrefs for these
 641  * trees always have a generation of zero.
 642  *
 643  * When a tree block is created, back references are inserted:
 644  *
 645  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
 646  *
 647  * When a tree block is cow'd in a reference counted root,
 648  * new back references are added for all the blocks it points to.
 649  * These are of the form (trans->transid will have increased since creation):
 650  *
 651  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
 652  *
 653  * Because the lowest_key_objectid and the level are just hints
 654  * they are not used when backrefs are deleted.  When a backref is deleted:
 655  *
 656  * if backref was for a tree root:
 657  *     root_objectid = root->root_key.objectid
 658  * else
 659  *     root_objectid = btrfs_header_owner(parent)
 660  *
 661  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
 662  *
 663  * Back Reference Key hashing:
 664  *
 665  * Back references have four fields, each 64 bits long.  Unfortunately,
 666  * This is hashed into a single 64 bit number and placed into the key offset.
 667  * The key objectid corresponds to the first byte in the extent, and the
 668  * key type is set to BTRFS_EXTENT_REF_KEY
 669  */
 670 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
 671                                  struct btrfs_root *root,
 672                                  struct btrfs_path *path, u64 bytenr,
 673                                  u64 root_objectid, u64 ref_generation,
 674                                  u64 owner, u64 owner_offset)
 675 {
 676         u64 hash;
 677         struct btrfs_key key;
 678         struct btrfs_extent_ref ref;
 679         struct btrfs_extent_ref *disk_ref;
 680         int ret;
 681
 682         btrfs_set_stack_ref_root(&ref, root_objectid);
 683         btrfs_set_stack_ref_generation(&ref, ref_generation);
 684         btrfs_set_stack_ref_objectid(&ref, owner);
 685         btrfs_set_stack_ref_offset(&ref, owner_offset);
 686
 687         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 688                                owner_offset);
 689         key.offset = hash;
 690         key.objectid = bytenr;
 691         key.type = BTRFS_EXTENT_REF_KEY;
 692
 693         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
 694         while (ret == -EEXIST) {
 695                 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 696                                           struct btrfs_extent_ref);
 697                 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
 698                         goto out;
 699                 key.offset++;
 700                 btrfs_release_path(root, path);
 701                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 702                                               sizeof(ref));
 703         }
 704         if (ret)
 705                 goto out;
 706         disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 707                                   struct btrfs_extent_ref);
 708         write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
 709                             sizeof(ref));
 710         btrfs_mark_buffer_dirty(path->nodes[0]);
 711 out:
 712         btrfs_release_path(root, path);
 713         return ret;
 714 }
 715
 716 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 717                                 struct btrfs_root *root,
 718                                 u64 bytenr, u64 num_bytes,
 719                                 u64 root_objectid, u64 ref_generation,
 720                                 u64 owner, u64 owner_offset)
 721 {
 722         struct btrfs_path *path;
 723         int ret;
 724         struct btrfs_key key;
 725         struct extent_buffer *l;
 726         struct btrfs_extent_item *item;
 727         u32 refs;
 728
 729         WARN_ON(num_bytes < root->sectorsize);
 730         path = btrfs_alloc_path();
 731         if (!path)
 732                 return -ENOMEM;
 733
 734         path->reada = 1;
 735         key.objectid = bytenr;
 736         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 737         key.offset = num_bytes;
 738         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 739                                 0, 1);
 740         if (ret < 0)
 741                 return ret;
 742         if (ret != 0) {
 743                 BUG();
 744         }
 745         BUG_ON(ret != 0);
 746         l = path->nodes[0];
 747         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 748         refs = btrfs_extent_refs(l, item);
 749         btrfs_set_extent_refs(l, item, refs + 1);
 750         btrfs_mark_buffer_dirty(path->nodes[0]);
 751
 752         btrfs_release_path(root->fs_info->extent_root, path);
 753
 754         path->reada = 1;
 755         ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
 756                                           path, bytenr, root_objectid,
 757                                           ref_generation, owner, owner_offset);
 758         BUG_ON(ret);
 759         finish_current_insert(trans, root->fs_info->extent_root);
 760         del_pending_extents(trans, root->fs_info->extent_root);
 761
 762         btrfs_free_path(path);
 763         return 0;
 764 }
 765
 766 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 767                                 struct btrfs_root *root,
 768                                 u64 bytenr, u64 num_bytes,
 769                                 u64 root_objectid, u64 ref_generation,
 770                                 u64 owner, u64 owner_offset)
 771 {
 772         int ret;
 773
 774         mutex_lock(&root->fs_info->alloc_mutex);
 775         ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
 776                                      root_objectid, ref_generation,
 777                                      owner, owner_offset);
 778         mutex_unlock(&root->fs_info->alloc_mutex);
 779         return ret;
 780 }
 781
 782 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 783                          struct btrfs_root *root)
 784 {
 785         finish_current_insert(trans, root->fs_info->extent_root);
 786         del_pending_extents(trans, root->fs_info->extent_root);
 787         return 0;
 788 }
 789
 790 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 791                              struct btrfs_root *root, u64 bytenr,
 792                              u64 num_bytes, u32 *refs)
 793 {
 794         struct btrfs_path *path;
 795         int ret;
 796         struct btrfs_key key;
 797         struct extent_buffer *l;
 798         struct btrfs_extent_item *item;
 799
 800         WARN_ON(num_bytes < root->sectorsize);
 801         path = btrfs_alloc_path();
 802         path->reada = 1;
 803         key.objectid = bytenr;
 804         key.offset = num_bytes;
 805         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 806         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 807                                 0, 0);
 808         if (ret < 0)
 809                 goto out;
 810         if (ret != 0) {
 811                 btrfs_print_leaf(root, path->nodes[0]);
 812                 printk("failed to find block number %Lu\n", bytenr);
 813                 BUG();
 814         }
 815         l = path->nodes[0];
 816         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 817         *refs = btrfs_extent_refs(l, item);
 818 out:
 819         btrfs_free_path(path);
 820         return 0;
 821 }
 822
 823
 824 static int get_reference_status(struct btrfs_root *root, u64 bytenr,
 825                                 u64 parent_gen, u64 ref_objectid,
 826                                 u64 *min_generation, u32 *ref_count)
 827 {
 828         struct btrfs_root *extent_root = root->fs_info->extent_root;
 829         struct btrfs_path *path;
 830         struct extent_buffer *leaf;
 831         struct btrfs_extent_ref *ref_item;
 832         struct btrfs_key key;
 833         struct btrfs_key found_key;
 834         u64 root_objectid = root->root_key.objectid;
 835         u64 ref_generation;
 836         u32 nritems;
 837         int ret;
 838
 839         key.objectid = bytenr;
 840         key.offset = 0;
 841         key.type = BTRFS_EXTENT_ITEM_KEY;
 842
 843         path = btrfs_alloc_path();
 844         mutex_lock(&root->fs_info->alloc_mutex);
 845         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 846         if (ret < 0)
 847                 goto out;
 848         BUG_ON(ret == 0);
 849
 850         leaf = path->nodes[0];
 851         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 852
 853         if (found_key.objectid != bytenr ||
 854             found_key.type != BTRFS_EXTENT_ITEM_KEY) {
 855                 ret = 1;
 856                 goto out;
 857         }
 858
 859         *ref_count = 0;
 860         *min_generation = (u64)-1;
 861
 862         while (1) {
 863                 leaf = path->nodes[0];
 864                 nritems = btrfs_header_nritems(leaf);
 865                 if (path->slots[0] >= nritems) {
 866                         ret = btrfs_next_leaf(extent_root, path);
 867                         if (ret < 0)
 868                                 goto out;
 869                         if (ret == 0)
 870                                 continue;
 871                         break;
 872                 }
 873                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 874                 if (found_key.objectid != bytenr)
 875                         break;
 876
 877                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 878                         path->slots[0]++;
 879                         continue;
 880                 }
 881
 882                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
 883                                           struct btrfs_extent_ref);
 884                 ref_generation = btrfs_ref_generation(leaf, ref_item);
 885                 /*
 886                  * For (parent_gen > 0 && parent_gen > ref_gen):
 887                  *
 888                  * we reach here through the oldest root, therefore
 889                  * all other reference from same snapshot should have
 890                  * a larger generation.
 891                  */
 892                 if ((root_objectid != btrfs_ref_root(leaf, ref_item)) ||
 893                     (parent_gen > 0 && parent_gen > ref_generation) ||
 894                     (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
 895                      ref_objectid != btrfs_ref_objectid(leaf, ref_item))) {
 896                         if (ref_count)
 897                                 *ref_count = 2;
 898                         break;
 899                 }
 900
 901                 *ref_count = 1;
 902                 if (*min_generation > ref_generation)
 903                         *min_generation = ref_generation;
 904
 905                 path->slots[0]++;
 906         }
 907         ret = 0;
 908 out:
 909         mutex_unlock(&root->fs_info->alloc_mutex);
 910         btrfs_free_path(path);
 911         return ret;
 912 }
 913
 914 int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans,
 915                            struct btrfs_root *root,
 916                            struct btrfs_key *key, u64 bytenr)
 917 {
 918         struct btrfs_root *old_root;
 919         struct btrfs_path *path = NULL;
 920         struct extent_buffer *eb;
 921         struct btrfs_file_extent_item *item;
 922         u64 ref_generation;
 923         u64 min_generation;
 924         u64 extent_start;
 925         u32 ref_count;
 926         int level;
 927         int ret;
 928
 929         BUG_ON(trans == NULL);
 930         BUG_ON(key->type != BTRFS_EXTENT_DATA_KEY);
 931         ret = get_reference_status(root, bytenr, 0, key->objectid,
 932                                    &min_generation, &ref_count);
 933         if (ret)
 934                 return ret;
 935
 936         if (ref_count != 1)
 937                 return 1;
 938
 939         old_root = root->dirty_root->root;
 940         ref_generation = old_root->root_key.offset;
 941
 942         /* all references are created in running transaction */
 943         if (min_generation > ref_generation) {
 944                 ret = 0;
 945                 goto out;
 946         }
 947
 948         path = btrfs_alloc_path();
 949         if (!path) {
 950                 ret = -ENOMEM;
 951                 goto out;
 952         }
 953
 954         path->skip_locking = 1;
 955         /* if no item found, the extent is referenced by other snapshot */
 956         ret = btrfs_search_slot(NULL, old_root, key, path, 0, 0);
 957         if (ret)
 958                 goto out;
 959
 960         eb = path->nodes[0];
 961         item = btrfs_item_ptr(eb, path->slots[0],
 962                               struct btrfs_file_extent_item);
 963         if (btrfs_file_extent_type(eb, item) != BTRFS_FILE_EXTENT_REG ||
 964             btrfs_file_extent_disk_bytenr(eb, item) != bytenr) {
 965                 ret = 1;
 966                 goto out;
 967         }
 968
 969         for (level = BTRFS_MAX_LEVEL - 1; level >= -1; level--) {
 970                 if (level >= 0) {
 971                         eb = path->nodes[level];
 972                         if (!eb)
 973                                 continue;
 974                         extent_start = eb->start;
 975                 } else
 976                         extent_start = bytenr;
 977
 978                 ret = get_reference_status(root, extent_start, ref_generation,
 979                                            0, &min_generation, &ref_count);
 980                 if (ret)
 981                         goto out;
 982
 983                 if (ref_count != 1) {
 984                         ret = 1;
 985                         goto out;
 986                 }
 987                 if (level >= 0)
 988                         ref_generation = btrfs_header_generation(eb);
 989         }
 990         ret = 0;
 991 out:
 992         if (path)
 993                 btrfs_free_path(path);
 994         return ret;
 995 }
 996
 997 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 998                   struct extent_buffer *buf, int cache_ref)
 999 {
1000         u64 bytenr;
1001         u32 nritems;
1002         struct btrfs_key key;
1003         struct btrfs_file_extent_item *fi;
1004         int i;
1005         int level;
1006         int ret;
1007         int faili;
1008         int nr_file_extents = 0;
1009
1010         if (!root->ref_cows)
1011                 return 0;
1012
1013         level = btrfs_header_level(buf);
1014         nritems = btrfs_header_nritems(buf);
1015         for (i = 0; i < nritems; i++) {
1016                 cond_resched();
1017                 if (level == 0) {
1018                         u64 disk_bytenr;
1019                         btrfs_item_key_to_cpu(buf, &key, i);
1020                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1021                                 continue;
1022                         fi = btrfs_item_ptr(buf, i,
1023                                             struct btrfs_file_extent_item);
1024                         if (btrfs_file_extent_type(buf, fi) ==
1025                             BTRFS_FILE_EXTENT_INLINE)
1026                                 continue;
1027                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1028                         if (disk_bytenr == 0)
1029                                 continue;
1030
1031                         if (buf != root->commit_root)
1032                                 nr_file_extents++;
1033
1034                         mutex_lock(&root->fs_info->alloc_mutex);
1035                         ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr,
1036                                     btrfs_file_extent_disk_num_bytes(buf, fi),
1037                                     root->root_key.objectid, trans->transid,
1038                                     key.objectid, key.offset);
1039                         mutex_unlock(&root->fs_info->alloc_mutex);
1040                         if (ret) {
1041                                 faili = i;
1042                                 WARN_ON(1);
1043                                 goto fail;
1044                         }
1045                 } else {
1046                         bytenr = btrfs_node_blockptr(buf, i);
1047                         btrfs_node_key_to_cpu(buf, &key, i);
1048
1049                         mutex_lock(&root->fs_info->alloc_mutex);
1050                         ret = __btrfs_inc_extent_ref(trans, root, bytenr,
1051                                            btrfs_level_size(root, level - 1),
1052                                            root->root_key.objectid,
1053                                            trans->transid,
1054                                            level - 1, key.objectid);
1055                         mutex_unlock(&root->fs_info->alloc_mutex);
1056                         if (ret) {
1057                                 faili = i;
1058                                 WARN_ON(1);
1059                                 goto fail;
1060                         }
1061                 }
1062         }
1063         /* cache orignal leaf block's references */
1064         if (level == 0 && cache_ref && buf != root->commit_root) {
1065                 struct btrfs_leaf_ref *ref;
1066                 struct btrfs_extent_info *info;
1067
1068                 ref = btrfs_alloc_leaf_ref(root, nr_file_extents);
1069                 if (!ref) {
1070                         WARN_ON(1);
1071                         goto out;
1072                 }
1073
1074                 ref->root_gen = root->root_key.offset;
1075                 ref->bytenr = buf->start;
1076                 ref->owner = btrfs_header_owner(buf);
1077                 ref->generation = btrfs_header_generation(buf);
1078                 ref->nritems = nr_file_extents;
1079                 info = ref->extents;
1080
1081                 for (i = 0; nr_file_extents > 0 && i < nritems; i++) {
1082                         u64 disk_bytenr;
1083                         btrfs_item_key_to_cpu(buf, &key, i);
1084                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1085                                 continue;
1086                         fi = btrfs_item_ptr(buf, i,
1087                                             struct btrfs_file_extent_item);
1088                         if (btrfs_file_extent_type(buf, fi) ==
1089                             BTRFS_FILE_EXTENT_INLINE)
1090                                 continue;
1091                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1092                         if (disk_bytenr == 0)
1093                                 continue;
1094
1095                         info->bytenr = disk_bytenr;
1096                         info->num_bytes =
1097                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1098                         info->objectid = key.objectid;
1099                         info->offset = key.offset;
1100                         info++;
1101                 }
1102
1103                 BUG_ON(!root->ref_tree);
1104                 ret = btrfs_add_leaf_ref(root, ref);
1105                 WARN_ON(ret);
1106                 btrfs_free_leaf_ref(root, ref);
1107         }
1108 out:
1109         return 0;
1110 fail:
1111         WARN_ON(1);
1112 #if 0
1113         for (i =0; i < faili; i++) {
1114                 if (level == 0) {
1115                         u64 disk_bytenr;
1116                         btrfs_item_key_to_cpu(buf, &key, i);
1117                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1118                                 continue;
1119                         fi = btrfs_item_ptr(buf, i,
1120                                             struct btrfs_file_extent_item);
1121                         if (btrfs_file_extent_type(buf, fi) ==
1122                             BTRFS_FILE_EXTENT_INLINE)
1123                                 continue;
1124                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1125                         if (disk_bytenr == 0)
1126                                 continue;
1127                         err = btrfs_free_extent(trans, root, disk_bytenr,
1128                                     btrfs_file_extent_disk_num_bytes(buf,
1129                                                                       fi), 0);
1130                         BUG_ON(err);
1131                 } else {
1132                         bytenr = btrfs_node_blockptr(buf, i);
1133                         err = btrfs_free_extent(trans, root, bytenr,
1134                                         btrfs_level_size(root, level - 1), 0);
1135                         BUG_ON(err);
1136                 }
1137         }
1138 #endif
1139         return ret;
1140 }
1141
1142 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1143                                  struct btrfs_root *root,
1144                                  struct btrfs_path *path,
1145                                  struct btrfs_block_group_cache *cache)
1146 {
1147         int ret;
1148         int pending_ret;
1149         struct btrfs_root *extent_root = root->fs_info->extent_root;
1150         unsigned long bi;
1151         struct extent_buffer *leaf;
1152
1153         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1154         if (ret < 0)
1155                 goto fail;
1156         BUG_ON(ret);
1157
1158         leaf = path->nodes[0];
1159         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1160         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1161         btrfs_mark_buffer_dirty(leaf);
1162         btrfs_release_path(extent_root, path);
1163 fail:
1164         finish_current_insert(trans, extent_root);
1165         pending_ret = del_pending_extents(trans, extent_root);
1166         if (ret)
1167                 return ret;
1168         if (pending_ret)
1169                 return pending_ret;
1170         return 0;
1171
1172 }
1173
1174 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1175                                    struct btrfs_root *root)
1176 {
1177         struct extent_io_tree *block_group_cache;
1178         struct btrfs_block_group_cache *cache;
1179         int ret;
1180         int err = 0;
1181         int werr = 0;
1182         struct btrfs_path *path;
1183         u64 last = 0;
1184         u64 start;
1185         u64 end;
1186         u64 ptr;
1187
1188         block_group_cache = &root->fs_info->block_group_cache;
1189         path = btrfs_alloc_path();
1190         if (!path)
1191                 return -ENOMEM;
1192
1193         mutex_lock(&root->fs_info->alloc_mutex);
1194         while(1) {
1195                 ret = find_first_extent_bit(block_group_cache, last,
1196                                             &start, &end, BLOCK_GROUP_DIRTY);
1197                 if (ret)
1198                         break;
1199
1200                 last = end + 1;
1201                 ret = get_state_private(block_group_cache, start, &ptr);
1202                 if (ret)
1203                         break;
1204                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1205                 err = write_one_cache_group(trans, root,
1206                                             path, cache);
1207                 /*
1208                  * if we fail to write the cache group, we want
1209                  * to keep it marked dirty in hopes that a later
1210                  * write will work
1211                  */
1212                 if (err) {
1213                         werr = err;
1214                         continue;
1215                 }
1216                 clear_extent_bits(block_group_cache, start, end,
1217                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
1218         }
1219         btrfs_free_path(path);
1220         mutex_unlock(&root->fs_info->alloc_mutex);
1221         return werr;
1222 }
1223
1224 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1225                                                   u64 flags)
1226 {
1227         struct list_head *head = &info->space_info;
1228         struct list_head *cur;
1229         struct btrfs_space_info *found;
1230         list_for_each(cur, head) {
1231                 found = list_entry(cur, struct btrfs_space_info, list);
1232                 if (found->flags == flags)
1233                         return found;
1234         }
1235         return NULL;
1236
1237 }
1238
1239 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1240                              u64 total_bytes, u64 bytes_used,
1241                              struct btrfs_space_info **space_info)
1242 {
1243         struct btrfs_space_info *found;
1244
1245         found = __find_space_info(info, flags);
1246         if (found) {
1247                 found->total_bytes += total_bytes;
1248                 found->bytes_used += bytes_used;
1249                 found->full = 0;
1250                 *space_info = found;
1251                 return 0;
1252         }
1253         found = kmalloc(sizeof(*found), GFP_NOFS);
1254         if (!found)
1255                 return -ENOMEM;
1256
1257         list_add(&found->list, &info->space_info);
1258         found->flags = flags;
1259         found->total_bytes = total_bytes;
1260         found->bytes_used = bytes_used;
1261         found->bytes_pinned = 0;
1262         found->full = 0;
1263         found->force_alloc = 0;
1264         *space_info = found;
1265         return 0;
1266 }
1267
1268 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1269 {
1270         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1271                                    BTRFS_BLOCK_GROUP_RAID1 |
1272                                    BTRFS_BLOCK_GROUP_RAID10 |
1273                                    BTRFS_BLOCK_GROUP_DUP);
1274         if (extra_flags) {
1275                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1276                         fs_info->avail_data_alloc_bits |= extra_flags;
1277                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1278                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1279                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1280                         fs_info->avail_system_alloc_bits |= extra_flags;
1281         }
1282 }
1283
1284 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1285 {
1286         u64 num_devices = root->fs_info->fs_devices->num_devices;
1287
1288         if (num_devices == 1)
1289                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1290         if (num_devices < 4)
1291                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1292
1293         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1294             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1295                       BTRFS_BLOCK_GROUP_RAID10))) {
1296                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1297         }
1298
1299         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1300             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1301                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1302         }
1303
1304         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1305             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1306              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1307              (flags & BTRFS_BLOCK_GROUP_DUP)))
1308                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1309         return flags;
1310 }
1311
1312 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1313                           struct btrfs_root *extent_root, u64 alloc_bytes,
1314                           u64 flags, int force)
1315 {
1316         struct btrfs_space_info *space_info;
1317         u64 thresh;
1318         u64 start;
1319         u64 num_bytes;
1320         int ret;
1321
1322         flags = reduce_alloc_profile(extent_root, flags);
1323
1324         space_info = __find_space_info(extent_root->fs_info, flags);
1325         if (!space_info) {
1326                 ret = update_space_info(extent_root->fs_info, flags,
1327                                         0, 0, &space_info);
1328                 BUG_ON(ret);
1329         }
1330         BUG_ON(!space_info);
1331
1332         if (space_info->force_alloc) {
1333                 force = 1;
1334                 space_info->force_alloc = 0;
1335         }
1336         if (space_info->full)
1337                 goto out;
1338
1339         thresh = div_factor(space_info->total_bytes, 6);
1340         if (!force &&
1341            (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1342             thresh)
1343                 goto out;
1344
1345         mutex_lock(&extent_root->fs_info->chunk_mutex);
1346         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1347         if (ret == -ENOSPC) {
1348 printk("space info full %Lu\n", flags);
1349                 space_info->full = 1;
1350                 goto out_unlock;
1351         }
1352         BUG_ON(ret);
1353
1354         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1355                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1356         BUG_ON(ret);
1357 out_unlock:
1358         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1359 out:
1360         return 0;
1361 }
1362
1363 static int update_block_group(struct btrfs_trans_handle *trans,
1364                               struct btrfs_root *root,
1365                               u64 bytenr, u64 num_bytes, int alloc,
1366                               int mark_free)
1367 {
1368         struct btrfs_block_group_cache *cache;
1369         struct btrfs_fs_info *info = root->fs_info;
1370         u64 total = num_bytes;
1371         u64 old_val;
1372         u64 byte_in_group;
1373         u64 start;
1374         u64 end;
1375
1376         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1377         while(total) {
1378                 cache = btrfs_lookup_block_group(info, bytenr);
1379                 if (!cache) {
1380                         return -1;
1381                 }
1382                 byte_in_group = bytenr - cache->key.objectid;
1383                 WARN_ON(byte_in_group > cache->key.offset);
1384                 start = cache->key.objectid;
1385                 end = start + cache->key.offset - 1;
1386                 set_extent_bits(&info->block_group_cache, start, end,
1387                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1388
1389                 spin_lock(&cache->lock);
1390                 old_val = btrfs_block_group_used(&cache->item);
1391                 num_bytes = min(total, cache->key.offset - byte_in_group);
1392                 if (alloc) {
1393                         old_val += num_bytes;
1394                         cache->space_info->bytes_used += num_bytes;
1395                         btrfs_set_block_group_used(&cache->item, old_val);
1396                         spin_unlock(&cache->lock);
1397                 } else {
1398                         old_val -= num_bytes;
1399                         cache->space_info->bytes_used -= num_bytes;
1400                         btrfs_set_block_group_used(&cache->item, old_val);
1401                         spin_unlock(&cache->lock);
1402                         if (mark_free) {
1403                                 set_extent_dirty(&info->free_space_cache,
1404                                                  bytenr, bytenr + num_bytes - 1,
1405                                                  GFP_NOFS);
1406                         }
1407                 }
1408                 total -= num_bytes;
1409                 bytenr += num_bytes;
1410         }
1411         return 0;
1412 }
1413
1414 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1415 {
1416         u64 start;
1417         u64 end;
1418         int ret;
1419         ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1420                                     search_start, &start, &end,
1421                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1422                                     BLOCK_GROUP_SYSTEM);
1423         if (ret)
1424                 return 0;
1425         return start;
1426 }
1427
1428
1429 int btrfs_update_pinned_extents(struct btrfs_root *root,
1430                                 u64 bytenr, u64 num, int pin)
1431 {
1432         u64 len;
1433         struct btrfs_block_group_cache *cache;
1434         struct btrfs_fs_info *fs_info = root->fs_info;
1435
1436         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1437         if (pin) {
1438                 set_extent_dirty(&fs_info->pinned_extents,
1439                                 bytenr, bytenr + num - 1, GFP_NOFS);
1440         } else {
1441                 clear_extent_dirty(&fs_info->pinned_extents,
1442                                 bytenr, bytenr + num - 1, GFP_NOFS);
1443         }
1444         while (num > 0) {
1445                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1446                 if (!cache) {
1447                         u64 first = first_logical_byte(root, bytenr);
1448                         WARN_ON(first < bytenr);
1449                         len = min(first - bytenr, num);
1450                 } else {
1451                         len = min(num, cache->key.offset -
1452                                   (bytenr - cache->key.objectid));
1453                 }
1454                 if (pin) {
1455                         if (cache) {
1456                                 spin_lock(&cache->lock);
1457                                 cache->pinned += len;
1458                                 cache->space_info->bytes_pinned += len;
1459                                 spin_unlock(&cache->lock);
1460                         }
1461                         fs_info->total_pinned += len;
1462                 } else {
1463                         if (cache) {
1464                                 spin_lock(&cache->lock);
1465                                 cache->pinned -= len;
1466                                 cache->space_info->bytes_pinned -= len;
1467                                 spin_unlock(&cache->lock);
1468                         }
1469                         fs_info->total_pinned -= len;
1470                 }
1471                 bytenr += len;
1472                 num -= len;
1473         }
1474         return 0;
1475 }
1476
1477 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1478 {
1479         u64 last = 0;
1480         u64 start;
1481         u64 end;
1482         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1483         int ret;
1484
1485         while(1) {
1486                 ret = find_first_extent_bit(pinned_extents, last,
1487                                             &start, &end, EXTENT_DIRTY);
1488                 if (ret)
1489                         break;
1490                 set_extent_dirty(copy, start, end, GFP_NOFS);
1491                 last = end + 1;
1492         }
1493         return 0;
1494 }
1495
1496 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1497                                struct btrfs_root *root,
1498                                struct extent_io_tree *unpin)
1499 {
1500         u64 start;
1501         u64 end;
1502         int ret;
1503         struct extent_io_tree *free_space_cache;
1504         free_space_cache = &root->fs_info->free_space_cache;
1505
1506         mutex_lock(&root->fs_info->alloc_mutex);
1507         while(1) {
1508                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1509                                             EXTENT_DIRTY);
1510                 if (ret)
1511                         break;
1512                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1513                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1514                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1515                 if (need_resched()) {
1516                         mutex_unlock(&root->fs_info->alloc_mutex);
1517                         cond_resched();
1518                         mutex_lock(&root->fs_info->alloc_mutex);
1519                 }
1520         }
1521         mutex_unlock(&root->fs_info->alloc_mutex);
1522         return 0;
1523 }
1524
1525 static int finish_current_insert(struct btrfs_trans_handle *trans,
1526                                  struct btrfs_root *extent_root)
1527 {
1528         u64 start;
1529         u64 end;
1530         struct btrfs_fs_info *info = extent_root->fs_info;
1531         struct extent_buffer *eb;
1532         struct btrfs_path *path;
1533         struct btrfs_key ins;
1534         struct btrfs_disk_key first;
1535         struct btrfs_extent_item extent_item;
1536         int ret;
1537         int level;
1538         int err = 0;
1539
1540         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1541         btrfs_set_stack_extent_refs(&extent_item, 1);
1542         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1543         path = btrfs_alloc_path();
1544
1545         while(1) {
1546                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1547                                             &end, EXTENT_LOCKED);
1548                 if (ret)
1549                         break;
1550
1551                 ins.objectid = start;
1552                 ins.offset = end + 1 - start;
1553                 err = btrfs_insert_item(trans, extent_root, &ins,
1554                                         &extent_item, sizeof(extent_item));
1555                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1556                                   GFP_NOFS);
1557
1558                 eb = btrfs_find_create_tree_block(extent_root, ins.objectid,
1559                                            ins.offset);
1560
1561                 if (!btrfs_buffer_uptodate(eb, trans->transid))
1562                         btrfs_read_buffer(eb, trans->transid);
1563
1564                 btrfs_tree_lock(eb);
1565                 level = btrfs_header_level(eb);
1566                 if (level == 0) {
1567                         btrfs_item_key(eb, &first, 0);
1568                 } else {
1569                         btrfs_node_key(eb, &first, 0);
1570                 }
1571                 btrfs_tree_unlock(eb);
1572                 free_extent_buffer(eb);
1573                 /*
1574                  * the first key is just a hint, so the race we've created
1575                  * against reading it is fine
1576                  */
1577                 err = btrfs_insert_extent_backref(trans, extent_root, path,
1578                                           start, extent_root->root_key.objectid,
1579                                           0, level,
1580                                           btrfs_disk_key_objectid(&first));
1581                 BUG_ON(err);
1582                 if (need_resched()) {
1583                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1584                         cond_resched();
1585                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1586                 }
1587         }
1588         btrfs_free_path(path);
1589         return 0;
1590 }
1591
1592 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1593                           int is_data, int pending)
1594 {
1595         int err = 0;
1596
1597         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1598         if (!pending) {
1599                 struct extent_buffer *buf;
1600
1601                 if (is_data)
1602                         goto pinit;
1603
1604                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1605                 if (buf) {
1606                         /* we can reuse a block if it hasn't been written
1607                          * and it is from this transaction.  We can't
1608                          * reuse anything from the tree log root because
1609                          * it has tiny sub-transactions.
1610                          */
1611                         if (btrfs_buffer_uptodate(buf, 0) &&
1612                             btrfs_try_tree_lock(buf)) {
1613                                 u64 transid =
1614                                     root->fs_info->running_transaction->transid;
1615                                 u64 header_transid =
1616                                         btrfs_header_generation(buf);
1617                                 if (btrfs_header_owner(buf) !=
1618                                     BTRFS_TREE_LOG_OBJECTID &&
1619                                     header_transid == transid &&
1620                                     !btrfs_header_flag(buf,
1621                                                BTRFS_HEADER_FLAG_WRITTEN)) {
1622                                         clean_tree_block(NULL, root, buf);
1623                                         btrfs_tree_unlock(buf);
1624                                         free_extent_buffer(buf);
1625                                         return 1;
1626                                 }
1627                                 btrfs_tree_unlock(buf);
1628                         }
1629                         free_extent_buffer(buf);
1630                 }
1631 pinit:
1632                 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1633         } else {
1634                 set_extent_bits(&root->fs_info->pending_del,
1635                                 bytenr, bytenr + num_bytes - 1,
1636                                 EXTENT_LOCKED, GFP_NOFS);
1637         }
1638         BUG_ON(err < 0);
1639         return 0;
1640 }
1641
1642 /*
1643  * remove an extent from the root, returns 0 on success
1644  */
1645 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1646                          *root, u64 bytenr, u64 num_bytes,
1647                          u64 root_objectid, u64 ref_generation,
1648                          u64 owner_objectid, u64 owner_offset, int pin,
1649                          int mark_free)
1650 {
1651         struct btrfs_path *path;
1652         struct btrfs_key key;
1653         struct btrfs_fs_info *info = root->fs_info;
1654         struct btrfs_root *extent_root = info->extent_root;
1655         struct extent_buffer *leaf;
1656         int ret;
1657         int extent_slot = 0;
1658         int found_extent = 0;
1659         int num_to_del = 1;
1660         struct btrfs_extent_item *ei;
1661         u32 refs;
1662
1663         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1664         key.objectid = bytenr;
1665         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1666         key.offset = num_bytes;
1667         path = btrfs_alloc_path();
1668         if (!path)
1669                 return -ENOMEM;
1670
1671         path->reada = 1;
1672         ret = lookup_extent_backref(trans, extent_root, path,
1673                                     bytenr, root_objectid,
1674                                     ref_generation,
1675                                     owner_objectid, owner_offset, 1);
1676         if (ret == 0) {
1677                 struct btrfs_key found_key;
1678                 extent_slot = path->slots[0];
1679                 while(extent_slot > 0) {
1680                         extent_slot--;
1681                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1682                                               extent_slot);
1683                         if (found_key.objectid != bytenr)
1684                                 break;
1685                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1686                             found_key.offset == num_bytes) {
1687                                 found_extent = 1;
1688                                 break;
1689                         }
1690                         if (path->slots[0] - extent_slot > 5)
1691                                 break;
1692                 }
1693                 if (!found_extent)
1694                         ret = btrfs_del_item(trans, extent_root, path);
1695         } else {
1696                 btrfs_print_leaf(extent_root, path->nodes[0]);
1697                 WARN_ON(1);
1698                 printk("Unable to find ref byte nr %Lu root %Lu "
1699                        " gen %Lu owner %Lu offset %Lu\n", bytenr,
1700                        root_objectid, ref_generation, owner_objectid,
1701                        owner_offset);
1702         }
1703         if (!found_extent) {
1704                 btrfs_release_path(extent_root, path);
1705                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1706                 if (ret < 0)
1707                         return ret;
1708                 BUG_ON(ret);
1709                 extent_slot = path->slots[0];
1710         }
1711
1712         leaf = path->nodes[0];
1713         ei = btrfs_item_ptr(leaf, extent_slot,
1714                             struct btrfs_extent_item);
1715         refs = btrfs_extent_refs(leaf, ei);
1716         BUG_ON(refs == 0);
1717         refs -= 1;
1718         btrfs_set_extent_refs(leaf, ei, refs);
1719
1720         btrfs_mark_buffer_dirty(leaf);
1721
1722         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1723                 /* if the back ref and the extent are next to each other
1724                  * they get deleted below in one shot
1725                  */
1726                 path->slots[0] = extent_slot;
1727                 num_to_del = 2;
1728         } else if (found_extent) {
1729                 /* otherwise delete the extent back ref */
1730                 ret = btrfs_del_item(trans, extent_root, path);
1731                 BUG_ON(ret);
1732                 /* if refs are 0, we need to setup the path for deletion */
1733                 if (refs == 0) {
1734                         btrfs_release_path(extent_root, path);
1735                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1736                                                 -1, 1);
1737                         if (ret < 0)
1738                                 return ret;
1739                         BUG_ON(ret);
1740                 }
1741         }
1742
1743         if (refs == 0) {
1744                 u64 super_used;
1745                 u64 root_used;
1746 #ifdef BIO_RW_DISCARD
1747                 u64 map_length = num_bytes;
1748                 struct btrfs_multi_bio *multi = NULL;
1749 #endif
1750
1751                 if (pin) {
1752                         ret = pin_down_bytes(root, bytenr, num_bytes,
1753                              owner_objectid >= BTRFS_FIRST_FREE_OBJECTID, 0);
1754                         if (ret > 0)
1755                                 mark_free = 1;
1756                         BUG_ON(ret < 0);
1757                 }
1758
1759                 /* block accounting for super block */
1760                 spin_lock_irq(&info->delalloc_lock);
1761                 super_used = btrfs_super_bytes_used(&info->super_copy);
1762                 btrfs_set_super_bytes_used(&info->super_copy,
1763                                            super_used - num_bytes);
1764                 spin_unlock_irq(&info->delalloc_lock);
1765
1766                 /* block accounting for root item */
1767                 root_used = btrfs_root_used(&root->root_item);
1768                 btrfs_set_root_used(&root->root_item,
1769                                            root_used - num_bytes);
1770                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1771                                       num_to_del);
1772                 if (ret) {
1773                         return ret;
1774                 }
1775                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1776                                          mark_free);
1777                 BUG_ON(ret);
1778
1779 #ifdef BIO_RW_DISCARD
1780                 /* Tell the block device(s) that the sectors can be discarded */
1781                 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1782                                       bytenr, &map_length, &multi, 0);
1783                 if (!ret) {
1784                         struct btrfs_bio_stripe *stripe = multi->stripes;
1785                         int i;
1786
1787                         if (map_length > num_bytes)
1788                                 map_length = num_bytes;
1789
1790                         for (i = 0; i < multi->num_stripes; i++, stripe++) {
1791                                 blkdev_issue_discard(stripe->dev->bdev,
1792                                                      stripe->physical >> 9,
1793                                                      map_length >> 9);
1794                         }
1795                         kfree(multi);
1796                 }
1797 #endif
1798         }
1799         btrfs_free_path(path);
1800         finish_current_insert(trans, extent_root);
1801         return ret;
1802 }
1803
1804 /*
1805  * find all the blocks marked as pending in the radix tree and remove
1806  * them from the extent map
1807  */
1808 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1809                                btrfs_root *extent_root)
1810 {
1811         int ret;
1812         int err = 0;
1813         u64 start;
1814         u64 end;
1815         struct extent_io_tree *pending_del;
1816         struct extent_io_tree *pinned_extents;
1817
1818         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1819         pending_del = &extent_root->fs_info->pending_del;
1820         pinned_extents = &extent_root->fs_info->pinned_extents;
1821
1822         while(1) {
1823                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1824                                             EXTENT_LOCKED);
1825                 if (ret)
1826                         break;
1827                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1828                                   GFP_NOFS);
1829                 if (!test_range_bit(&extent_root->fs_info->extent_ins,
1830                                     start, end, EXTENT_LOCKED, 0)) {
1831                         btrfs_update_pinned_extents(extent_root, start,
1832                                               end + 1 - start, 1);
1833                         ret = __free_extent(trans, extent_root,
1834                                              start, end + 1 - start,
1835                                              extent_root->root_key.objectid,
1836                                              0, 0, 0, 0, 0);
1837                 } else {
1838                         clear_extent_bits(&extent_root->fs_info->extent_ins,
1839                                           start, end, EXTENT_LOCKED, GFP_NOFS);
1840                 }
1841                 if (ret)
1842                         err = ret;
1843
1844                 if (need_resched()) {
1845                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1846                         cond_resched();
1847                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1848                 }
1849         }
1850         return err;
1851 }
1852
1853 /*
1854  * remove an extent from the root, returns 0 on success
1855  */
1856 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1857                                struct btrfs_root *root, u64 bytenr,
1858                                u64 num_bytes, u64 root_objectid,
1859                                u64 ref_generation, u64 owner_objectid,
1860                                u64 owner_offset, int pin)
1861 {
1862         struct btrfs_root *extent_root = root->fs_info->extent_root;
1863         int pending_ret;
1864         int ret;
1865
1866         WARN_ON(num_bytes < root->sectorsize);
1867         if (!root->ref_cows)
1868                 ref_generation = 0;
1869
1870         if (root == extent_root) {
1871                 pin_down_bytes(root, bytenr, num_bytes, 0, 1);
1872                 return 0;
1873         }
1874         /* if metadata always pin */
1875         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
1876                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1877                         /* btrfs_free_reserved_extent */
1878                         set_extent_dirty(&root->fs_info->free_space_cache,
1879                                  bytenr, bytenr + num_bytes - 1, GFP_NOFS);
1880                         return 0;
1881                 }
1882                 pin = 1;
1883         }
1884
1885         /* if data pin when any transaction has committed this */
1886         if (ref_generation != trans->transid)
1887                 pin = 1;
1888
1889         ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1890                             ref_generation, owner_objectid, owner_offset,
1891                             pin, pin == 0);
1892
1893         finish_current_insert(trans, root->fs_info->extent_root);
1894         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1895         return ret ? ret : pending_ret;
1896 }
1897
1898 int btrfs_free_extent(struct btrfs_trans_handle *trans,
1899                       struct btrfs_root *root, u64 bytenr,
1900                       u64 num_bytes, u64 root_objectid,
1901                       u64 ref_generation, u64 owner_objectid,
1902                       u64 owner_offset, int pin)
1903 {
1904         int ret;
1905
1906         maybe_lock_mutex(root);
1907         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1908                                   root_objectid, ref_generation,
1909                                   owner_objectid, owner_offset, pin);
1910         maybe_unlock_mutex(root);
1911         return ret;
1912 }
1913
1914 static u64 stripe_align(struct btrfs_root *root, u64 val)
1915 {
1916         u64 mask = ((u64)root->stripesize - 1);
1917         u64 ret = (val + mask) & ~mask;
1918         return ret;
1919 }
1920
1921 /*
1922  * walks the btree of allocated extents and find a hole of a given size.
1923  * The key ins is changed to record the hole:
1924  * ins->objectid == block start
1925  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1926  * ins->offset == number of blocks
1927  * Any available blocks before search_start are skipped.
1928  */
1929 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1930                                      struct btrfs_root *orig_root,
1931                                      u64 num_bytes, u64 empty_size,
1932                                      u64 search_start, u64 search_end,
1933                                      u64 hint_byte, struct btrfs_key *ins,
1934                                      u64 exclude_start, u64 exclude_nr,
1935                                      int data)
1936 {
1937         int ret;
1938         u64 orig_search_start;
1939         struct btrfs_root * root = orig_root->fs_info->extent_root;
1940         struct btrfs_fs_info *info = root->fs_info;
1941         u64 total_needed = num_bytes;
1942         u64 *last_ptr = NULL;
1943         struct btrfs_block_group_cache *block_group;
1944         int full_scan = 0;
1945         int wrapped = 0;
1946         int chunk_alloc_done = 0;
1947         int empty_cluster = 2 * 1024 * 1024;
1948         int allowed_chunk_alloc = 0;
1949
1950         WARN_ON(num_bytes < root->sectorsize);
1951         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1952
1953         if (orig_root->ref_cows || empty_size)
1954                 allowed_chunk_alloc = 1;
1955
1956         if (data & BTRFS_BLOCK_GROUP_METADATA) {
1957                 last_ptr = &root->fs_info->last_alloc;
1958                 empty_cluster = 256 * 1024;
1959         }
1960
1961         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1962                 last_ptr = &root->fs_info->last_data_alloc;
1963         }
1964         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1965                 last_ptr = &root->fs_info->last_log_alloc;
1966                 if (!last_ptr == 0 && root->fs_info->last_alloc) {
1967                         *last_ptr = root->fs_info->last_alloc + empty_cluster;
1968                 }
1969         }
1970
1971         if (last_ptr) {
1972                 if (*last_ptr)
1973                         hint_byte = *last_ptr;
1974                 else {
1975                         empty_size += empty_cluster;
1976                 }
1977         }
1978
1979         search_start = max(search_start, first_logical_byte(root, 0));
1980         orig_search_start = search_start;
1981
1982         if (search_end == (u64)-1)
1983                 search_end = btrfs_super_total_bytes(&info->super_copy);
1984
1985         if (hint_byte) {
1986                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
1987                 if (!block_group)
1988                         hint_byte = search_start;
1989                 block_group = btrfs_find_block_group(root, block_group,
1990                                                      hint_byte, data, 1);
1991                 if (last_ptr && *last_ptr == 0 && block_group)
1992                         hint_byte = block_group->key.objectid;
1993         } else {
1994                 block_group = btrfs_find_block_group(root,
1995                                                      trans->block_group,
1996                                                      search_start, data, 1);
1997         }
1998         search_start = max(search_start, hint_byte);
1999
2000         total_needed += empty_size;
2001
2002 check_failed:
2003         if (!block_group) {
2004                 block_group = btrfs_lookup_first_block_group(info,
2005                                                              search_start);
2006                 if (!block_group)
2007                         block_group = btrfs_lookup_first_block_group(info,
2008                                                        orig_search_start);
2009         }
2010         if (full_scan && !chunk_alloc_done) {
2011                 if (allowed_chunk_alloc) {
2012                         do_chunk_alloc(trans, root,
2013                                      num_bytes + 2 * 1024 * 1024, data, 1);
2014                         allowed_chunk_alloc = 0;
2015                 } else if (block_group && block_group_bits(block_group, data)) {
2016                         block_group->space_info->force_alloc = 1;
2017                 }
2018                 chunk_alloc_done = 1;
2019         }
2020         ret = find_search_start(root, &block_group, &search_start,
2021                                 total_needed, data);
2022         if (ret == -ENOSPC && last_ptr && *last_ptr) {
2023                 *last_ptr = 0;
2024                 block_group = btrfs_lookup_first_block_group(info,
2025                                                              orig_search_start);
2026                 search_start = orig_search_start;
2027                 ret = find_search_start(root, &block_group, &search_start,
2028                                         total_needed, data);
2029         }
2030         if (ret == -ENOSPC)
2031                 goto enospc;
2032         if (ret)
2033                 goto error;
2034
2035         if (last_ptr && *last_ptr && search_start != *last_ptr) {
2036                 *last_ptr = 0;
2037                 if (!empty_size) {
2038                         empty_size += empty_cluster;
2039                         total_needed += empty_size;
2040                 }
2041                 block_group = btrfs_lookup_first_block_group(info,
2042                                                        orig_search_start);
2043                 search_start = orig_search_start;
2044                 ret = find_search_start(root, &block_group,
2045                                         &search_start, total_needed, data);
2046                 if (ret == -ENOSPC)
2047                         goto enospc;
2048                 if (ret)
2049                         goto error;
2050         }
2051
2052         search_start = stripe_align(root, search_start);
2053         ins->objectid = search_start;
2054         ins->offset = num_bytes;
2055
2056         if (ins->objectid + num_bytes >= search_end)
2057                 goto enospc;
2058
2059         if (ins->objectid + num_bytes >
2060             block_group->key.objectid + block_group->key.offset) {
2061                 search_start = block_group->key.objectid +
2062                         block_group->key.offset;
2063                 goto new_group;
2064         }
2065
2066         if (test_range_bit(&info->extent_ins, ins->objectid,
2067                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2068                 search_start = ins->objectid + num_bytes;
2069                 goto new_group;
2070         }
2071
2072         if (test_range_bit(&info->pinned_extents, ins->objectid,
2073                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2074                 search_start = ins->objectid + num_bytes;
2075                 goto new_group;
2076         }
2077
2078         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2079             ins->objectid < exclude_start + exclude_nr)) {
2080                 search_start = exclude_start + exclude_nr;
2081                 goto new_group;
2082         }
2083
2084         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2085                 block_group = btrfs_lookup_block_group(info, ins->objectid);
2086                 if (block_group)
2087                         trans->block_group = block_group;
2088         }
2089         ins->offset = num_bytes;
2090         if (last_ptr) {
2091                 *last_ptr = ins->objectid + ins->offset;
2092                 if (*last_ptr ==
2093                     btrfs_super_total_bytes(&root->fs_info->super_copy)) {
2094                         *last_ptr = 0;
2095                 }
2096         }
2097         return 0;
2098
2099 new_group:
2100         if (search_start + num_bytes >= search_end) {
2101 enospc:
2102                 search_start = orig_search_start;
2103                 if (full_scan) {
2104                         ret = -ENOSPC;
2105                         goto error;
2106                 }
2107                 if (wrapped) {
2108                         if (!full_scan)
2109                                 total_needed -= empty_size;
2110                         full_scan = 1;
2111                 } else
2112                         wrapped = 1;
2113         }
2114         block_group = btrfs_lookup_first_block_group(info, search_start);
2115         cond_resched();
2116         block_group = btrfs_find_block_group(root, block_group,
2117                                              search_start, data, 0);
2118         goto check_failed;
2119
2120 error:
2121         return ret;
2122 }
2123
2124 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2125                                   struct btrfs_root *root,
2126                                   u64 num_bytes, u64 min_alloc_size,
2127                                   u64 empty_size, u64 hint_byte,
2128                                   u64 search_end, struct btrfs_key *ins,
2129                                   u64 data)
2130 {
2131         int ret;
2132         u64 search_start = 0;
2133         u64 alloc_profile;
2134         struct btrfs_fs_info *info = root->fs_info;
2135
2136         if (data) {
2137                 alloc_profile = info->avail_data_alloc_bits &
2138                                 info->data_alloc_profile;
2139                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2140         } else if (root == root->fs_info->chunk_root) {
2141                 alloc_profile = info->avail_system_alloc_bits &
2142                                 info->system_alloc_profile;
2143                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2144         } else {
2145                 alloc_profile = info->avail_metadata_alloc_bits &
2146                                 info->metadata_alloc_profile;
2147                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2148         }
2149 again:
2150         data = reduce_alloc_profile(root, data);
2151         /*
2152          * the only place that sets empty_size is btrfs_realloc_node, which
2153          * is not called recursively on allocations
2154          */
2155         if (empty_size || root->ref_cows) {
2156                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2157                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2158                                      2 * 1024 * 1024,
2159                                      BTRFS_BLOCK_GROUP_METADATA |
2160                                      (info->metadata_alloc_profile &
2161                                       info->avail_metadata_alloc_bits), 0);
2162                         BUG_ON(ret);
2163                 }
2164                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2165                                      num_bytes + 2 * 1024 * 1024, data, 0);
2166                 BUG_ON(ret);
2167         }
2168
2169         WARN_ON(num_bytes < root->sectorsize);
2170         ret = find_free_extent(trans, root, num_bytes, empty_size,
2171                                search_start, search_end, hint_byte, ins,
2172                                trans->alloc_exclude_start,
2173                                trans->alloc_exclude_nr, data);
2174
2175         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2176                 num_bytes = num_bytes >> 1;
2177                 num_bytes = max(num_bytes, min_alloc_size);
2178                 do_chunk_alloc(trans, root->fs_info->extent_root,
2179                                num_bytes, data, 1);
2180                 goto again;
2181         }
2182         if (ret) {
2183                 printk("allocation failed flags %Lu\n", data);
2184                 BUG();
2185         }
2186         clear_extent_dirty(&root->fs_info->free_space_cache,
2187                            ins->objectid, ins->objectid + ins->offset - 1,
2188                            GFP_NOFS);
2189         return 0;
2190 }
2191
2192 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2193 {
2194         maybe_lock_mutex(root);
2195         set_extent_dirty(&root->fs_info->free_space_cache,
2196                          start, start + len - 1, GFP_NOFS);
2197         maybe_unlock_mutex(root);
2198         return 0;
2199 }
2200
2201 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2202                                   struct btrfs_root *root,
2203                                   u64 num_bytes, u64 min_alloc_size,
2204                                   u64 empty_size, u64 hint_byte,
2205                                   u64 search_end, struct btrfs_key *ins,
2206                                   u64 data)
2207 {
2208         int ret;
2209         maybe_lock_mutex(root);
2210         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2211                                      empty_size, hint_byte, search_end, ins,
2212                                      data);
2213         maybe_unlock_mutex(root);
2214         return ret;
2215 }
2216
2217 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2218                                          struct btrfs_root *root,
2219                                          u64 root_objectid, u64 ref_generation,
2220                                          u64 owner, u64 owner_offset,
2221                                          struct btrfs_key *ins)
2222 {
2223         int ret;
2224         int pending_ret;
2225         u64 super_used;
2226         u64 root_used;
2227         u64 num_bytes = ins->offset;
2228         u32 sizes[2];
2229         struct btrfs_fs_info *info = root->fs_info;
2230         struct btrfs_root *extent_root = info->extent_root;
2231         struct btrfs_extent_item *extent_item;
2232         struct btrfs_extent_ref *ref;
2233         struct btrfs_path *path;
2234         struct btrfs_key keys[2];
2235
2236         /* block accounting for super block */
2237         spin_lock_irq(&info->delalloc_lock);
2238         super_used = btrfs_super_bytes_used(&info->super_copy);
2239         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2240         spin_unlock_irq(&info->delalloc_lock);
2241
2242         /* block accounting for root item */
2243         root_used = btrfs_root_used(&root->root_item);
2244         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2245
2246         if (root == extent_root) {
2247                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2248                                 ins->objectid + ins->offset - 1,
2249                                 EXTENT_LOCKED, GFP_NOFS);
2250                 goto update_block;
2251         }
2252
2253         memcpy(&keys[0], ins, sizeof(*ins));
2254         keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
2255                                          owner, owner_offset);
2256         keys[1].objectid = ins->objectid;
2257         keys[1].type = BTRFS_EXTENT_REF_KEY;
2258         sizes[0] = sizeof(*extent_item);
2259         sizes[1] = sizeof(*ref);
2260
2261         path = btrfs_alloc_path();
2262         BUG_ON(!path);
2263
2264         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2265                                        sizes, 2);
2266
2267         BUG_ON(ret);
2268         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2269                                      struct btrfs_extent_item);
2270         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2271         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2272                              struct btrfs_extent_ref);
2273
2274         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2275         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2276         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2277         btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2278
2279         btrfs_mark_buffer_dirty(path->nodes[0]);
2280
2281         trans->alloc_exclude_start = 0;
2282         trans->alloc_exclude_nr = 0;
2283         btrfs_free_path(path);
2284         finish_current_insert(trans, extent_root);
2285         pending_ret = del_pending_extents(trans, extent_root);
2286
2287         if (ret)
2288                 goto out;
2289         if (pending_ret) {
2290                 ret = pending_ret;
2291                 goto out;
2292         }
2293
2294 update_block:
2295         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2296         if (ret) {
2297                 printk("update block group failed for %Lu %Lu\n",
2298                        ins->objectid, ins->offset);
2299                 BUG();
2300         }
2301 out:
2302         return ret;
2303 }
2304
2305 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2306                                 struct btrfs_root *root,
2307                                 u64 root_objectid, u64 ref_generation,
2308                                 u64 owner, u64 owner_offset,
2309                                 struct btrfs_key *ins)
2310 {
2311         int ret;
2312         maybe_lock_mutex(root);
2313         ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2314                                             ref_generation, owner,
2315                                             owner_offset, ins);
2316         maybe_unlock_mutex(root);
2317         return ret;
2318 }
2319
2320 /*
2321  * this is used by the tree logging recovery code.  It records that
2322  * an extent has been allocated and makes sure to clear the free
2323  * space cache bits as well
2324  */
2325 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2326                                 struct btrfs_root *root,
2327                                 u64 root_objectid, u64 ref_generation,
2328                                 u64 owner, u64 owner_offset,
2329                                 struct btrfs_key *ins)
2330 {
2331         int ret;
2332         struct btrfs_block_group_cache *block_group;
2333
2334         maybe_lock_mutex(root);
2335         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2336         cache_block_group(root, block_group);
2337
2338         clear_extent_dirty(&root->fs_info->free_space_cache,
2339                            ins->objectid, ins->objectid + ins->offset - 1,
2340                            GFP_NOFS);
2341         ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2342                                             ref_generation, owner,
2343                                             owner_offset, ins);
2344         maybe_unlock_mutex(root);
2345         return ret;
2346 }
2347
2348 /*
2349  * finds a free extent and does all the dirty work required for allocation
2350  * returns the key for the extent through ins, and a tree buffer for
2351  * the first block of the extent through buf.
2352  *
2353  * returns 0 if everything worked, non-zero otherwise.
2354  */
2355 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2356                        struct btrfs_root *root,
2357                        u64 num_bytes, u64 min_alloc_size,
2358                        u64 root_objectid, u64 ref_generation,
2359                        u64 owner, u64 owner_offset,
2360                        u64 empty_size, u64 hint_byte,
2361                        u64 search_end, struct btrfs_key *ins, u64 data)
2362 {
2363         int ret;
2364
2365         maybe_lock_mutex(root);
2366
2367         ret = __btrfs_reserve_extent(trans, root, num_bytes,
2368                                      min_alloc_size, empty_size, hint_byte,
2369                                      search_end, ins, data);
2370         BUG_ON(ret);
2371         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2372                 ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2373                                                     ref_generation, owner,
2374                                                     owner_offset, ins);
2375                 BUG_ON(ret);
2376
2377         }
2378         maybe_unlock_mutex(root);
2379         return ret;
2380 }
2381
2382 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2383                                             struct btrfs_root *root,
2384                                             u64 bytenr, u32 blocksize)
2385 {
2386         struct extent_buffer *buf;
2387
2388         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2389         if (!buf)
2390                 return ERR_PTR(-ENOMEM);
2391         btrfs_set_header_generation(buf, trans->transid);
2392         btrfs_tree_lock(buf);
2393         clean_tree_block(trans, root, buf);
2394         btrfs_set_buffer_uptodate(buf);
2395         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2396                          buf->start + buf->len - 1, GFP_NOFS);
2397         trans->blocks_used++;
2398         return buf;
2399 }
2400
2401 /*
2402  * helper function to allocate a block for a given tree
2403  * returns the tree buffer or NULL.
2404  */
2405 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2406                                              struct btrfs_root *root,
2407                                              u32 blocksize,
2408                                              u64 root_objectid,
2409                                              u64 ref_generation,
2410                                              u64 first_objectid,
2411                                              int level,
2412                                              u64 hint,
2413                                              u64 empty_size)
2414 {
2415         struct btrfs_key ins;
2416         int ret;
2417         struct extent_buffer *buf;
2418
2419         ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2420                                  root_objectid, ref_generation,
2421                                  level, first_objectid, empty_size, hint,
2422                                  (u64)-1, &ins, 0);
2423         if (ret) {
2424                 BUG_ON(ret > 0);
2425                 return ERR_PTR(ret);
2426         }
2427
2428         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2429         return buf;
2430 }
2431
2432 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2433                         struct btrfs_root *root, struct extent_buffer *leaf)
2434 {
2435         u64 leaf_owner;
2436         u64 leaf_generation;
2437         struct btrfs_key key;
2438         struct btrfs_file_extent_item *fi;
2439         int i;
2440         int nritems;
2441         int ret;
2442
2443         BUG_ON(!btrfs_is_leaf(leaf));
2444         nritems = btrfs_header_nritems(leaf);
2445         leaf_owner = btrfs_header_owner(leaf);
2446         leaf_generation = btrfs_header_generation(leaf);
2447
2448         for (i = 0; i < nritems; i++) {
2449                 u64 disk_bytenr;
2450                 cond_resched();
2451
2452                 btrfs_item_key_to_cpu(leaf, &key, i);
2453                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2454                         continue;
2455                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2456                 if (btrfs_file_extent_type(leaf, fi) ==
2457                     BTRFS_FILE_EXTENT_INLINE)
2458                         continue;
2459                 /*
2460                  * FIXME make sure to insert a trans record that
2461                  * repeats the snapshot del on crash
2462                  */
2463                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2464                 if (disk_bytenr == 0)
2465                         continue;
2466
2467                 mutex_lock(&root->fs_info->alloc_mutex);
2468                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2469                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2470                                 leaf_owner, leaf_generation,
2471                                 key.objectid, key.offset, 0);
2472                 mutex_unlock(&root->fs_info->alloc_mutex);
2473
2474                 atomic_inc(&root->fs_info->throttle_gen);
2475                 wake_up(&root->fs_info->transaction_throttle);
2476                 cond_resched();
2477
2478                 BUG_ON(ret);
2479         }
2480         return 0;
2481 }
2482
2483 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2484                                         struct btrfs_root *root,
2485                                         struct btrfs_leaf_ref *ref)
2486 {
2487         int i;
2488         int ret;
2489         struct btrfs_extent_info *info = ref->extents;
2490
2491         for (i = 0; i < ref->nritems; i++) {
2492                 mutex_lock(&root->fs_info->alloc_mutex);
2493                 ret = __btrfs_free_extent(trans, root,
2494                                         info->bytenr, info->num_bytes,
2495                                         ref->owner, ref->generation,
2496                                         info->objectid, info->offset, 0);
2497                 mutex_unlock(&root->fs_info->alloc_mutex);
2498
2499                 atomic_inc(&root->fs_info->throttle_gen);
2500                 wake_up(&root->fs_info->transaction_throttle);
2501                 cond_resched();
2502
2503                 BUG_ON(ret);
2504                 info++;
2505         }
2506
2507         return 0;
2508 }
2509
2510 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2511                               u32 *refs)
2512 {
2513         int ret;
2514
2515         ret = lookup_extent_ref(NULL, root, start, len, refs);
2516         BUG_ON(ret);
2517
2518 #if 0 // some debugging code in case we see problems here
2519         /* if the refs count is one, it won't get increased again.  But
2520          * if the ref count is > 1, someone may be decreasing it at
2521          * the same time we are.
2522          */
2523         if (*refs != 1) {
2524                 struct extent_buffer *eb = NULL;
2525                 eb = btrfs_find_create_tree_block(root, start, len);
2526                 if (eb)
2527                         btrfs_tree_lock(eb);
2528
2529                 mutex_lock(&root->fs_info->alloc_mutex);
2530                 ret = lookup_extent_ref(NULL, root, start, len, refs);
2531                 BUG_ON(ret);
2532                 mutex_unlock(&root->fs_info->alloc_mutex);
2533
2534                 if (eb) {
2535                         btrfs_tree_unlock(eb);
2536                         free_extent_buffer(eb);
2537                 }
2538                 if (*refs == 1) {
2539                         printk("block %llu went down to one during drop_snap\n",
2540                                (unsigned long long)start);
2541                 }
2542
2543         }
2544 #endif
2545
2546         cond_resched();
2547         return ret;
2548 }
2549
2550 /*
2551  * helper function for drop_snapshot, this walks down the tree dropping ref
2552  * counts as it goes.
2553  */
2554 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2555                                    struct btrfs_root *root,
2556                                    struct btrfs_path *path, int *level)
2557 {
2558         u64 root_owner;
2559         u64 root_gen;
2560         u64 bytenr;
2561         u64 ptr_gen;
2562         struct extent_buffer *next;
2563         struct extent_buffer *cur;
2564         struct extent_buffer *parent;
2565         struct btrfs_leaf_ref *ref;
2566         u32 blocksize;
2567         int ret;
2568         u32 refs;
2569
2570         WARN_ON(*level < 0);
2571         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2572         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2573                                 path->nodes[*level]->len, &refs);
2574         BUG_ON(ret);
2575         if (refs > 1)
2576                 goto out;
2577
2578         /*
2579          * walk down to the last node level and free all the leaves
2580          */
2581         while(*level >= 0) {
2582                 WARN_ON(*level < 0);
2583                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2584                 cur = path->nodes[*level];
2585
2586                 if (btrfs_header_level(cur) != *level)
2587                         WARN_ON(1);
2588
2589                 if (path->slots[*level] >=
2590                     btrfs_header_nritems(cur))
2591                         break;
2592                 if (*level == 0) {
2593                         ret = btrfs_drop_leaf_ref(trans, root, cur);
2594                         BUG_ON(ret);
2595                         break;
2596                 }
2597                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2598                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2599                 blocksize = btrfs_level_size(root, *level - 1);
2600
2601                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2602                 BUG_ON(ret);
2603                 if (refs != 1) {
2604                         parent = path->nodes[*level];
2605                         root_owner = btrfs_header_owner(parent);
2606                         root_gen = btrfs_header_generation(parent);
2607                         path->slots[*level]++;
2608
2609                         mutex_lock(&root->fs_info->alloc_mutex);
2610                         ret = __btrfs_free_extent(trans, root, bytenr,
2611                                                 blocksize, root_owner,
2612                                                 root_gen, 0, 0, 1);
2613                         BUG_ON(ret);
2614                         mutex_unlock(&root->fs_info->alloc_mutex);
2615
2616                         atomic_inc(&root->fs_info->throttle_gen);
2617                         wake_up(&root->fs_info->transaction_throttle);
2618                         cond_resched();
2619
2620                         continue;
2621                 }
2622                 /*
2623                  * at this point, we have a single ref, and since the
2624                  * only place referencing this extent is a dead root
2625                  * the reference count should never go higher.
2626                  * So, we don't need to check it again
2627                  */
2628                 if (*level == 1) {
2629                         struct btrfs_key key;
2630                         btrfs_node_key_to_cpu(cur, &key, path->slots[*level]);
2631                         ref = btrfs_lookup_leaf_ref(root, bytenr);
2632                         if (ref) {
2633                                 ret = cache_drop_leaf_ref(trans, root, ref);
2634                                 BUG_ON(ret);
2635                                 btrfs_remove_leaf_ref(root, ref);
2636                                 btrfs_free_leaf_ref(root, ref);
2637                                 *level = 0;
2638                                 break;
2639                         }
2640                         if (printk_ratelimit())
2641                                 printk("leaf ref miss for bytenr %llu\n",
2642                                        (unsigned long long)bytenr);
2643                 }
2644                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2645                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2646                         free_extent_buffer(next);
2647
2648                         next = read_tree_block(root, bytenr, blocksize,
2649                                                ptr_gen);
2650                         cond_resched();
2651 #if 0
2652                         /*
2653                          * this is a debugging check and can go away
2654                          * the ref should never go all the way down to 1
2655                          * at this point
2656                          */
2657                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2658                                                 &refs);
2659                         BUG_ON(ret);
2660                         WARN_ON(refs != 1);
2661 #endif
2662                 }
2663                 WARN_ON(*level <= 0);
2664                 if (path->nodes[*level-1])
2665                         free_extent_buffer(path->nodes[*level-1]);
2666                 path->nodes[*level-1] = next;
2667                 *level = btrfs_header_level(next);
2668                 path->slots[*level] = 0;
2669                 cond_resched();
2670         }
2671 out:
2672         WARN_ON(*level < 0);
2673         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2674
2675         if (path->nodes[*level] == root->node) {
2676                 parent = path->nodes[*level];
2677                 bytenr = path->nodes[*level]->start;
2678         } else {
2679                 parent = path->nodes[*level + 1];
2680                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
2681         }
2682
2683         blocksize = btrfs_level_size(root, *level);
2684         root_owner = btrfs_header_owner(parent);
2685         root_gen = btrfs_header_generation(parent);
2686
2687         mutex_lock(&root->fs_info->alloc_mutex);
2688         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
2689                                   root_owner, root_gen, 0, 0, 1);
2690         free_extent_buffer(path->nodes[*level]);
2691         path->nodes[*level] = NULL;
2692         *level += 1;
2693         BUG_ON(ret);
2694         mutex_unlock(&root->fs_info->alloc_mutex);
2695
2696         cond_resched();
2697         return 0;
2698 }
2699
2700 /*
2701  * helper for dropping snapshots.  This walks back up the tree in the path
2702  * to find the first node higher up where we haven't yet gone through
2703  * all the slots
2704  */
2705 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2706                                  struct btrfs_root *root,
2707                                  struct btrfs_path *path, int *level)
2708 {
2709         u64 root_owner;
2710         u64 root_gen;
2711         struct btrfs_root_item *root_item = &root->root_item;
2712         int i;
2713         int slot;
2714         int ret;
2715
2716         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2717                 slot = path->slots[i];
2718                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2719                         struct extent_buffer *node;
2720                         struct btrfs_disk_key disk_key;
2721                         node = path->nodes[i];
2722                         path->slots[i]++;
2723                         *level = i;
2724                         WARN_ON(*level == 0);
2725                         btrfs_node_key(node, &disk_key, path->slots[i]);
2726                         memcpy(&root_item->drop_progress,
2727                                &disk_key, sizeof(disk_key));
2728                         root_item->drop_level = i;
2729                         return 0;
2730                 } else {
2731                         if (path->nodes[*level] == root->node) {
2732                                 root_owner = root->root_key.objectid;
2733                                 root_gen =
2734                                    btrfs_header_generation(path->nodes[*level]);
2735                         } else {
2736                                 struct extent_buffer *node;
2737                                 node = path->nodes[*level + 1];
2738                                 root_owner = btrfs_header_owner(node);
2739                                 root_gen = btrfs_header_generation(node);
2740                         }
2741                         ret = btrfs_free_extent(trans, root,
2742                                                 path->nodes[*level]->start,
2743                                                 path->nodes[*level]->len,
2744                                                 root_owner, root_gen, 0, 0, 1);
2745                         BUG_ON(ret);
2746                         free_extent_buffer(path->nodes[*level]);
2747                         path->nodes[*level] = NULL;
2748                         *level = i + 1;
2749                 }
2750         }
2751         return 1;
2752 }
2753
2754 /*
2755  * drop the reference count on the tree rooted at 'snap'.  This traverses
2756  * the tree freeing any blocks that have a ref count of zero after being
2757  * decremented.
2758  */
2759 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2760                         *root)
2761 {
2762         int ret = 0;
2763         int wret;
2764         int level;
2765         struct btrfs_path *path;
2766         int i;
2767         int orig_level;
2768         struct btrfs_root_item *root_item = &root->root_item;
2769
2770         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2771         path = btrfs_alloc_path();
2772         BUG_ON(!path);
2773
2774         level = btrfs_header_level(root->node);
2775         orig_level = level;
2776         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2777                 path->nodes[level] = root->node;
2778                 extent_buffer_get(root->node);
2779                 path->slots[level] = 0;
2780         } else {
2781                 struct btrfs_key key;
2782                 struct btrfs_disk_key found_key;
2783                 struct extent_buffer *node;
2784
2785                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2786                 level = root_item->drop_level;
2787                 path->lowest_level = level;
2788                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2789                 if (wret < 0) {
2790                         ret = wret;
2791                         goto out;
2792                 }
2793                 node = path->nodes[level];
2794                 btrfs_node_key(node, &found_key, path->slots[level]);
2795                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2796                                sizeof(found_key)));
2797                 /*
2798                  * unlock our path, this is safe because only this
2799                  * function is allowed to delete this snapshot
2800                  */
2801                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2802                         if (path->nodes[i] && path->locks[i]) {
2803                                 path->locks[i] = 0;
2804                                 btrfs_tree_unlock(path->nodes[i]);
2805                         }
2806                 }
2807         }
2808         while(1) {
2809                 wret = walk_down_tree(trans, root, path, &level);
2810                 if (wret > 0)
2811                         break;
2812                 if (wret < 0)
2813                         ret = wret;
2814
2815                 wret = walk_up_tree(trans, root, path, &level);
2816                 if (wret > 0)
2817                         break;
2818                 if (wret < 0)
2819                         ret = wret;
2820                 if (trans->transaction->in_commit) {
2821                         ret = -EAGAIN;
2822                         break;
2823                 }
2824                 atomic_inc(&root->fs_info->throttle_gen);
2825                 wake_up(&root->fs_info->transaction_throttle);
2826         }
2827         for (i = 0; i <= orig_level; i++) {
2828                 if (path->nodes[i]) {
2829                         free_extent_buffer(path->nodes[i]);
2830                         path->nodes[i] = NULL;
2831                 }
2832         }
2833 out:
2834         btrfs_free_path(path);
2835         return ret;
2836 }
2837
2838 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2839 {
2840         u64 start;
2841         u64 end;
2842         u64 ptr;
2843         int ret;
2844
2845         mutex_lock(&info->alloc_mutex);
2846         while(1) {
2847                 ret = find_first_extent_bit(&info->block_group_cache, 0,
2848                                             &start, &end, (unsigned int)-1);
2849                 if (ret)
2850                         break;
2851                 ret = get_state_private(&info->block_group_cache, start, &ptr);
2852                 if (!ret)
2853                         kfree((void *)(unsigned long)ptr);
2854                 clear_extent_bits(&info->block_group_cache, start,
2855                                   end, (unsigned int)-1, GFP_NOFS);
2856         }
2857         while(1) {
2858                 ret = find_first_extent_bit(&info->free_space_cache, 0,
2859                                             &start, &end, EXTENT_DIRTY);
2860                 if (ret)
2861                         break;
2862                 clear_extent_dirty(&info->free_space_cache, start,
2863                                    end, GFP_NOFS);
2864         }
2865         mutex_unlock(&info->alloc_mutex);
2866         return 0;
2867 }
2868
2869 static unsigned long calc_ra(unsigned long start, unsigned long last,
2870                              unsigned long nr)
2871 {
2872         return min(last, start + nr - 1);
2873 }
2874
2875 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2876                                          u64 len)
2877 {
2878         u64 page_start;
2879         u64 page_end;
2880         unsigned long last_index;
2881         unsigned long i;
2882         struct page *page;
2883         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2884         struct file_ra_state *ra;
2885         unsigned long total_read = 0;
2886         unsigned long ra_pages;
2887         struct btrfs_ordered_extent *ordered;
2888         struct btrfs_trans_handle *trans;
2889
2890         ra = kzalloc(sizeof(*ra), GFP_NOFS);
2891
2892         mutex_lock(&inode->i_mutex);
2893         i = start >> PAGE_CACHE_SHIFT;
2894         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2895
2896         ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2897
2898         file_ra_state_init(ra, inode->i_mapping);
2899
2900         for (; i <= last_index; i++) {
2901                 if (total_read % ra_pages == 0) {
2902                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2903                                        calc_ra(i, last_index, ra_pages));
2904                 }
2905                 total_read++;
2906 again:
2907                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
2908                         goto truncate_racing;
2909                 page = grab_cache_page(inode->i_mapping, i);
2910                 if (!page) {
2911                         goto out_unlock;
2912                 }
2913                 if (!PageUptodate(page)) {
2914                         btrfs_readpage(NULL, page);
2915                         lock_page(page);
2916                         if (!PageUptodate(page)) {
2917                                 unlock_page(page);
2918                                 page_cache_release(page);
2919                                 goto out_unlock;
2920                         }
2921                 }
2922                 wait_on_page_writeback(page);
2923
2924                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2925                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2926                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2927
2928                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2929                 if (ordered) {
2930                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2931                         unlock_page(page);
2932                         page_cache_release(page);
2933                         btrfs_start_ordered_extent(inode, ordered, 1);
2934                         btrfs_put_ordered_extent(ordered);
2935                         goto again;
2936                 }
2937                 set_page_extent_mapped(page);
2938
2939                 /*
2940                  * make sure page_mkwrite is called for this page if userland
2941                  * wants to change it from mmap
2942                  */
2943                 clear_page_dirty_for_io(page);
2944
2945                 btrfs_set_extent_delalloc(inode, page_start, page_end);
2946                 set_page_dirty(page);
2947
2948                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2949                 unlock_page(page);
2950                 page_cache_release(page);
2951         }
2952
2953 out_unlock:
2954         /* we have to start the IO in order to get the ordered extents
2955          * instantiated.  This allows the relocation to code to wait
2956          * for all the ordered extents to hit the disk.
2957          *
2958          * Otherwise, it would constantly loop over the same extents
2959          * because the old ones don't get deleted  until the IO is
2960          * started
2961          */
2962         btrfs_fdatawrite_range(inode->i_mapping, start, start + len - 1,
2963                                WB_SYNC_NONE);
2964         kfree(ra);
2965         trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2966         if (trans) {
2967                 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2968                 mark_inode_dirty(inode);
2969         }
2970         mutex_unlock(&inode->i_mutex);
2971         return 0;
2972
2973 truncate_racing:
2974         vmtruncate(inode, inode->i_size);
2975         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2976                                            total_read);
2977         goto out_unlock;
2978 }
2979
2980 /*
2981  * The back references tell us which tree holds a ref on a block,
2982  * but it is possible for the tree root field in the reference to
2983  * reflect the original root before a snapshot was made.  In this
2984  * case we should search through all the children of a given root
2985  * to find potential holders of references on a block.
2986  *
2987  * Instead, we do something a little less fancy and just search
2988  * all the roots for a given key/block combination.
2989  */
2990 static int find_root_for_ref(struct btrfs_root *root,
2991                              struct btrfs_path *path,
2992                              struct btrfs_key *key0,
2993                              int level,
2994                              int file_key,
2995                              struct btrfs_root **found_root,
2996                              u64 bytenr)
2997 {
2998         struct btrfs_key root_location;
2999         struct btrfs_root *cur_root = *found_root;
3000         struct btrfs_file_extent_item *file_extent;
3001         u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
3002         u64 found_bytenr;
3003         int ret;
3004
3005         root_location.offset = (u64)-1;
3006         root_location.type = BTRFS_ROOT_ITEM_KEY;
3007         path->lowest_level = level;
3008         path->reada = 0;
3009         while(1) {
3010                 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
3011                 found_bytenr = 0;
3012                 if (ret == 0 && file_key) {
3013                         struct extent_buffer *leaf = path->nodes[0];
3014                         file_extent = btrfs_item_ptr(leaf, path->slots[0],
3015                                              struct btrfs_file_extent_item);
3016                         if (btrfs_file_extent_type(leaf, file_extent) ==
3017                             BTRFS_FILE_EXTENT_REG) {
3018                                 found_bytenr =
3019                                         btrfs_file_extent_disk_bytenr(leaf,
3020                                                                file_extent);
3021                        }
3022                 } else if (!file_key) {
3023                         if (path->nodes[level])
3024                                 found_bytenr = path->nodes[level]->start;
3025                 }
3026
3027                 btrfs_release_path(cur_root, path);
3028
3029                 if (found_bytenr == bytenr) {
3030                         *found_root = cur_root;
3031                         ret = 0;
3032                         goto out;
3033                 }
3034                 ret = btrfs_search_root(root->fs_info->tree_root,
3035                                         root_search_start, &root_search_start);
3036                 if (ret)
3037                         break;
3038
3039                 root_location.objectid = root_search_start;
3040                 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
3041                                                       &root_location);
3042                 if (!cur_root) {
3043                         ret = 1;
3044                         break;
3045                 }
3046         }
3047 out:
3048         path->lowest_level = 0;
3049         return ret;
3050 }
3051
3052 /*
3053  * note, this releases the path
3054  */
3055 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
3056                                   struct btrfs_path *path,
3057                                   struct btrfs_key *extent_key,
3058                                   u64 *last_file_objectid,
3059                                   u64 *last_file_offset,
3060                                   u64 *last_file_root,
3061                                   u64 last_extent)
3062 {
3063         struct inode *inode;
3064         struct btrfs_root *found_root;
3065         struct btrfs_key root_location;
3066         struct btrfs_key found_key;
3067         struct btrfs_extent_ref *ref;
3068         u64 ref_root;
3069         u64 ref_gen;
3070         u64 ref_objectid;
3071         u64 ref_offset;
3072         int ret;
3073         int level;
3074
3075         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
3076
3077         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
3078                              struct btrfs_extent_ref);
3079         ref_root = btrfs_ref_root(path->nodes[0], ref);
3080         ref_gen = btrfs_ref_generation(path->nodes[0], ref);
3081         ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
3082         ref_offset = btrfs_ref_offset(path->nodes[0], ref);
3083         btrfs_release_path(extent_root, path);
3084
3085         root_location.objectid = ref_root;
3086         if (ref_gen == 0)
3087                 root_location.offset = 0;
3088         else
3089                 root_location.offset = (u64)-1;
3090         root_location.type = BTRFS_ROOT_ITEM_KEY;
3091
3092         found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
3093                                                 &root_location);
3094         BUG_ON(!found_root);
3095         mutex_unlock(&extent_root->fs_info->alloc_mutex);
3096
3097         if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
3098                 found_key.objectid = ref_objectid;
3099                 found_key.type = BTRFS_EXTENT_DATA_KEY;
3100                 found_key.offset = ref_offset;
3101                 level = 0;
3102
3103                 if (last_extent == extent_key->objectid &&
3104                     *last_file_objectid == ref_objectid &&
3105                     *last_file_offset == ref_offset &&
3106                     *last_file_root == ref_root)
3107                         goto out;
3108
3109                 ret = find_root_for_ref(extent_root, path, &found_key,
3110                                         level, 1, &found_root,
3111                                         extent_key->objectid);
3112
3113                 if (ret)
3114                         goto out;
3115
3116                 if (last_extent == extent_key->objectid &&
3117                     *last_file_objectid == ref_objectid &&
3118                     *last_file_offset == ref_offset &&
3119                     *last_file_root == ref_root)
3120                         goto out;
3121
3122                 inode = btrfs_iget_locked(extent_root->fs_info->sb,
3123                                           ref_objectid, found_root);
3124                 if (inode->i_state & I_NEW) {
3125                         /* the inode and parent dir are two different roots */
3126                         BTRFS_I(inode)->root = found_root;
3127                         BTRFS_I(inode)->location.objectid = ref_objectid;
3128                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
3129                         BTRFS_I(inode)->location.offset = 0;
3130                         btrfs_read_locked_inode(inode);
3131                         unlock_new_inode(inode);
3132
3133                 }
3134                 /* this can happen if the reference is not against
3135                  * the latest version of the tree root
3136                  */
3137                 if (is_bad_inode(inode))
3138                         goto out;
3139
3140                 *last_file_objectid = inode->i_ino;
3141                 *last_file_root = found_root->root_key.objectid;
3142                 *last_file_offset = ref_offset;
3143
3144                 relocate_inode_pages(inode, ref_offset, extent_key->offset);
3145                 iput(inode);
3146         } else {
3147                 struct btrfs_trans_handle *trans;
3148                 struct extent_buffer *eb;
3149                 int needs_lock = 0;
3150
3151                 eb = read_tree_block(found_root, extent_key->objectid,
3152                                      extent_key->offset, 0);
3153                 btrfs_tree_lock(eb);
3154                 level = btrfs_header_level(eb);
3155
3156                 if (level == 0)
3157                         btrfs_item_key_to_cpu(eb, &found_key, 0);
3158                 else
3159                         btrfs_node_key_to_cpu(eb, &found_key, 0);
3160
3161                 btrfs_tree_unlock(eb);
3162                 free_extent_buffer(eb);
3163
3164                 ret = find_root_for_ref(extent_root, path, &found_key,
3165                                         level, 0, &found_root,
3166                                         extent_key->objectid);
3167
3168                 if (ret)
3169                         goto out;
3170
3171                 /*
3172                  * right here almost anything could happen to our key,
3173                  * but that's ok.  The cow below will either relocate it
3174                  * or someone else will have relocated it.  Either way,
3175                  * it is in a different spot than it was before and
3176                  * we're happy.
3177                  */
3178
3179                 trans = btrfs_start_transaction(found_root, 1);
3180
3181                 if (found_root == extent_root->fs_info->extent_root ||
3182                     found_root == extent_root->fs_info->chunk_root ||
3183                     found_root == extent_root->fs_info->dev_root) {
3184                         needs_lock = 1;
3185                         mutex_lock(&extent_root->fs_info->alloc_mutex);
3186                 }
3187
3188                 path->lowest_level = level;
3189                 path->reada = 2;
3190                 ret = btrfs_search_slot(trans, found_root, &found_key, path,
3191                                         0, 1);
3192                 path->lowest_level = 0;
3193                 btrfs_release_path(found_root, path);
3194
3195                 if (found_root == found_root->fs_info->extent_root)
3196                         btrfs_extent_post_op(trans, found_root);
3197                 if (needs_lock)
3198                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
3199
3200                 btrfs_end_transaction(trans, found_root);
3201
3202         }
3203 out:
3204         mutex_lock(&extent_root->fs_info->alloc_mutex);
3205         return 0;
3206 }
3207
3208 static int noinline del_extent_zero(struct btrfs_root *extent_root,
3209                                     struct btrfs_path *path,
3210                                     struct btrfs_key *extent_key)
3211 {
3212         int ret;
3213         struct btrfs_trans_handle *trans;
3214
3215         trans = btrfs_start_transaction(extent_root, 1);
3216         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
3217         if (ret > 0) {
3218                 ret = -EIO;
3219                 goto out;
3220         }
3221         if (ret < 0)
3222                 goto out;
3223         ret = btrfs_del_item(trans, extent_root, path);
3224 out:
3225         btrfs_end_transaction(trans, extent_root);
3226         return ret;
3227 }
3228
3229 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
3230                                         struct btrfs_path *path,
3231                                         struct btrfs_key *extent_key)
3232 {
3233         struct btrfs_key key;
3234         struct btrfs_key found_key;
3235         struct extent_buffer *leaf;
3236         u64 last_file_objectid = 0;
3237         u64 last_file_root = 0;
3238         u64 last_file_offset = (u64)-1;
3239         u64 last_extent = 0;
3240         u32 nritems;
3241         u32 item_size;
3242         int ret = 0;
3243
3244         if (extent_key->objectid == 0) {
3245                 ret = del_extent_zero(extent_root, path, extent_key);
3246                 goto out;
3247         }
3248         key.objectid = extent_key->objectid;
3249         key.type = BTRFS_EXTENT_REF_KEY;
3250         key.offset = 0;
3251
3252         while(1) {
3253                 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3254
3255                 if (ret < 0)
3256                         goto out;
3257
3258                 ret = 0;
3259                 leaf = path->nodes[0];
3260                 nritems = btrfs_header_nritems(leaf);
3261                 if (path->slots[0] == nritems) {
3262                         ret = btrfs_next_leaf(extent_root, path);
3263                         if (ret > 0) {
3264                                 ret = 0;
3265                                 goto out;
3266                         }
3267                         if (ret < 0)
3268                                 goto out;
3269                         leaf = path->nodes[0];
3270                 }
3271
3272                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3273                 if (found_key.objectid != extent_key->objectid) {
3274                         break;
3275                 }
3276
3277                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
3278                         break;
3279                 }
3280
3281                 key.offset = found_key.offset + 1;
3282                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3283
3284                 ret = relocate_one_reference(extent_root, path, extent_key,
3285                                              &last_file_objectid,
3286                                              &last_file_offset,
3287                                              &last_file_root, last_extent);
3288                 if (ret)
3289                         goto out;
3290                 last_extent = extent_key->objectid;
3291         }
3292         ret = 0;
3293 out:
3294         btrfs_release_path(extent_root, path);
3295         return ret;
3296 }
3297
3298 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
3299 {
3300         u64 num_devices;
3301         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
3302                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
3303
3304         num_devices = root->fs_info->fs_devices->num_devices;
3305         if (num_devices == 1) {
3306                 stripped |= BTRFS_BLOCK_GROUP_DUP;
3307                 stripped = flags & ~stripped;
3308
3309                 /* turn raid0 into single device chunks */
3310                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
3311                         return stripped;
3312
3313                 /* turn mirroring into duplication */
3314                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3315                              BTRFS_BLOCK_GROUP_RAID10))
3316                         return stripped | BTRFS_BLOCK_GROUP_DUP;
3317                 return flags;
3318         } else {
3319                 /* they already had raid on here, just return */
3320                 if (flags & stripped)
3321                         return flags;
3322
3323                 stripped |= BTRFS_BLOCK_GROUP_DUP;
3324                 stripped = flags & ~stripped;
3325
3326                 /* switch duplicated blocks with raid1 */
3327                 if (flags & BTRFS_BLOCK_GROUP_DUP)
3328                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
3329
3330                 /* turn single device chunks into raid0 */
3331                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
3332         }
3333         return flags;
3334 }
3335
3336 int __alloc_chunk_for_shrink(struct btrfs_root *root,
3337                      struct btrfs_block_group_cache *shrink_block_group,
3338                      int force)
3339 {
3340         struct btrfs_trans_handle *trans;
3341         u64 new_alloc_flags;
3342         u64 calc;
3343
3344         spin_lock(&shrink_block_group->lock);
3345         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
3346                 spin_unlock(&shrink_block_group->lock);
3347                 mutex_unlock(&root->fs_info->alloc_mutex);
3348
3349                 trans = btrfs_start_transaction(root, 1);
3350                 mutex_lock(&root->fs_info->alloc_mutex);
3351                 spin_lock(&shrink_block_group->lock);
3352
3353                 new_alloc_flags = update_block_group_flags(root,
3354                                                    shrink_block_group->flags);
3355                 if (new_alloc_flags != shrink_block_group->flags) {
3356                         calc =
3357                              btrfs_block_group_used(&shrink_block_group->item);
3358                 } else {
3359                         calc = shrink_block_group->key.offset;
3360                 }
3361                 spin_unlock(&shrink_block_group->lock);
3362
3363                 do_chunk_alloc(trans, root->fs_info->extent_root,
3364                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
3365
3366                 mutex_unlock(&root->fs_info->alloc_mutex);
3367                 btrfs_end_transaction(trans, root);
3368                 mutex_lock(&root->fs_info->alloc_mutex);
3369         } else
3370                 spin_unlock(&shrink_block_group->lock);
3371         return 0;
3372 }
3373
3374 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
3375 {
3376         struct btrfs_trans_handle *trans;
3377         struct btrfs_root *tree_root = root->fs_info->tree_root;
3378         struct btrfs_path *path;
3379         u64 cur_byte;
3380         u64 total_found;
3381         u64 shrink_last_byte;
3382         struct btrfs_block_group_cache *shrink_block_group;
3383         struct btrfs_fs_info *info = root->fs_info;
3384         struct btrfs_key key;
3385         struct btrfs_key found_key;
3386         struct extent_buffer *leaf;
3387         u32 nritems;
3388         int ret;
3389         int progress;
3390
3391         mutex_lock(&root->fs_info->alloc_mutex);
3392         shrink_block_group = btrfs_lookup_block_group(root->fs_info,
3393                                                       shrink_start);
3394         BUG_ON(!shrink_block_group);
3395
3396         shrink_last_byte = shrink_block_group->key.objectid +
3397                 shrink_block_group->key.offset;
3398
3399         shrink_block_group->space_info->total_bytes -=
3400                 shrink_block_group->key.offset;
3401         path = btrfs_alloc_path();
3402         root = root->fs_info->extent_root;
3403         path->reada = 2;
3404
3405         printk("btrfs relocating block group %llu flags %llu\n",
3406                (unsigned long long)shrink_start,
3407                (unsigned long long)shrink_block_group->flags);
3408
3409         __alloc_chunk_for_shrink(root, shrink_block_group, 1);
3410
3411 again:
3412
3413         shrink_block_group->ro = 1;
3414
3415         total_found = 0;
3416         progress = 0;
3417         key.objectid = shrink_start;
3418         key.offset = 0;
3419         key.type = 0;
3420         cur_byte = key.objectid;
3421
3422         mutex_unlock(&root->fs_info->alloc_mutex);
3423
3424         btrfs_start_delalloc_inodes(root);
3425         btrfs_wait_ordered_extents(tree_root, 0);
3426
3427         mutex_lock(&root->fs_info->alloc_mutex);
3428
3429         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3430         if (ret < 0)
3431                 goto out;
3432
3433         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
3434         if (ret < 0)
3435                 goto out;
3436
3437         if (ret == 0) {
3438                 leaf = path->nodes[0];
3439                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3440                 if (found_key.objectid + found_key.offset > shrink_start &&
3441                     found_key.objectid < shrink_last_byte) {
3442                         cur_byte = found_key.objectid;
3443                         key.objectid = cur_byte;
3444                 }
3445         }
3446         btrfs_release_path(root, path);
3447
3448         while(1) {
3449                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3450                 if (ret < 0)
3451                         goto out;
3452
3453 next:
3454                 leaf = path->nodes[0];
3455                 nritems = btrfs_header_nritems(leaf);
3456                 if (path->slots[0] >= nritems) {
3457                         ret = btrfs_next_leaf(root, path);
3458                         if (ret < 0)
3459                                 goto out;
3460                         if (ret == 1) {
3461                                 ret = 0;
3462                                 break;
3463                         }
3464                         leaf = path->nodes[0];
3465                         nritems = btrfs_header_nritems(leaf);
3466                 }
3467
3468                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3469
3470                 if (found_key.objectid >= shrink_last_byte)
3471                         break;
3472
3473                 if (progress && need_resched()) {
3474                         memcpy(&key, &found_key, sizeof(key));
3475                         cond_resched();
3476                         btrfs_release_path(root, path);
3477                         btrfs_search_slot(NULL, root, &key, path, 0, 0);
3478                         progress = 0;
3479                         goto next;
3480                 }
3481                 progress = 1;
3482
3483                 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3484                     found_key.objectid + found_key.offset <= cur_byte) {
3485                         memcpy(&key, &found_key, sizeof(key));
3486                         key.offset++;
3487                         path->slots[0]++;
3488                         goto next;
3489                 }
3490
3491                 total_found++;
3492                 cur_byte = found_key.objectid + found_key.offset;
3493                 key.objectid = cur_byte;
3494                 btrfs_release_path(root, path);
3495                 ret = relocate_one_extent(root, path, &found_key);
3496                 __alloc_chunk_for_shrink(root, shrink_block_group, 0);
3497         }
3498
3499         btrfs_release_path(root, path);
3500
3501         if (total_found > 0) {
3502                 printk("btrfs relocate found %llu last extent was %llu\n",
3503                        (unsigned long long)total_found,
3504                        (unsigned long long)found_key.objectid);
3505                 mutex_unlock(&root->fs_info->alloc_mutex);
3506                 trans = btrfs_start_transaction(tree_root, 1);
3507                 btrfs_commit_transaction(trans, tree_root);
3508
3509                 btrfs_clean_old_snapshots(tree_root);
3510
3511                 btrfs_start_delalloc_inodes(root);
3512                 btrfs_wait_ordered_extents(tree_root, 0);
3513
3514                 trans = btrfs_start_transaction(tree_root, 1);
3515                 btrfs_commit_transaction(trans, tree_root);
3516                 mutex_lock(&root->fs_info->alloc_mutex);
3517                 goto again;
3518         }
3519
3520         /*
3521          * we've freed all the extents, now remove the block
3522          * group item from the tree
3523          */
3524         mutex_unlock(&root->fs_info->alloc_mutex);
3525
3526         trans = btrfs_start_transaction(root, 1);
3527
3528         mutex_lock(&root->fs_info->alloc_mutex);
3529         memcpy(&key, &shrink_block_group->key, sizeof(key));
3530
3531         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3532         if (ret > 0)
3533                 ret = -EIO;
3534         if (ret < 0) {
3535                 btrfs_end_transaction(trans, root);
3536                 goto out;
3537         }
3538
3539         clear_extent_bits(&info->block_group_cache, key.objectid,
3540                           key.objectid + key.offset - 1,
3541                           (unsigned int)-1, GFP_NOFS);
3542
3543
3544         clear_extent_bits(&info->free_space_cache,
3545                            key.objectid, key.objectid + key.offset - 1,
3546                            (unsigned int)-1, GFP_NOFS);
3547
3548         /*
3549         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3550         kfree(shrink_block_group);
3551         */
3552
3553         btrfs_del_item(trans, root, path);
3554         btrfs_release_path(root, path);
3555         mutex_unlock(&root->fs_info->alloc_mutex);
3556         btrfs_commit_transaction(trans, root);
3557
3558         mutex_lock(&root->fs_info->alloc_mutex);
3559
3560         /* the code to unpin extents might set a few bits in the free
3561          * space cache for this range again
3562          */
3563         clear_extent_bits(&info->free_space_cache,
3564                            key.objectid, key.objectid + key.offset - 1,
3565                            (unsigned int)-1, GFP_NOFS);
3566 out:
3567         btrfs_free_path(path);
3568         mutex_unlock(&root->fs_info->alloc_mutex);
3569         return ret;
3570 }
3571
3572 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3573                            struct btrfs_key *key)
3574 {
3575         int ret = 0;
3576         struct btrfs_key found_key;
3577         struct extent_buffer *leaf;
3578         int slot;
3579
3580         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3581         if (ret < 0)
3582                 goto out;
3583
3584         while(1) {
3585                 slot = path->slots[0];
3586                 leaf = path->nodes[0];
3587                 if (slot >= btrfs_header_nritems(leaf)) {
3588                         ret = btrfs_next_leaf(root, path);
3589                         if (ret == 0)
3590                                 continue;
3591                         if (ret < 0)
3592                                 goto out;
3593                         break;
3594                 }
3595                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3596
3597                 if (found_key.objectid >= key->objectid &&
3598                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3599                         ret = 0;
3600                         goto out;
3601                 }
3602                 path->slots[0]++;
3603         }
3604         ret = -ENOENT;
3605 out:
3606         return ret;
3607 }
3608
3609 int btrfs_read_block_groups(struct btrfs_root *root)
3610 {
3611         struct btrfs_path *path;
3612         int ret;
3613         int bit;
3614         struct btrfs_block_group_cache *cache;
3615         struct btrfs_fs_info *info = root->fs_info;
3616         struct btrfs_space_info *space_info;
3617         struct extent_io_tree *block_group_cache;
3618         struct btrfs_key key;
3619         struct btrfs_key found_key;
3620         struct extent_buffer *leaf;
3621
3622         block_group_cache = &info->block_group_cache;
3623         root = info->extent_root;
3624         key.objectid = 0;
3625         key.offset = 0;
3626         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3627         path = btrfs_alloc_path();
3628         if (!path)
3629                 return -ENOMEM;
3630
3631         mutex_lock(&root->fs_info->alloc_mutex);
3632         while(1) {
3633                 ret = find_first_block_group(root, path, &key);
3634                 if (ret > 0) {
3635                         ret = 0;
3636                         goto error;
3637                 }
3638                 if (ret != 0)
3639                         goto error;
3640
3641                 leaf = path->nodes[0];
3642                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3643                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3644                 if (!cache) {
3645                         ret = -ENOMEM;
3646                         break;
3647                 }
3648
3649                 spin_lock_init(&cache->lock);
3650                 read_extent_buffer(leaf, &cache->item,
3651                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3652                                    sizeof(cache->item));
3653                 memcpy(&cache->key, &found_key, sizeof(found_key));
3654
3655                 key.objectid = found_key.objectid + found_key.offset;
3656                 btrfs_release_path(root, path);
3657                 cache->flags = btrfs_block_group_flags(&cache->item);
3658                 bit = 0;
3659                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3660                         bit = BLOCK_GROUP_DATA;
3661                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3662                         bit = BLOCK_GROUP_SYSTEM;
3663                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3664                         bit = BLOCK_GROUP_METADATA;
3665                 }
3666                 set_avail_alloc_bits(info, cache->flags);
3667
3668                 ret = update_space_info(info, cache->flags, found_key.offset,
3669                                         btrfs_block_group_used(&cache->item),
3670                                         &space_info);
3671                 BUG_ON(ret);
3672                 cache->space_info = space_info;
3673
3674                 /* use EXTENT_LOCKED to prevent merging */
3675                 set_extent_bits(block_group_cache, found_key.objectid,
3676                                 found_key.objectid + found_key.offset - 1,
3677                                 EXTENT_LOCKED, GFP_NOFS);
3678                 set_state_private(block_group_cache, found_key.objectid,
3679                                   (unsigned long)cache);
3680                 set_extent_bits(block_group_cache, found_key.objectid,
3681                                 found_key.objectid + found_key.offset - 1,
3682                                 bit | EXTENT_LOCKED, GFP_NOFS);
3683                 if (key.objectid >=
3684                     btrfs_super_total_bytes(&info->super_copy))
3685                         break;
3686         }
3687         ret = 0;
3688 error:
3689         btrfs_free_path(path);
3690         mutex_unlock(&root->fs_info->alloc_mutex);
3691         return ret;
3692 }
3693
3694 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3695                            struct btrfs_root *root, u64 bytes_used,
3696                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3697                            u64 size)
3698 {
3699         int ret;
3700         int bit = 0;
3701         struct btrfs_root *extent_root;
3702         struct btrfs_block_group_cache *cache;
3703         struct extent_io_tree *block_group_cache;
3704
3705         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
3706         extent_root = root->fs_info->extent_root;
3707         block_group_cache = &root->fs_info->block_group_cache;
3708
3709         root->fs_info->last_trans_new_blockgroup = trans->transid;
3710
3711         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3712         BUG_ON(!cache);
3713         cache->key.objectid = chunk_offset;
3714         cache->key.offset = size;
3715         spin_lock_init(&cache->lock);
3716         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3717
3718         btrfs_set_block_group_used(&cache->item, bytes_used);
3719         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3720         cache->flags = type;
3721         btrfs_set_block_group_flags(&cache->item, type);
3722
3723         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3724                                 &cache->space_info);
3725         BUG_ON(ret);
3726
3727         bit = block_group_state_bits(type);
3728         set_extent_bits(block_group_cache, chunk_offset,
3729                         chunk_offset + size - 1,
3730                         EXTENT_LOCKED, GFP_NOFS);
3731         set_state_private(block_group_cache, chunk_offset,
3732                           (unsigned long)cache);
3733         set_extent_bits(block_group_cache, chunk_offset,
3734                         chunk_offset + size - 1,
3735                         bit | EXTENT_LOCKED, GFP_NOFS);
3736
3737         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3738                                 sizeof(cache->item));
3739         BUG_ON(ret);
3740
3741         finish_current_insert(trans, extent_root);
3742         ret = del_pending_extents(trans, extent_root);
3743         BUG_ON(ret);
3744         set_avail_alloc_bits(extent_root->fs_info, type);
3745
3746         return 0;
3747 }