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 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  33                                  btrfs_root *extent_root);
  34 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  35                                btrfs_root *extent_root);
  36 static struct btrfs_block_group_cache *
  37 __btrfs_find_block_group(struct btrfs_root *root,
  38                          struct btrfs_block_group_cache *hint,
  39                          u64 search_start, int data, int owner);
  40
  41 void maybe_lock_mutex(struct btrfs_root *root)
  42 {
  43         if (root != root->fs_info->extent_root &&
  44             root != root->fs_info->chunk_root &&
  45             root != root->fs_info->dev_root) {
  46                 mutex_lock(&root->fs_info->alloc_mutex);
  47         }
  48 }
  49
  50 void maybe_unlock_mutex(struct btrfs_root *root)
  51 {
  52         if (root != root->fs_info->extent_root &&
  53             root != root->fs_info->chunk_root &&
  54             root != root->fs_info->dev_root) {
  55                 mutex_unlock(&root->fs_info->alloc_mutex);
  56         }
  57 }
  58
  59 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  60 {
  61         return (cache->flags & bits) == bits;
  62 }
  63
  64 /*
  65  * this adds the block group to the fs_info rb tree for the block group
  66  * cache
  67  */
  68 int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  69                                 struct btrfs_block_group_cache *block_group)
  70 {
  71         struct rb_node **p;
  72         struct rb_node *parent = NULL;
  73         struct btrfs_block_group_cache *cache;
  74
  75         spin_lock(&info->block_group_cache_lock);
  76         p = &info->block_group_cache_tree.rb_node;
  77
  78         while (*p) {
  79                 parent = *p;
  80                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  81                                  cache_node);
  82                 if (block_group->key.objectid < cache->key.objectid) {
  83                         p = &(*p)->rb_left;
  84                 } else if (block_group->key.objectid > cache->key.objectid) {
  85                         p = &(*p)->rb_right;
  86                 } else {
  87                         spin_unlock(&info->block_group_cache_lock);
  88                         return -EEXIST;
  89                 }
  90         }
  91
  92         rb_link_node(&block_group->cache_node, parent, p);
  93         rb_insert_color(&block_group->cache_node,
  94                         &info->block_group_cache_tree);
  95         spin_unlock(&info->block_group_cache_lock);
  96
  97         return 0;
  98 }
  99
 100 /*
 101  * This will return the block group at or after bytenr if contains is 0, else
 102  * it will return the block group that contains the bytenr
 103  */
 104 static struct btrfs_block_group_cache *
 105 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 106                               int contains)
 107 {
 108         struct btrfs_block_group_cache *cache, *ret = NULL;
 109         struct rb_node *n;
 110         u64 end, start;
 111
 112         spin_lock(&info->block_group_cache_lock);
 113         n = info->block_group_cache_tree.rb_node;
 114
 115         while (n) {
 116                 cache = rb_entry(n, struct btrfs_block_group_cache,
 117                                  cache_node);
 118                 end = cache->key.objectid + cache->key.offset - 1;
 119                 start = cache->key.objectid;
 120
 121                 if (bytenr < start) {
 122                         if (!contains && (!ret || start < ret->key.objectid))
 123                                 ret = cache;
 124                         n = n->rb_left;
 125                 } else if (bytenr > start) {
 126                         if (contains && bytenr <= end) {
 127                                 ret = cache;
 128                                 break;
 129                         }
 130                         n = n->rb_right;
 131                 } else {
 132                         ret = cache;
 133                         break;
 134                 }
 135         }
 136         spin_unlock(&info->block_group_cache_lock);
 137
 138         return ret;
 139 }
 140
 141 /*
 142  * this is only called by cache_block_group, since we could have freed extents
 143  * we need to check the pinned_extents for any extents that can't be used yet
 144  * since their free space will be released as soon as the transaction commits.
 145  */
 146 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 147                               struct btrfs_fs_info *info, u64 start, u64 end)
 148 {
 149         u64 extent_start, extent_end, size;
 150         int ret;
 151
 152         while (start < end) {
 153                 ret = find_first_extent_bit(&info->pinned_extents, start,
 154                                             &extent_start, &extent_end,
 155                                             EXTENT_DIRTY);
 156                 if (ret)
 157                         break;
 158
 159                 if (extent_start == start) {
 160                         start = extent_end + 1;
 161                 } else if (extent_start > start && extent_start < end) {
 162                         size = extent_start - start;
 163                         ret = btrfs_add_free_space(block_group, start, size);
 164                         BUG_ON(ret);
 165                         start = extent_end + 1;
 166                 } else {
 167                         break;
 168                 }
 169         }
 170
 171         if (start < end) {
 172                 size = end - start;
 173                 ret = btrfs_add_free_space(block_group, start, size);
 174                 BUG_ON(ret);
 175         }
 176
 177         return 0;
 178 }
 179
 180 static int cache_block_group(struct btrfs_root *root,
 181                              struct btrfs_block_group_cache *block_group)
 182 {
 183         struct btrfs_path *path;
 184         int ret = 0;
 185         struct btrfs_key key;
 186         struct extent_buffer *leaf;
 187         int slot;
 188         u64 last = 0;
 189         u64 first_free;
 190         int found = 0;
 191
 192         if (!block_group)
 193                 return 0;
 194
 195         root = root->fs_info->extent_root;
 196
 197         if (block_group->cached)
 198                 return 0;
 199
 200         path = btrfs_alloc_path();
 201         if (!path)
 202                 return -ENOMEM;
 203
 204         path->reada = 2;
 205         /*
 206          * we get into deadlocks with paths held by callers of this function.
 207          * since the alloc_mutex is protecting things right now, just
 208          * skip the locking here
 209          */
 210         path->skip_locking = 1;
 211         first_free = max_t(u64, block_group->key.objectid,
 212                            BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 213         key.objectid = block_group->key.objectid;
 214         key.offset = 0;
 215         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 216         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 217         if (ret < 0)
 218                 goto err;
 219         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
 220         if (ret < 0)
 221                 goto err;
 222         if (ret == 0) {
 223                 leaf = path->nodes[0];
 224                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 225                 if (key.objectid + key.offset > first_free)
 226                         first_free = key.objectid + key.offset;
 227         }
 228         while(1) {
 229                 leaf = path->nodes[0];
 230                 slot = path->slots[0];
 231                 if (slot >= btrfs_header_nritems(leaf)) {
 232                         ret = btrfs_next_leaf(root, path);
 233                         if (ret < 0)
 234                                 goto err;
 235                         if (ret == 0)
 236                                 continue;
 237                         else
 238                                 break;
 239                 }
 240                 btrfs_item_key_to_cpu(leaf, &key, slot);
 241                 if (key.objectid < block_group->key.objectid)
 242                         goto next;
 243
 244                 if (key.objectid >= block_group->key.objectid +
 245                     block_group->key.offset)
 246                         break;
 247
 248                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 249                         if (!found) {
 250                                 last = first_free;
 251                                 found = 1;
 252                         }
 253
 254                         add_new_free_space(block_group, root->fs_info, last,
 255                                            key.objectid);
 256
 257                         last = key.objectid + key.offset;
 258                 }
 259 next:
 260                 path->slots[0]++;
 261         }
 262
 263         if (!found)
 264                 last = first_free;
 265
 266         add_new_free_space(block_group, root->fs_info, last,
 267                            block_group->key.objectid +
 268                            block_group->key.offset);
 269
 270         block_group->cached = 1;
 271         ret = 0;
 272 err:
 273         btrfs_free_path(path);
 274         return ret;
 275 }
 276
 277 /*
 278  * return the block group that starts at or after bytenr
 279  */
 280 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 281                                                        btrfs_fs_info *info,
 282                                                          u64 bytenr)
 283 {
 284         struct btrfs_block_group_cache *cache;
 285
 286         cache = block_group_cache_tree_search(info, bytenr, 0);
 287
 288         return cache;
 289 }
 290
 291 /*
 292  * return the block group that contains teh given bytenr
 293  */
 294 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 295                                                          btrfs_fs_info *info,
 296                                                          u64 bytenr)
 297 {
 298         struct btrfs_block_group_cache *cache;
 299
 300         cache = block_group_cache_tree_search(info, bytenr, 1);
 301
 302         return cache;
 303 }
 304
 305 static int noinline find_free_space(struct btrfs_root *root,
 306                                     struct btrfs_block_group_cache **cache_ret,
 307                                     u64 *start_ret, u64 num, int data)
 308 {
 309         int ret;
 310         struct btrfs_block_group_cache *cache = *cache_ret;
 311         struct btrfs_free_space *info = NULL;
 312         u64 last;
 313         u64 total_fs_bytes;
 314         u64 search_start = *start_ret;
 315
 316         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
 317         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 318
 319         if (!cache)
 320                 goto out;
 321
 322         last = max(search_start, cache->key.objectid);
 323
 324 again:
 325         ret = cache_block_group(root, cache);
 326         if (ret)
 327                 goto out;
 328
 329         if (cache->ro || !block_group_bits(cache, data))
 330                 goto new_group;
 331
 332         info = btrfs_find_free_space(cache, last, num);
 333         if (info) {
 334                 *start_ret = info->offset;
 335                 return 0;
 336         }
 337
 338 new_group:
 339         last = cache->key.objectid + cache->key.offset;
 340
 341         cache = btrfs_lookup_first_block_group(root->fs_info, last);
 342         if (!cache || cache->key.objectid >= total_fs_bytes)
 343                 goto out;
 344
 345         *cache_ret = cache;
 346         goto again;
 347
 348 out:
 349         return -ENOSPC;
 350 }
 351
 352 static u64 div_factor(u64 num, int factor)
 353 {
 354         if (factor == 10)
 355                 return num;
 356         num *= factor;
 357         do_div(num, 10);
 358         return num;
 359 }
 360
 361 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 362                                                   u64 flags)
 363 {
 364         struct list_head *head = &info->space_info;
 365         struct list_head *cur;
 366         struct btrfs_space_info *found;
 367         list_for_each(cur, head) {
 368                 found = list_entry(cur, struct btrfs_space_info, list);
 369                 if (found->flags == flags)
 370                         return found;
 371         }
 372         return NULL;
 373
 374 }
 375
 376 static struct btrfs_block_group_cache *
 377 __btrfs_find_block_group(struct btrfs_root *root,
 378                          struct btrfs_block_group_cache *hint,
 379                          u64 search_start, int data, int owner)
 380 {
 381         struct btrfs_block_group_cache *cache;
 382         struct btrfs_block_group_cache *found_group = NULL;
 383         struct btrfs_fs_info *info = root->fs_info;
 384         struct btrfs_space_info *sinfo;
 385         u64 used;
 386         u64 last = 0;
 387         u64 free_check;
 388         int full_search = 0;
 389         int factor = 10;
 390         int wrapped = 0;
 391
 392         if (data & BTRFS_BLOCK_GROUP_METADATA)
 393                 factor = 9;
 394
 395         if (search_start) {
 396                 struct btrfs_block_group_cache *shint;
 397                 shint = btrfs_lookup_first_block_group(info, search_start);
 398                 if (shint && block_group_bits(shint, data) && !shint->ro) {
 399                         spin_lock(&shint->lock);
 400                         used = btrfs_block_group_used(&shint->item);
 401                         if (used + shint->pinned <
 402                             div_factor(shint->key.offset, factor)) {
 403                                 spin_unlock(&shint->lock);
 404                                 return shint;
 405                         }
 406                         spin_unlock(&shint->lock);
 407                 }
 408         }
 409         if (hint && !hint->ro && block_group_bits(hint, data)) {
 410                 spin_lock(&hint->lock);
 411                 used = btrfs_block_group_used(&hint->item);
 412                 if (used + hint->pinned <
 413                     div_factor(hint->key.offset, factor)) {
 414                         spin_unlock(&hint->lock);
 415                         return hint;
 416                 }
 417                 spin_unlock(&hint->lock);
 418                 last = hint->key.objectid + hint->key.offset;
 419         } else {
 420                 if (hint)
 421                         last = max(hint->key.objectid, search_start);
 422                 else
 423                         last = search_start;
 424         }
 425         sinfo = __find_space_info(root->fs_info, data);
 426         if (!sinfo)
 427                 goto found;
 428 again:
 429         while(1) {
 430                 struct list_head *l;
 431
 432                 cache = NULL;
 433
 434                 spin_lock(&sinfo->lock);
 435                 list_for_each(l, &sinfo->block_groups) {
 436                         struct btrfs_block_group_cache *entry;
 437                         entry = list_entry(l, struct btrfs_block_group_cache,
 438                                            list);
 439                         if ((entry->key.objectid >= last) &&
 440                             (!cache || (entry->key.objectid <
 441                                         cache->key.objectid)))
 442                                 cache = entry;
 443                 }
 444                 spin_unlock(&sinfo->lock);
 445
 446                 if (!cache)
 447                         break;
 448
 449                 spin_lock(&cache->lock);
 450                 last = cache->key.objectid + cache->key.offset;
 451                 used = btrfs_block_group_used(&cache->item);
 452
 453                 if (!cache->ro && block_group_bits(cache, data)) {
 454                         free_check = div_factor(cache->key.offset, factor);
 455                         if (used + cache->pinned < free_check) {
 456                                 found_group = cache;
 457                                 spin_unlock(&cache->lock);
 458                                 goto found;
 459                         }
 460                 }
 461                 spin_unlock(&cache->lock);
 462                 cond_resched();
 463         }
 464         if (!wrapped) {
 465                 last = search_start;
 466                 wrapped = 1;
 467                 goto again;
 468         }
 469         if (!full_search && factor < 10) {
 470                 last = search_start;
 471                 full_search = 1;
 472                 factor = 10;
 473                 goto again;
 474         }
 475 found:
 476         return found_group;
 477 }
 478
 479 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 480                                                  struct btrfs_block_group_cache
 481                                                  *hint, u64 search_start,
 482                                                  int data, int owner)
 483 {
 484
 485         struct btrfs_block_group_cache *ret;
 486         ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
 487         return ret;
 488 }
 489
 490 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
 491                            u64 owner, u64 owner_offset)
 492 {
 493         u32 high_crc = ~(u32)0;
 494         u32 low_crc = ~(u32)0;
 495         __le64 lenum;
 496         lenum = cpu_to_le64(root_objectid);
 497         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
 498         lenum = cpu_to_le64(ref_generation);
 499         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 500         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
 501                 lenum = cpu_to_le64(owner);
 502                 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 503                 lenum = cpu_to_le64(owner_offset);
 504                 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 505         }
 506         return ((u64)high_crc << 32) | (u64)low_crc;
 507 }
 508
 509 static int match_extent_ref(struct extent_buffer *leaf,
 510                             struct btrfs_extent_ref *disk_ref,
 511                             struct btrfs_extent_ref *cpu_ref)
 512 {
 513         int ret;
 514         int len;
 515
 516         if (cpu_ref->objectid)
 517                 len = sizeof(*cpu_ref);
 518         else
 519                 len = 2 * sizeof(u64);
 520         ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
 521                                    len);
 522         return ret == 0;
 523 }
 524
 525 /* simple helper to search for an existing extent at a given offset */
 526 int btrfs_lookup_extent(struct btrfs_root *root, struct btrfs_path *path,
 527                         u64 start, u64 len)
 528 {
 529         int ret;
 530         struct btrfs_key key;
 531
 532         maybe_lock_mutex(root);
 533         key.objectid = start;
 534         key.offset = len;
 535         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 536         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 537                                 0, 0);
 538         maybe_unlock_mutex(root);
 539         return ret;
 540 }
 541
 542 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 543                                           struct btrfs_root *root,
 544                                           struct btrfs_path *path, u64 bytenr,
 545                                           u64 root_objectid,
 546                                           u64 ref_generation, u64 owner,
 547                                           u64 owner_offset, int del)
 548 {
 549         u64 hash;
 550         struct btrfs_key key;
 551         struct btrfs_key found_key;
 552         struct btrfs_extent_ref ref;
 553         struct extent_buffer *leaf;
 554         struct btrfs_extent_ref *disk_ref;
 555         int ret;
 556         int ret2;
 557
 558         btrfs_set_stack_ref_root(&ref, root_objectid);
 559         btrfs_set_stack_ref_generation(&ref, ref_generation);
 560         btrfs_set_stack_ref_objectid(&ref, owner);
 561         btrfs_set_stack_ref_offset(&ref, owner_offset);
 562
 563         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 564                                owner_offset);
 565         key.offset = hash;
 566         key.objectid = bytenr;
 567         key.type = BTRFS_EXTENT_REF_KEY;
 568
 569         while (1) {
 570                 ret = btrfs_search_slot(trans, root, &key, path,
 571                                         del ? -1 : 0, del);
 572                 if (ret < 0)
 573                         goto out;
 574                 leaf = path->nodes[0];
 575                 if (ret != 0) {
 576                         u32 nritems = btrfs_header_nritems(leaf);
 577                         if (path->slots[0] >= nritems) {
 578                                 ret2 = btrfs_next_leaf(root, path);
 579                                 if (ret2)
 580                                         goto out;
 581                                 leaf = path->nodes[0];
 582                         }
 583                         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 584                         if (found_key.objectid != bytenr ||
 585                             found_key.type != BTRFS_EXTENT_REF_KEY)
 586                                 goto out;
 587                         key.offset = found_key.offset;
 588                         if (del) {
 589                                 btrfs_release_path(root, path);
 590                                 continue;
 591                         }
 592                 }
 593                 disk_ref = btrfs_item_ptr(path->nodes[0],
 594                                           path->slots[0],
 595                                           struct btrfs_extent_ref);
 596                 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
 597                         ret = 0;
 598                         goto out;
 599                 }
 600                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 601                 key.offset = found_key.offset + 1;
 602                 btrfs_release_path(root, path);
 603         }
 604 out:
 605         return ret;
 606 }
 607
 608 /*
 609  * Back reference rules.  Back refs have three main goals:
 610  *
 611  * 1) differentiate between all holders of references to an extent so that
 612  *    when a reference is dropped we can make sure it was a valid reference
 613  *    before freeing the extent.
 614  *
 615  * 2) Provide enough information to quickly find the holders of an extent
 616  *    if we notice a given block is corrupted or bad.
 617  *
 618  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 619  *    maintenance.  This is actually the same as #2, but with a slightly
 620  *    different use case.
 621  *
 622  * File extents can be referenced by:
 623  *
 624  * - multiple snapshots, subvolumes, or different generations in one subvol
 625  * - different files inside a single subvolume (in theory, not implemented yet)
 626  * - different offsets inside a file (bookend extents in file.c)
 627  *
 628  * The extent ref structure has fields for:
 629  *
 630  * - Objectid of the subvolume root
 631  * - Generation number of the tree holding the reference
 632  * - objectid of the file holding the reference
 633  * - offset in the file corresponding to the key holding the reference
 634  *
 635  * When a file extent is allocated the fields are filled in:
 636  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 637  *
 638  * When a leaf is cow'd new references are added for every file extent found
 639  * in the leaf.  It looks the same as the create case, but trans->transid
 640  * will be different when the block is cow'd.
 641  *
 642  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
 643  *
 644  * When a file extent is removed either during snapshot deletion or file
 645  * truncation, the corresponding back reference is found
 646  * by searching for:
 647  *
 648  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 649  *      inode objectid, offset in file)
 650  *
 651  * Btree extents can be referenced by:
 652  *
 653  * - Different subvolumes
 654  * - Different generations of the same subvolume
 655  *
 656  * Storing sufficient information for a full reverse mapping of a btree
 657  * block would require storing the lowest key of the block in the backref,
 658  * and it would require updating that lowest key either before write out or
 659  * every time it changed.  Instead, the objectid of the lowest key is stored
 660  * along with the level of the tree block.  This provides a hint
 661  * about where in the btree the block can be found.  Searches through the
 662  * btree only need to look for a pointer to that block, so they stop one
 663  * level higher than the level recorded in the backref.
 664  *
 665  * Some btrees do not do reference counting on their extents.  These
 666  * include the extent tree and the tree of tree roots.  Backrefs for these
 667  * trees always have a generation of zero.
 668  *
 669  * When a tree block is created, back references are inserted:
 670  *
 671  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
 672  *
 673  * When a tree block is cow'd in a reference counted root,
 674  * new back references are added for all the blocks it points to.
 675  * These are of the form (trans->transid will have increased since creation):
 676  *
 677  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
 678  *
 679  * Because the lowest_key_objectid and the level are just hints
 680  * they are not used when backrefs are deleted.  When a backref is deleted:
 681  *
 682  * if backref was for a tree root:
 683  *     root_objectid = root->root_key.objectid
 684  * else
 685  *     root_objectid = btrfs_header_owner(parent)
 686  *
 687  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
 688  *
 689  * Back Reference Key hashing:
 690  *
 691  * Back references have four fields, each 64 bits long.  Unfortunately,
 692  * This is hashed into a single 64 bit number and placed into the key offset.
 693  * The key objectid corresponds to the first byte in the extent, and the
 694  * key type is set to BTRFS_EXTENT_REF_KEY
 695  */
 696 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
 697                                  struct btrfs_root *root,
 698                                  struct btrfs_path *path, u64 bytenr,
 699                                  u64 root_objectid, u64 ref_generation,
 700                                  u64 owner, u64 owner_offset)
 701 {
 702         u64 hash;
 703         struct btrfs_key key;
 704         struct btrfs_extent_ref ref;
 705         struct btrfs_extent_ref *disk_ref;
 706         int ret;
 707
 708         btrfs_set_stack_ref_root(&ref, root_objectid);
 709         btrfs_set_stack_ref_generation(&ref, ref_generation);
 710         btrfs_set_stack_ref_objectid(&ref, owner);
 711         btrfs_set_stack_ref_offset(&ref, owner_offset);
 712
 713         hash = hash_extent_ref(root_objectid, ref_generation, owner,
 714                                owner_offset);
 715         key.offset = hash;
 716         key.objectid = bytenr;
 717         key.type = BTRFS_EXTENT_REF_KEY;
 718
 719         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
 720         while (ret == -EEXIST) {
 721                 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 722                                           struct btrfs_extent_ref);
 723                 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
 724                         goto out;
 725                 key.offset++;
 726                 btrfs_release_path(root, path);
 727                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 728                                               sizeof(ref));
 729         }
 730         if (ret)
 731                 goto out;
 732         disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 733                                   struct btrfs_extent_ref);
 734         write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
 735                             sizeof(ref));
 736         btrfs_mark_buffer_dirty(path->nodes[0]);
 737 out:
 738         btrfs_release_path(root, path);
 739         return ret;
 740 }
 741
 742 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 743                                 struct btrfs_root *root,
 744                                 u64 bytenr, u64 num_bytes,
 745                                 u64 root_objectid, u64 ref_generation,
 746                                 u64 owner, u64 owner_offset)
 747 {
 748         struct btrfs_path *path;
 749         int ret;
 750         struct btrfs_key key;
 751         struct extent_buffer *l;
 752         struct btrfs_extent_item *item;
 753         u32 refs;
 754
 755         WARN_ON(num_bytes < root->sectorsize);
 756         path = btrfs_alloc_path();
 757         if (!path)
 758                 return -ENOMEM;
 759
 760         path->reada = 1;
 761         key.objectid = bytenr;
 762         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 763         key.offset = num_bytes;
 764         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 765                                 0, 1);
 766         if (ret < 0)
 767                 return ret;
 768         if (ret != 0) {
 769                 BUG();
 770         }
 771         BUG_ON(ret != 0);
 772         l = path->nodes[0];
 773         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 774         refs = btrfs_extent_refs(l, item);
 775         btrfs_set_extent_refs(l, item, refs + 1);
 776         btrfs_mark_buffer_dirty(path->nodes[0]);
 777
 778         btrfs_release_path(root->fs_info->extent_root, path);
 779
 780         path->reada = 1;
 781         ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
 782                                           path, bytenr, root_objectid,
 783                                           ref_generation, owner, owner_offset);
 784         BUG_ON(ret);
 785         finish_current_insert(trans, root->fs_info->extent_root);
 786         del_pending_extents(trans, root->fs_info->extent_root);
 787
 788         btrfs_free_path(path);
 789         return 0;
 790 }
 791
 792 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 793                                 struct btrfs_root *root,
 794                                 u64 bytenr, u64 num_bytes,
 795                                 u64 root_objectid, u64 ref_generation,
 796                                 u64 owner, u64 owner_offset)
 797 {
 798         int ret;
 799
 800         mutex_lock(&root->fs_info->alloc_mutex);
 801         ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
 802                                      root_objectid, ref_generation,
 803                                      owner, owner_offset);
 804         mutex_unlock(&root->fs_info->alloc_mutex);
 805         return ret;
 806 }
 807
 808 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 809                          struct btrfs_root *root)
 810 {
 811         finish_current_insert(trans, root->fs_info->extent_root);
 812         del_pending_extents(trans, root->fs_info->extent_root);
 813         return 0;
 814 }
 815
 816 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
 817                              struct btrfs_root *root, u64 bytenr,
 818                              u64 num_bytes, u32 *refs)
 819 {
 820         struct btrfs_path *path;
 821         int ret;
 822         struct btrfs_key key;
 823         struct extent_buffer *l;
 824         struct btrfs_extent_item *item;
 825
 826         WARN_ON(num_bytes < root->sectorsize);
 827         path = btrfs_alloc_path();
 828         path->reada = 1;
 829         key.objectid = bytenr;
 830         key.offset = num_bytes;
 831         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 832         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 833                                 0, 0);
 834         if (ret < 0)
 835                 goto out;
 836         if (ret != 0) {
 837                 btrfs_print_leaf(root, path->nodes[0]);
 838                 printk("failed to find block number %Lu\n", bytenr);
 839                 BUG();
 840         }
 841         l = path->nodes[0];
 842         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 843         *refs = btrfs_extent_refs(l, item);
 844 out:
 845         btrfs_free_path(path);
 846         return 0;
 847 }
 848
 849
 850 static int get_reference_status(struct btrfs_root *root, u64 bytenr,
 851                                 u64 parent_gen, u64 ref_objectid,
 852                                 u64 *min_generation, u32 *ref_count)
 853 {
 854         struct btrfs_root *extent_root = root->fs_info->extent_root;
 855         struct btrfs_path *path;
 856         struct extent_buffer *leaf;
 857         struct btrfs_extent_ref *ref_item;
 858         struct btrfs_key key;
 859         struct btrfs_key found_key;
 860         u64 root_objectid = root->root_key.objectid;
 861         u64 ref_generation;
 862         u32 nritems;
 863         int ret;
 864
 865         key.objectid = bytenr;
 866         key.offset = 0;
 867         key.type = BTRFS_EXTENT_ITEM_KEY;
 868
 869         path = btrfs_alloc_path();
 870         mutex_lock(&root->fs_info->alloc_mutex);
 871         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 872         if (ret < 0)
 873                 goto out;
 874         BUG_ON(ret == 0);
 875
 876         leaf = path->nodes[0];
 877         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 878
 879         if (found_key.objectid != bytenr ||
 880             found_key.type != BTRFS_EXTENT_ITEM_KEY) {
 881                 ret = 1;
 882                 goto out;
 883         }
 884
 885         *ref_count = 0;
 886         *min_generation = (u64)-1;
 887
 888         while (1) {
 889                 leaf = path->nodes[0];
 890                 nritems = btrfs_header_nritems(leaf);
 891                 if (path->slots[0] >= nritems) {
 892                         ret = btrfs_next_leaf(extent_root, path);
 893                         if (ret < 0)
 894                                 goto out;
 895                         if (ret == 0)
 896                                 continue;
 897                         break;
 898                 }
 899                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 900                 if (found_key.objectid != bytenr)
 901                         break;
 902
 903                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 904                         path->slots[0]++;
 905                         continue;
 906                 }
 907
 908                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
 909                                           struct btrfs_extent_ref);
 910                 ref_generation = btrfs_ref_generation(leaf, ref_item);
 911                 /*
 912                  * For (parent_gen > 0 && parent_gen > ref_gen):
 913                  *
 914                  * we reach here through the oldest root, therefore
 915                  * all other reference from same snapshot should have
 916                  * a larger generation.
 917                  */
 918                 if ((root_objectid != btrfs_ref_root(leaf, ref_item)) ||
 919                     (parent_gen > 0 && parent_gen > ref_generation) ||
 920                     (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
 921                      ref_objectid != btrfs_ref_objectid(leaf, ref_item))) {
 922                         if (ref_count)
 923                                 *ref_count = 2;
 924                         break;
 925                 }
 926
 927                 *ref_count = 1;
 928                 if (*min_generation > ref_generation)
 929                         *min_generation = ref_generation;
 930
 931                 path->slots[0]++;
 932         }
 933         ret = 0;
 934 out:
 935         mutex_unlock(&root->fs_info->alloc_mutex);
 936         btrfs_free_path(path);
 937         return ret;
 938 }
 939
 940 int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans,
 941                            struct btrfs_root *root,
 942                            struct btrfs_key *key, u64 bytenr)
 943 {
 944         struct btrfs_root *old_root;
 945         struct btrfs_path *path = NULL;
 946         struct extent_buffer *eb;
 947         struct btrfs_file_extent_item *item;
 948         u64 ref_generation;
 949         u64 min_generation;
 950         u64 extent_start;
 951         u32 ref_count;
 952         int level;
 953         int ret;
 954
 955         BUG_ON(trans == NULL);
 956         BUG_ON(key->type != BTRFS_EXTENT_DATA_KEY);
 957         ret = get_reference_status(root, bytenr, 0, key->objectid,
 958                                    &min_generation, &ref_count);
 959         if (ret)
 960                 return ret;
 961
 962         if (ref_count != 1)
 963                 return 1;
 964
 965         old_root = root->dirty_root->root;
 966         ref_generation = old_root->root_key.offset;
 967
 968         /* all references are created in running transaction */
 969         if (min_generation > ref_generation) {
 970                 ret = 0;
 971                 goto out;
 972         }
 973
 974         path = btrfs_alloc_path();
 975         if (!path) {
 976                 ret = -ENOMEM;
 977                 goto out;
 978         }
 979
 980         path->skip_locking = 1;
 981         /* if no item found, the extent is referenced by other snapshot */
 982         ret = btrfs_search_slot(NULL, old_root, key, path, 0, 0);
 983         if (ret)
 984                 goto out;
 985
 986         eb = path->nodes[0];
 987         item = btrfs_item_ptr(eb, path->slots[0],
 988                               struct btrfs_file_extent_item);
 989         if (btrfs_file_extent_type(eb, item) != BTRFS_FILE_EXTENT_REG ||
 990             btrfs_file_extent_disk_bytenr(eb, item) != bytenr) {
 991                 ret = 1;
 992                 goto out;
 993         }
 994
 995         for (level = BTRFS_MAX_LEVEL - 1; level >= -1; level--) {
 996                 if (level >= 0) {
 997                         eb = path->nodes[level];
 998                         if (!eb)
 999                                 continue;
1000                         extent_start = eb->start;
1001                 } else
1002                         extent_start = bytenr;
1003
1004                 ret = get_reference_status(root, extent_start, ref_generation,
1005                                            0, &min_generation, &ref_count);
1006                 if (ret)
1007                         goto out;
1008
1009                 if (ref_count != 1) {
1010                         ret = 1;
1011                         goto out;
1012                 }
1013                 if (level >= 0)
1014                         ref_generation = btrfs_header_generation(eb);
1015         }
1016         ret = 0;
1017 out:
1018         if (path)
1019                 btrfs_free_path(path);
1020         return ret;
1021 }
1022
1023 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1024                   struct extent_buffer *buf, int cache_ref)
1025 {
1026         u64 bytenr;
1027         u32 nritems;
1028         struct btrfs_key key;
1029         struct btrfs_file_extent_item *fi;
1030         int i;
1031         int level;
1032         int ret;
1033         int faili;
1034         int nr_file_extents = 0;
1035
1036         if (!root->ref_cows)
1037                 return 0;
1038
1039         level = btrfs_header_level(buf);
1040         nritems = btrfs_header_nritems(buf);
1041         for (i = 0; i < nritems; i++) {
1042                 cond_resched();
1043                 if (level == 0) {
1044                         u64 disk_bytenr;
1045                         btrfs_item_key_to_cpu(buf, &key, i);
1046                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1047                                 continue;
1048                         fi = btrfs_item_ptr(buf, i,
1049                                             struct btrfs_file_extent_item);
1050                         if (btrfs_file_extent_type(buf, fi) ==
1051                             BTRFS_FILE_EXTENT_INLINE)
1052                                 continue;
1053                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1054                         if (disk_bytenr == 0)
1055                                 continue;
1056
1057                         if (buf != root->commit_root)
1058                                 nr_file_extents++;
1059
1060                         mutex_lock(&root->fs_info->alloc_mutex);
1061                         ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr,
1062                                     btrfs_file_extent_disk_num_bytes(buf, fi),
1063                                     root->root_key.objectid, trans->transid,
1064                                     key.objectid, key.offset);
1065                         mutex_unlock(&root->fs_info->alloc_mutex);
1066                         if (ret) {
1067                                 faili = i;
1068                                 WARN_ON(1);
1069                                 goto fail;
1070                         }
1071                 } else {
1072                         bytenr = btrfs_node_blockptr(buf, i);
1073                         btrfs_node_key_to_cpu(buf, &key, i);
1074
1075                         mutex_lock(&root->fs_info->alloc_mutex);
1076                         ret = __btrfs_inc_extent_ref(trans, root, bytenr,
1077                                            btrfs_level_size(root, level - 1),
1078                                            root->root_key.objectid,
1079                                            trans->transid,
1080                                            level - 1, key.objectid);
1081                         mutex_unlock(&root->fs_info->alloc_mutex);
1082                         if (ret) {
1083                                 faili = i;
1084                                 WARN_ON(1);
1085                                 goto fail;
1086                         }
1087                 }
1088         }
1089         /* cache orignal leaf block's references */
1090         if (level == 0 && cache_ref && buf != root->commit_root) {
1091                 struct btrfs_leaf_ref *ref;
1092                 struct btrfs_extent_info *info;
1093
1094                 ref = btrfs_alloc_leaf_ref(root, nr_file_extents);
1095                 if (!ref) {
1096                         WARN_ON(1);
1097                         goto out;
1098                 }
1099
1100                 ref->root_gen = root->root_key.offset;
1101                 ref->bytenr = buf->start;
1102                 ref->owner = btrfs_header_owner(buf);
1103                 ref->generation = btrfs_header_generation(buf);
1104                 ref->nritems = nr_file_extents;
1105                 info = ref->extents;
1106
1107                 for (i = 0; nr_file_extents > 0 && i < nritems; i++) {
1108                         u64 disk_bytenr;
1109                         btrfs_item_key_to_cpu(buf, &key, i);
1110                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1111                                 continue;
1112                         fi = btrfs_item_ptr(buf, i,
1113                                             struct btrfs_file_extent_item);
1114                         if (btrfs_file_extent_type(buf, fi) ==
1115                             BTRFS_FILE_EXTENT_INLINE)
1116                                 continue;
1117                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1118                         if (disk_bytenr == 0)
1119                                 continue;
1120
1121                         info->bytenr = disk_bytenr;
1122                         info->num_bytes =
1123                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1124                         info->objectid = key.objectid;
1125                         info->offset = key.offset;
1126                         info++;
1127                 }
1128
1129                 BUG_ON(!root->ref_tree);
1130                 ret = btrfs_add_leaf_ref(root, ref);
1131                 WARN_ON(ret);
1132                 btrfs_free_leaf_ref(root, ref);
1133         }
1134 out:
1135         return 0;
1136 fail:
1137         WARN_ON(1);
1138 #if 0
1139         for (i =0; i < faili; i++) {
1140                 if (level == 0) {
1141                         u64 disk_bytenr;
1142                         btrfs_item_key_to_cpu(buf, &key, i);
1143                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1144                                 continue;
1145                         fi = btrfs_item_ptr(buf, i,
1146                                             struct btrfs_file_extent_item);
1147                         if (btrfs_file_extent_type(buf, fi) ==
1148                             BTRFS_FILE_EXTENT_INLINE)
1149                                 continue;
1150                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1151                         if (disk_bytenr == 0)
1152                                 continue;
1153                         err = btrfs_free_extent(trans, root, disk_bytenr,
1154                                     btrfs_file_extent_disk_num_bytes(buf,
1155                                                                       fi), 0);
1156                         BUG_ON(err);
1157                 } else {
1158                         bytenr = btrfs_node_blockptr(buf, i);
1159                         err = btrfs_free_extent(trans, root, bytenr,
1160                                         btrfs_level_size(root, level - 1), 0);
1161                         BUG_ON(err);
1162                 }
1163         }
1164 #endif
1165         return ret;
1166 }
1167
1168 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1169                                  struct btrfs_root *root,
1170                                  struct btrfs_path *path,
1171                                  struct btrfs_block_group_cache *cache)
1172 {
1173         int ret;
1174         int pending_ret;
1175         struct btrfs_root *extent_root = root->fs_info->extent_root;
1176         unsigned long bi;
1177         struct extent_buffer *leaf;
1178
1179         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1180         if (ret < 0)
1181                 goto fail;
1182         BUG_ON(ret);
1183
1184         leaf = path->nodes[0];
1185         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1186         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1187         btrfs_mark_buffer_dirty(leaf);
1188         btrfs_release_path(extent_root, path);
1189 fail:
1190         finish_current_insert(trans, extent_root);
1191         pending_ret = del_pending_extents(trans, extent_root);
1192         if (ret)
1193                 return ret;
1194         if (pending_ret)
1195                 return pending_ret;
1196         return 0;
1197
1198 }
1199
1200 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1201                                    struct btrfs_root *root)
1202 {
1203         struct btrfs_block_group_cache *cache, *entry;
1204         struct rb_node *n;
1205         int err = 0;
1206         int werr = 0;
1207         struct btrfs_path *path;
1208         u64 last = 0;
1209
1210         path = btrfs_alloc_path();
1211         if (!path)
1212                 return -ENOMEM;
1213
1214         mutex_lock(&root->fs_info->alloc_mutex);
1215         while(1) {
1216                 cache = NULL;
1217                 spin_lock(&root->fs_info->block_group_cache_lock);
1218                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1219                      n; n = rb_next(n)) {
1220                         entry = rb_entry(n, struct btrfs_block_group_cache,
1221                                          cache_node);
1222                         if (entry->dirty) {
1223                                 cache = entry;
1224                                 break;
1225                         }
1226                 }
1227                 spin_unlock(&root->fs_info->block_group_cache_lock);
1228
1229                 if (!cache)
1230                         break;
1231
1232                 last += cache->key.offset;
1233
1234                 err = write_one_cache_group(trans, root,
1235                                             path, cache);
1236                 /*
1237                  * if we fail to write the cache group, we want
1238                  * to keep it marked dirty in hopes that a later
1239                  * write will work
1240                  */
1241                 if (err) {
1242                         werr = err;
1243                         continue;
1244                 }
1245
1246                 cache->dirty = 0;
1247         }
1248         btrfs_free_path(path);
1249         mutex_unlock(&root->fs_info->alloc_mutex);
1250         return werr;
1251 }
1252
1253 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1254                              u64 total_bytes, u64 bytes_used,
1255                              struct btrfs_space_info **space_info)
1256 {
1257         struct btrfs_space_info *found;
1258
1259         found = __find_space_info(info, flags);
1260         if (found) {
1261                 found->total_bytes += total_bytes;
1262                 found->bytes_used += bytes_used;
1263                 found->full = 0;
1264                 *space_info = found;
1265                 return 0;
1266         }
1267         found = kmalloc(sizeof(*found), GFP_NOFS);
1268         if (!found)
1269                 return -ENOMEM;
1270
1271         list_add(&found->list, &info->space_info);
1272         INIT_LIST_HEAD(&found->block_groups);
1273         spin_lock_init(&found->lock);
1274         found->flags = flags;
1275         found->total_bytes = total_bytes;
1276         found->bytes_used = bytes_used;
1277         found->bytes_pinned = 0;
1278         found->full = 0;
1279         found->force_alloc = 0;
1280         *space_info = found;
1281         return 0;
1282 }
1283
1284 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1285 {
1286         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1287                                    BTRFS_BLOCK_GROUP_RAID1 |
1288                                    BTRFS_BLOCK_GROUP_RAID10 |
1289                                    BTRFS_BLOCK_GROUP_DUP);
1290         if (extra_flags) {
1291                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1292                         fs_info->avail_data_alloc_bits |= extra_flags;
1293                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1294                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1295                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1296                         fs_info->avail_system_alloc_bits |= extra_flags;
1297         }
1298 }
1299
1300 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1301 {
1302         u64 num_devices = root->fs_info->fs_devices->num_devices;
1303
1304         if (num_devices == 1)
1305                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1306         if (num_devices < 4)
1307                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1308
1309         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1310             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1311                       BTRFS_BLOCK_GROUP_RAID10))) {
1312                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1313         }
1314
1315         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1316             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1317                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1318         }
1319
1320         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1321             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1322              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1323              (flags & BTRFS_BLOCK_GROUP_DUP)))
1324                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1325         return flags;
1326 }
1327
1328 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1329                           struct btrfs_root *extent_root, u64 alloc_bytes,
1330                           u64 flags, int force)
1331 {
1332         struct btrfs_space_info *space_info;
1333         u64 thresh;
1334         u64 start;
1335         u64 num_bytes;
1336         int ret = 0;
1337
1338         flags = reduce_alloc_profile(extent_root, flags);
1339
1340         space_info = __find_space_info(extent_root->fs_info, flags);
1341         if (!space_info) {
1342                 ret = update_space_info(extent_root->fs_info, flags,
1343                                         0, 0, &space_info);
1344                 BUG_ON(ret);
1345         }
1346         BUG_ON(!space_info);
1347
1348         if (space_info->force_alloc) {
1349                 force = 1;
1350                 space_info->force_alloc = 0;
1351         }
1352         if (space_info->full)
1353                 goto out;
1354
1355         thresh = div_factor(space_info->total_bytes, 6);
1356         if (!force &&
1357            (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1358             thresh)
1359                 goto out;
1360
1361         mutex_lock(&extent_root->fs_info->chunk_mutex);
1362         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1363         if (ret == -ENOSPC) {
1364 printk("space info full %Lu\n", flags);
1365                 space_info->full = 1;
1366                 goto out_unlock;
1367         }
1368         BUG_ON(ret);
1369
1370         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1371                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1372         BUG_ON(ret);
1373
1374 out_unlock:
1375         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1376 out:
1377         return ret;
1378 }
1379
1380 static int update_block_group(struct btrfs_trans_handle *trans,
1381                               struct btrfs_root *root,
1382                               u64 bytenr, u64 num_bytes, int alloc,
1383                               int mark_free)
1384 {
1385         struct btrfs_block_group_cache *cache;
1386         struct btrfs_fs_info *info = root->fs_info;
1387         u64 total = num_bytes;
1388         u64 old_val;
1389         u64 byte_in_group;
1390
1391         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1392         while(total) {
1393                 cache = btrfs_lookup_block_group(info, bytenr);
1394                 if (!cache) {
1395                         return -1;
1396                 }
1397                 byte_in_group = bytenr - cache->key.objectid;
1398                 WARN_ON(byte_in_group > cache->key.offset);
1399
1400                 spin_lock(&cache->lock);
1401                 cache->dirty = 1;
1402                 old_val = btrfs_block_group_used(&cache->item);
1403                 num_bytes = min(total, cache->key.offset - byte_in_group);
1404                 if (alloc) {
1405                         old_val += num_bytes;
1406                         cache->space_info->bytes_used += num_bytes;
1407                         btrfs_set_block_group_used(&cache->item, old_val);
1408                         spin_unlock(&cache->lock);
1409                 } else {
1410                         old_val -= num_bytes;
1411                         cache->space_info->bytes_used -= num_bytes;
1412                         btrfs_set_block_group_used(&cache->item, old_val);
1413                         spin_unlock(&cache->lock);
1414                         if (mark_free) {
1415                                 int ret;
1416                                 ret = btrfs_add_free_space(cache, bytenr,
1417                                                            num_bytes);
1418                                 if (ret)
1419                                         return -1;
1420                         }
1421                 }
1422                 total -= num_bytes;
1423                 bytenr += num_bytes;
1424         }
1425         return 0;
1426 }
1427
1428 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1429 {
1430         struct btrfs_block_group_cache *cache;
1431
1432         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1433         if (!cache)
1434                 return 0;
1435
1436         return cache->key.objectid;
1437 }
1438
1439
1440 int btrfs_update_pinned_extents(struct btrfs_root *root,
1441                                 u64 bytenr, u64 num, int pin)
1442 {
1443         u64 len;
1444         struct btrfs_block_group_cache *cache;
1445         struct btrfs_fs_info *fs_info = root->fs_info;
1446
1447         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1448         if (pin) {
1449                 set_extent_dirty(&fs_info->pinned_extents,
1450                                 bytenr, bytenr + num - 1, GFP_NOFS);
1451         } else {
1452                 clear_extent_dirty(&fs_info->pinned_extents,
1453                                 bytenr, bytenr + num - 1, GFP_NOFS);
1454         }
1455         while (num > 0) {
1456                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1457                 if (!cache) {
1458                         u64 first = first_logical_byte(root, bytenr);
1459                         WARN_ON(first < bytenr);
1460                         len = min(first - bytenr, num);
1461                 } else {
1462                         len = min(num, cache->key.offset -
1463                                   (bytenr - cache->key.objectid));
1464                 }
1465                 if (pin) {
1466                         if (cache) {
1467                                 spin_lock(&cache->lock);
1468                                 cache->pinned += len;
1469                                 cache->space_info->bytes_pinned += len;
1470                                 spin_unlock(&cache->lock);
1471                         }
1472                         fs_info->total_pinned += len;
1473                 } else {
1474                         if (cache) {
1475                                 spin_lock(&cache->lock);
1476                                 cache->pinned -= len;
1477                                 cache->space_info->bytes_pinned -= len;
1478                                 spin_unlock(&cache->lock);
1479                         }
1480                         fs_info->total_pinned -= len;
1481                 }
1482                 bytenr += len;
1483                 num -= len;
1484         }
1485         return 0;
1486 }
1487
1488 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1489 {
1490         u64 last = 0;
1491         u64 start;
1492         u64 end;
1493         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1494         int ret;
1495
1496         while(1) {
1497                 ret = find_first_extent_bit(pinned_extents, last,
1498                                             &start, &end, EXTENT_DIRTY);
1499                 if (ret)
1500                         break;
1501                 set_extent_dirty(copy, start, end, GFP_NOFS);
1502                 last = end + 1;
1503         }
1504         return 0;
1505 }
1506
1507 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1508                                struct btrfs_root *root,
1509                                struct extent_io_tree *unpin)
1510 {
1511         u64 start;
1512         u64 end;
1513         int ret;
1514         struct btrfs_block_group_cache *cache;
1515
1516         mutex_lock(&root->fs_info->alloc_mutex);
1517         while(1) {
1518                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1519                                             EXTENT_DIRTY);
1520                 if (ret)
1521                         break;
1522                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1523                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1524                 cache = btrfs_lookup_block_group(root->fs_info, start);
1525                 if (cache->cached)
1526                         btrfs_add_free_space(cache, start, end - start + 1);
1527                 if (need_resched()) {
1528                         mutex_unlock(&root->fs_info->alloc_mutex);
1529                         cond_resched();
1530                         mutex_lock(&root->fs_info->alloc_mutex);
1531                 }
1532         }
1533         mutex_unlock(&root->fs_info->alloc_mutex);
1534         return 0;
1535 }
1536
1537 static int finish_current_insert(struct btrfs_trans_handle *trans,
1538                                  struct btrfs_root *extent_root)
1539 {
1540         u64 start;
1541         u64 end;
1542         struct btrfs_fs_info *info = extent_root->fs_info;
1543         struct extent_buffer *eb;
1544         struct btrfs_path *path;
1545         struct btrfs_key ins;
1546         struct btrfs_disk_key first;
1547         struct btrfs_extent_item extent_item;
1548         int ret;
1549         int level;
1550         int err = 0;
1551
1552         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1553         btrfs_set_stack_extent_refs(&extent_item, 1);
1554         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1555         path = btrfs_alloc_path();
1556
1557         while(1) {
1558                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1559                                             &end, EXTENT_LOCKED);
1560                 if (ret)
1561                         break;
1562
1563                 ins.objectid = start;
1564                 ins.offset = end + 1 - start;
1565                 err = btrfs_insert_item(trans, extent_root, &ins,
1566                                         &extent_item, sizeof(extent_item));
1567                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1568                                   GFP_NOFS);
1569
1570                 eb = btrfs_find_create_tree_block(extent_root, ins.objectid,
1571                                            ins.offset);
1572
1573                 if (!btrfs_buffer_uptodate(eb, trans->transid))
1574                         btrfs_read_buffer(eb, trans->transid);
1575
1576                 btrfs_tree_lock(eb);
1577                 level = btrfs_header_level(eb);
1578                 if (level == 0) {
1579                         btrfs_item_key(eb, &first, 0);
1580                 } else {
1581                         btrfs_node_key(eb, &first, 0);
1582                 }
1583                 btrfs_tree_unlock(eb);
1584                 free_extent_buffer(eb);
1585                 /*
1586                  * the first key is just a hint, so the race we've created
1587                  * against reading it is fine
1588                  */
1589                 err = btrfs_insert_extent_backref(trans, extent_root, path,
1590                                           start, extent_root->root_key.objectid,
1591                                           0, level,
1592                                           btrfs_disk_key_objectid(&first));
1593                 BUG_ON(err);
1594                 if (need_resched()) {
1595                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1596                         cond_resched();
1597                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1598                 }
1599         }
1600         btrfs_free_path(path);
1601         return 0;
1602 }
1603
1604 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1605                           int is_data, int pending)
1606 {
1607         int err = 0;
1608
1609         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1610         if (!pending) {
1611                 struct extent_buffer *buf;
1612
1613                 if (is_data)
1614                         goto pinit;
1615
1616                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1617                 if (buf) {
1618                         /* we can reuse a block if it hasn't been written
1619                          * and it is from this transaction.  We can't
1620                          * reuse anything from the tree log root because
1621                          * it has tiny sub-transactions.
1622                          */
1623                         if (btrfs_buffer_uptodate(buf, 0) &&
1624                             btrfs_try_tree_lock(buf)) {
1625                                 u64 transid =
1626                                     root->fs_info->running_transaction->transid;
1627                                 u64 header_transid =
1628                                         btrfs_header_generation(buf);
1629                                 if (btrfs_header_owner(buf) !=
1630                                     BTRFS_TREE_LOG_OBJECTID &&
1631                                     header_transid == transid &&
1632                                     !btrfs_header_flag(buf,
1633                                                BTRFS_HEADER_FLAG_WRITTEN)) {
1634                                         clean_tree_block(NULL, root, buf);
1635                                         btrfs_tree_unlock(buf);
1636                                         free_extent_buffer(buf);
1637                                         return 1;
1638                                 }
1639                                 btrfs_tree_unlock(buf);
1640                         }
1641                         free_extent_buffer(buf);
1642                 }
1643 pinit:
1644                 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1645         } else {
1646                 set_extent_bits(&root->fs_info->pending_del,
1647                                 bytenr, bytenr + num_bytes - 1,
1648                                 EXTENT_LOCKED, GFP_NOFS);
1649         }
1650         BUG_ON(err < 0);
1651         return 0;
1652 }
1653
1654 /*
1655  * remove an extent from the root, returns 0 on success
1656  */
1657 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1658                          *root, u64 bytenr, u64 num_bytes,
1659                          u64 root_objectid, u64 ref_generation,
1660                          u64 owner_objectid, u64 owner_offset, int pin,
1661                          int mark_free)
1662 {
1663         struct btrfs_path *path;
1664         struct btrfs_key key;
1665         struct btrfs_fs_info *info = root->fs_info;
1666         struct btrfs_root *extent_root = info->extent_root;
1667         struct extent_buffer *leaf;
1668         int ret;
1669         int extent_slot = 0;
1670         int found_extent = 0;
1671         int num_to_del = 1;
1672         struct btrfs_extent_item *ei;
1673         u32 refs;
1674
1675         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1676         key.objectid = bytenr;
1677         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1678         key.offset = num_bytes;
1679         path = btrfs_alloc_path();
1680         if (!path)
1681                 return -ENOMEM;
1682
1683         path->reada = 1;
1684         ret = lookup_extent_backref(trans, extent_root, path,
1685                                     bytenr, root_objectid,
1686                                     ref_generation,
1687                                     owner_objectid, owner_offset, 1);
1688         if (ret == 0) {
1689                 struct btrfs_key found_key;
1690                 extent_slot = path->slots[0];
1691                 while(extent_slot > 0) {
1692                         extent_slot--;
1693                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1694                                               extent_slot);
1695                         if (found_key.objectid != bytenr)
1696                                 break;
1697                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1698                             found_key.offset == num_bytes) {
1699                                 found_extent = 1;
1700                                 break;
1701                         }
1702                         if (path->slots[0] - extent_slot > 5)
1703                                 break;
1704                 }
1705                 if (!found_extent)
1706                         ret = btrfs_del_item(trans, extent_root, path);
1707         } else {
1708                 btrfs_print_leaf(extent_root, path->nodes[0]);
1709                 WARN_ON(1);
1710                 printk("Unable to find ref byte nr %Lu root %Lu "
1711                        " gen %Lu owner %Lu offset %Lu\n", bytenr,
1712                        root_objectid, ref_generation, owner_objectid,
1713                        owner_offset);
1714         }
1715         if (!found_extent) {
1716                 btrfs_release_path(extent_root, path);
1717                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1718                 if (ret < 0)
1719                         return ret;
1720                 BUG_ON(ret);
1721                 extent_slot = path->slots[0];
1722         }
1723
1724         leaf = path->nodes[0];
1725         ei = btrfs_item_ptr(leaf, extent_slot,
1726                             struct btrfs_extent_item);
1727         refs = btrfs_extent_refs(leaf, ei);
1728         BUG_ON(refs == 0);
1729         refs -= 1;
1730         btrfs_set_extent_refs(leaf, ei, refs);
1731
1732         btrfs_mark_buffer_dirty(leaf);
1733
1734         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1735                 /* if the back ref and the extent are next to each other
1736                  * they get deleted below in one shot
1737                  */
1738                 path->slots[0] = extent_slot;
1739                 num_to_del = 2;
1740         } else if (found_extent) {
1741                 /* otherwise delete the extent back ref */
1742                 ret = btrfs_del_item(trans, extent_root, path);
1743                 BUG_ON(ret);
1744                 /* if refs are 0, we need to setup the path for deletion */
1745                 if (refs == 0) {
1746                         btrfs_release_path(extent_root, path);
1747                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1748                                                 -1, 1);
1749                         if (ret < 0)
1750                                 return ret;
1751                         BUG_ON(ret);
1752                 }
1753         }
1754
1755         if (refs == 0) {
1756                 u64 super_used;
1757                 u64 root_used;
1758 #ifdef BIO_RW_DISCARD
1759                 u64 map_length = num_bytes;
1760                 struct btrfs_multi_bio *multi = NULL;
1761 #endif
1762
1763                 if (pin) {
1764                         ret = pin_down_bytes(root, bytenr, num_bytes,
1765                              owner_objectid >= BTRFS_FIRST_FREE_OBJECTID, 0);
1766                         if (ret > 0)
1767                                 mark_free = 1;
1768                         BUG_ON(ret < 0);
1769                 }
1770
1771                 /* block accounting for super block */
1772                 spin_lock_irq(&info->delalloc_lock);
1773                 super_used = btrfs_super_bytes_used(&info->super_copy);
1774                 btrfs_set_super_bytes_used(&info->super_copy,
1775                                            super_used - num_bytes);
1776                 spin_unlock_irq(&info->delalloc_lock);
1777
1778                 /* block accounting for root item */
1779                 root_used = btrfs_root_used(&root->root_item);
1780                 btrfs_set_root_used(&root->root_item,
1781                                            root_used - num_bytes);
1782                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1783                                       num_to_del);
1784                 if (ret) {
1785                         return ret;
1786                 }
1787                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1788                                          mark_free);
1789                 BUG_ON(ret);
1790
1791 #ifdef BIO_RW_DISCARD
1792                 /* Tell the block device(s) that the sectors can be discarded */
1793                 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1794                                       bytenr, &map_length, &multi, 0);
1795                 if (!ret) {
1796                         struct btrfs_bio_stripe *stripe = multi->stripes;
1797                         int i;
1798
1799                         if (map_length > num_bytes)
1800                                 map_length = num_bytes;
1801
1802                         for (i = 0; i < multi->num_stripes; i++, stripe++) {
1803                                 blkdev_issue_discard(stripe->dev->bdev,
1804                                                      stripe->physical >> 9,
1805                                                      map_length >> 9);
1806                         }
1807                         kfree(multi);
1808                 }
1809 #endif
1810         }
1811         btrfs_free_path(path);
1812         finish_current_insert(trans, extent_root);
1813         return ret;
1814 }
1815
1816 /*
1817  * find all the blocks marked as pending in the radix tree and remove
1818  * them from the extent map
1819  */
1820 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1821                                btrfs_root *extent_root)
1822 {
1823         int ret;
1824         int err = 0;
1825         u64 start;
1826         u64 end;
1827         struct extent_io_tree *pending_del;
1828         struct extent_io_tree *pinned_extents;
1829
1830         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1831         pending_del = &extent_root->fs_info->pending_del;
1832         pinned_extents = &extent_root->fs_info->pinned_extents;
1833
1834         while(1) {
1835                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1836                                             EXTENT_LOCKED);
1837                 if (ret)
1838                         break;
1839                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1840                                   GFP_NOFS);
1841                 if (!test_range_bit(&extent_root->fs_info->extent_ins,
1842                                     start, end, EXTENT_LOCKED, 0)) {
1843                         btrfs_update_pinned_extents(extent_root, start,
1844                                               end + 1 - start, 1);
1845                         ret = __free_extent(trans, extent_root,
1846                                              start, end + 1 - start,
1847                                              extent_root->root_key.objectid,
1848                                              0, 0, 0, 0, 0);
1849                 } else {
1850                         clear_extent_bits(&extent_root->fs_info->extent_ins,
1851                                           start, end, EXTENT_LOCKED, GFP_NOFS);
1852                 }
1853                 if (ret)
1854                         err = ret;
1855
1856                 if (need_resched()) {
1857                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
1858                         cond_resched();
1859                         mutex_lock(&extent_root->fs_info->alloc_mutex);
1860                 }
1861         }
1862         return err;
1863 }
1864
1865 /*
1866  * remove an extent from the root, returns 0 on success
1867  */
1868 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1869                                struct btrfs_root *root, u64 bytenr,
1870                                u64 num_bytes, u64 root_objectid,
1871                                u64 ref_generation, u64 owner_objectid,
1872                                u64 owner_offset, int pin)
1873 {
1874         struct btrfs_root *extent_root = root->fs_info->extent_root;
1875         int pending_ret;
1876         int ret;
1877
1878         WARN_ON(num_bytes < root->sectorsize);
1879         if (!root->ref_cows)
1880                 ref_generation = 0;
1881
1882         if (root == extent_root) {
1883                 pin_down_bytes(root, bytenr, num_bytes, 0, 1);
1884                 return 0;
1885         }
1886         /* if metadata always pin */
1887         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
1888                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1889                         struct btrfs_block_group_cache *cache;
1890
1891                         /* btrfs_free_reserved_extent */
1892                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
1893                         BUG_ON(!cache);
1894                         btrfs_add_free_space(cache, bytenr, num_bytes);
1895                         return 0;
1896                 }
1897                 pin = 1;
1898         }
1899
1900         /* if data pin when any transaction has committed this */
1901         if (ref_generation != trans->transid)
1902                 pin = 1;
1903
1904         ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1905                             ref_generation, owner_objectid, owner_offset,
1906                             pin, pin == 0);
1907
1908         finish_current_insert(trans, root->fs_info->extent_root);
1909         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1910         return ret ? ret : pending_ret;
1911 }
1912
1913 int btrfs_free_extent(struct btrfs_trans_handle *trans,
1914                       struct btrfs_root *root, u64 bytenr,
1915                       u64 num_bytes, u64 root_objectid,
1916                       u64 ref_generation, u64 owner_objectid,
1917                       u64 owner_offset, int pin)
1918 {
1919         int ret;
1920
1921         maybe_lock_mutex(root);
1922         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1923                                   root_objectid, ref_generation,
1924                                   owner_objectid, owner_offset, pin);
1925         maybe_unlock_mutex(root);
1926         return ret;
1927 }
1928
1929 static u64 stripe_align(struct btrfs_root *root, u64 val)
1930 {
1931         u64 mask = ((u64)root->stripesize - 1);
1932         u64 ret = (val + mask) & ~mask;
1933         return ret;
1934 }
1935
1936 /*
1937  * walks the btree of allocated extents and find a hole of a given size.
1938  * The key ins is changed to record the hole:
1939  * ins->objectid == block start
1940  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1941  * ins->offset == number of blocks
1942  * Any available blocks before search_start are skipped.
1943  */
1944 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1945                                      struct btrfs_root *orig_root,
1946                                      u64 num_bytes, u64 empty_size,
1947                                      u64 search_start, u64 search_end,
1948                                      u64 hint_byte, struct btrfs_key *ins,
1949                                      u64 exclude_start, u64 exclude_nr,
1950                                      int data)
1951 {
1952         int ret;
1953         u64 orig_search_start;
1954         struct btrfs_root * root = orig_root->fs_info->extent_root;
1955         struct btrfs_fs_info *info = root->fs_info;
1956         u64 total_needed = num_bytes;
1957         u64 *last_ptr = NULL;
1958         struct btrfs_block_group_cache *block_group;
1959         int chunk_alloc_done = 0;
1960         int empty_cluster = 2 * 1024 * 1024;
1961         int allowed_chunk_alloc = 0;
1962
1963         WARN_ON(num_bytes < root->sectorsize);
1964         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1965
1966         if (orig_root->ref_cows || empty_size)
1967                 allowed_chunk_alloc = 1;
1968
1969         if (data & BTRFS_BLOCK_GROUP_METADATA) {
1970                 last_ptr = &root->fs_info->last_alloc;
1971                 empty_cluster = 256 * 1024;
1972         }
1973
1974         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
1975                 last_ptr = &root->fs_info->last_data_alloc;
1976
1977         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1978                 last_ptr = &root->fs_info->last_log_alloc;
1979                 if (!last_ptr == 0 && root->fs_info->last_alloc) {
1980                         *last_ptr = root->fs_info->last_alloc + empty_cluster;
1981                 }
1982         }
1983
1984         if (last_ptr) {
1985                 if (*last_ptr)
1986                         hint_byte = *last_ptr;
1987                 else
1988                         empty_size += empty_cluster;
1989         }
1990
1991         search_start = max(search_start, first_logical_byte(root, 0));
1992         orig_search_start = search_start;
1993
1994         if (search_end == (u64)-1)
1995                 search_end = btrfs_super_total_bytes(&info->super_copy);
1996
1997         search_start = max(search_start, hint_byte);
1998         total_needed += empty_size;
1999
2000 new_group:
2001         block_group = btrfs_lookup_block_group(info, search_start);
2002
2003         /*
2004          * Ok this looks a little tricky, buts its really simple.  First if we
2005          * didn't find a block group obviously we want to start over.
2006          * Secondly, if the block group we found does not match the type we
2007          * need, and we have a last_ptr and its not 0, chances are the last
2008          * allocation we made was at the end of the block group, so lets go
2009          * ahead and skip the looking through the rest of the block groups and
2010          * start at the beginning.  This helps with metadata allocations,
2011          * since you are likely to have a bunch of data block groups to search
2012          * through first before you realize that you need to start over, so go
2013          * ahead and start over and save the time.
2014          */
2015         if (!block_group || (!block_group_bits(block_group, data) &&
2016                              last_ptr && *last_ptr)) {
2017                 if (search_start != orig_search_start) {
2018                         if (last_ptr && *last_ptr)
2019                                 *last_ptr = 0;
2020                         search_start = orig_search_start;
2021                         goto new_group;
2022                 } else if (!chunk_alloc_done && allowed_chunk_alloc) {
2023                         ret = do_chunk_alloc(trans, root,
2024                                              num_bytes + 2 * 1024 * 1024,
2025                                              data, 1);
2026                         if (ret < 0) {
2027                                 struct btrfs_space_info *info;
2028
2029                                 info = __find_space_info(root->fs_info, data);
2030                                 goto error;
2031                         }
2032                         BUG_ON(ret);
2033                         chunk_alloc_done = 1;
2034                         search_start = orig_search_start;
2035                         goto new_group;
2036                 } else {
2037                         ret = -ENOSPC;
2038                         goto error;
2039                 }
2040         }
2041
2042         /*
2043          * this is going to seach through all of the existing block groups it
2044          * can find, so if we don't find something we need to see if we can
2045          * allocate what we need.
2046          */
2047         ret = find_free_space(root, &block_group, &search_start,
2048                               total_needed, data);
2049         if (ret == -ENOSPC) {
2050                 /*
2051                  * instead of allocating, start at the original search start
2052                  * and see if there is something to be found, if not then we
2053                  * allocate
2054                  */
2055                 if (search_start != orig_search_start) {
2056                         if (last_ptr && *last_ptr) {
2057                                 *last_ptr = 0;
2058                                 total_needed += empty_cluster;
2059                         }
2060                         search_start = orig_search_start;
2061                         goto new_group;
2062                 }
2063
2064                 /*
2065                  * we've already allocated, we're pretty screwed
2066                  */
2067                 if (chunk_alloc_done) {
2068                         goto error;
2069                 } else if (!allowed_chunk_alloc && block_group &&
2070                            block_group_bits(block_group, data)) {
2071                         block_group->space_info->force_alloc = 1;
2072                         goto error;
2073                 } else if (!allowed_chunk_alloc) {
2074                         goto error;
2075                 }
2076
2077                 ret = do_chunk_alloc(trans, root, num_bytes + 2 * 1024 * 1024,
2078                                      data, 1);
2079                 if (ret < 0)
2080                         goto error;
2081
2082                 BUG_ON(ret);
2083                 chunk_alloc_done = 1;
2084                 if (block_group)
2085                         search_start = block_group->key.objectid +
2086                                 block_group->key.offset;
2087                 else
2088                         search_start = orig_search_start;
2089                 goto new_group;
2090         }
2091
2092         if (ret)
2093                 goto error;
2094
2095         search_start = stripe_align(root, search_start);
2096         ins->objectid = search_start;
2097         ins->offset = num_bytes;
2098
2099         if (ins->objectid + num_bytes >= search_end) {
2100                 search_start = orig_search_start;
2101                 if (chunk_alloc_done) {
2102                         ret = -ENOSPC;
2103                         goto error;
2104                 }
2105                 goto new_group;
2106         }
2107
2108         if (ins->objectid + num_bytes >
2109             block_group->key.objectid + block_group->key.offset) {
2110                 if (search_start == orig_search_start && chunk_alloc_done) {
2111                         ret = -ENOSPC;
2112                         goto error;
2113                 }
2114                 search_start = block_group->key.objectid +
2115                         block_group->key.offset;
2116                 goto new_group;
2117         }
2118
2119         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2120             ins->objectid < exclude_start + exclude_nr)) {
2121                 search_start = exclude_start + exclude_nr;
2122                 goto new_group;
2123         }
2124
2125         if (!(data & BTRFS_BLOCK_GROUP_DATA))
2126                 trans->block_group = block_group;
2127
2128         ins->offset = num_bytes;
2129         if (last_ptr) {
2130                 *last_ptr = ins->objectid + ins->offset;
2131                 if (*last_ptr ==
2132                     btrfs_super_total_bytes(&root->fs_info->super_copy))
2133                         *last_ptr = 0;
2134         }
2135
2136         ret = 0;
2137 error:
2138         return ret;
2139 }
2140
2141 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2142 {
2143         struct btrfs_block_group_cache *cache;
2144         struct list_head *l;
2145
2146         printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
2147                info->total_bytes - info->bytes_used - info->bytes_pinned,
2148                (info->full) ? "" : "not ");
2149
2150         spin_lock(&info->lock);
2151         list_for_each(l, &info->block_groups) {
2152                 cache = list_entry(l, struct btrfs_block_group_cache, list);
2153                 spin_lock(&cache->lock);
2154                 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
2155                        "%Lu pinned\n",
2156                        cache->key.objectid, cache->key.offset,
2157                        btrfs_block_group_used(&cache->item), cache->pinned);
2158                 btrfs_dump_free_space(cache, bytes);
2159                 spin_unlock(&cache->lock);
2160         }
2161         spin_unlock(&info->lock);
2162 }
2163 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2164                                   struct btrfs_root *root,
2165                                   u64 num_bytes, u64 min_alloc_size,
2166                                   u64 empty_size, u64 hint_byte,
2167                                   u64 search_end, struct btrfs_key *ins,
2168                                   u64 data)
2169 {
2170         int ret;
2171         u64 search_start = 0;
2172         u64 alloc_profile;
2173         struct btrfs_fs_info *info = root->fs_info;
2174         struct btrfs_block_group_cache *cache;
2175
2176         if (data) {
2177                 alloc_profile = info->avail_data_alloc_bits &
2178                                 info->data_alloc_profile;
2179                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2180         } else if (root == root->fs_info->chunk_root) {
2181                 alloc_profile = info->avail_system_alloc_bits &
2182                                 info->system_alloc_profile;
2183                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2184         } else {
2185                 alloc_profile = info->avail_metadata_alloc_bits &
2186                                 info->metadata_alloc_profile;
2187                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2188         }
2189 again:
2190         data = reduce_alloc_profile(root, data);
2191         /*
2192          * the only place that sets empty_size is btrfs_realloc_node, which
2193          * is not called recursively on allocations
2194          */
2195         if (empty_size || root->ref_cows) {
2196                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2197                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2198                                      2 * 1024 * 1024,
2199                                      BTRFS_BLOCK_GROUP_METADATA |
2200                                      (info->metadata_alloc_profile &
2201                                       info->avail_metadata_alloc_bits), 0);
2202                 }
2203                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2204                                      num_bytes + 2 * 1024 * 1024, data, 0);
2205         }
2206
2207         WARN_ON(num_bytes < root->sectorsize);
2208         ret = find_free_extent(trans, root, num_bytes, empty_size,
2209                                search_start, search_end, hint_byte, ins,
2210                                trans->alloc_exclude_start,
2211                                trans->alloc_exclude_nr, data);
2212
2213         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2214                 num_bytes = num_bytes >> 1;
2215                 num_bytes = num_bytes & ~(root->sectorsize - 1);
2216                 num_bytes = max(num_bytes, min_alloc_size);
2217                 do_chunk_alloc(trans, root->fs_info->extent_root,
2218                                num_bytes, data, 1);
2219                 goto again;
2220         }
2221         if (ret) {
2222                 struct btrfs_space_info *sinfo;
2223
2224                 sinfo = __find_space_info(root->fs_info, data);
2225                 printk("allocation failed flags %Lu, wanted %Lu\n",
2226                        data, num_bytes);
2227                 dump_space_info(sinfo, num_bytes);
2228                 BUG();
2229         }
2230         cache = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2231         if (!cache) {
2232                 printk(KERN_ERR "Unable to find block group for %Lu\n", ins->objectid);
2233                 return -ENOSPC;
2234         }
2235
2236         ret = btrfs_remove_free_space(cache, ins->objectid, ins->offset);
2237
2238         return ret;
2239 }
2240
2241 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2242 {
2243         struct btrfs_block_group_cache *cache;
2244
2245         maybe_lock_mutex(root);
2246         cache = btrfs_lookup_block_group(root->fs_info, start);
2247         if (!cache) {
2248                 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
2249                 maybe_unlock_mutex(root);
2250                 return -ENOSPC;
2251         }
2252         btrfs_add_free_space(cache, start, len);
2253         maybe_unlock_mutex(root);
2254         return 0;
2255 }
2256
2257 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2258                                   struct btrfs_root *root,
2259                                   u64 num_bytes, u64 min_alloc_size,
2260                                   u64 empty_size, u64 hint_byte,
2261                                   u64 search_end, struct btrfs_key *ins,
2262                                   u64 data)
2263 {
2264         int ret;
2265         maybe_lock_mutex(root);
2266         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2267                                      empty_size, hint_byte, search_end, ins,
2268                                      data);
2269         maybe_unlock_mutex(root);
2270         return ret;
2271 }
2272
2273 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2274                                          struct btrfs_root *root,
2275                                          u64 root_objectid, u64 ref_generation,
2276                                          u64 owner, u64 owner_offset,
2277                                          struct btrfs_key *ins)
2278 {
2279         int ret;
2280         int pending_ret;
2281         u64 super_used;
2282         u64 root_used;
2283         u64 num_bytes = ins->offset;
2284         u32 sizes[2];
2285         struct btrfs_fs_info *info = root->fs_info;
2286         struct btrfs_root *extent_root = info->extent_root;
2287         struct btrfs_extent_item *extent_item;
2288         struct btrfs_extent_ref *ref;
2289         struct btrfs_path *path;
2290         struct btrfs_key keys[2];
2291
2292         /* block accounting for super block */
2293         spin_lock_irq(&info->delalloc_lock);
2294         super_used = btrfs_super_bytes_used(&info->super_copy);
2295         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2296         spin_unlock_irq(&info->delalloc_lock);
2297
2298         /* block accounting for root item */
2299         root_used = btrfs_root_used(&root->root_item);
2300         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2301
2302         if (root == extent_root) {
2303                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2304                                 ins->objectid + ins->offset - 1,
2305                                 EXTENT_LOCKED, GFP_NOFS);
2306                 goto update_block;
2307         }
2308
2309         memcpy(&keys[0], ins, sizeof(*ins));
2310         keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
2311                                          owner, owner_offset);
2312         keys[1].objectid = ins->objectid;
2313         keys[1].type = BTRFS_EXTENT_REF_KEY;
2314         sizes[0] = sizeof(*extent_item);
2315         sizes[1] = sizeof(*ref);
2316
2317         path = btrfs_alloc_path();
2318         BUG_ON(!path);
2319
2320         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2321                                        sizes, 2);
2322         BUG_ON(ret);
2323
2324         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2325                                      struct btrfs_extent_item);
2326         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2327         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2328                              struct btrfs_extent_ref);
2329
2330         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2331         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2332         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2333         btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2334
2335         btrfs_mark_buffer_dirty(path->nodes[0]);
2336
2337         trans->alloc_exclude_start = 0;
2338         trans->alloc_exclude_nr = 0;
2339         btrfs_free_path(path);
2340         finish_current_insert(trans, extent_root);
2341         pending_ret = del_pending_extents(trans, extent_root);
2342
2343         if (ret)
2344                 goto out;
2345         if (pending_ret) {
2346                 ret = pending_ret;
2347                 goto out;
2348         }
2349
2350 update_block:
2351         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2352         if (ret) {
2353                 printk("update block group failed for %Lu %Lu\n",
2354                        ins->objectid, ins->offset);
2355                 BUG();
2356         }
2357 out:
2358         return ret;
2359 }
2360
2361 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2362                                 struct btrfs_root *root,
2363                                 u64 root_objectid, u64 ref_generation,
2364                                 u64 owner, u64 owner_offset,
2365                                 struct btrfs_key *ins)
2366 {
2367         int ret;
2368         maybe_lock_mutex(root);
2369         ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2370                                             ref_generation, owner,
2371                                             owner_offset, ins);
2372         maybe_unlock_mutex(root);
2373         return ret;
2374 }
2375
2376 /*
2377  * this is used by the tree logging recovery code.  It records that
2378  * an extent has been allocated and makes sure to clear the free
2379  * space cache bits as well
2380  */
2381 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2382                                 struct btrfs_root *root,
2383                                 u64 root_objectid, u64 ref_generation,
2384                                 u64 owner, u64 owner_offset,
2385                                 struct btrfs_key *ins)
2386 {
2387         int ret;
2388         struct btrfs_block_group_cache *block_group;
2389
2390         maybe_lock_mutex(root);
2391         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2392         cache_block_group(root, block_group);
2393
2394         ret = btrfs_remove_free_space(block_group, ins->objectid, ins->offset);
2395         BUG_ON(ret);
2396
2397         ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2398                                             ref_generation, owner,
2399                                             owner_offset, ins);
2400         maybe_unlock_mutex(root);
2401         return ret;
2402 }
2403
2404 /*
2405  * finds a free extent and does all the dirty work required for allocation
2406  * returns the key for the extent through ins, and a tree buffer for
2407  * the first block of the extent through buf.
2408  *
2409  * returns 0 if everything worked, non-zero otherwise.
2410  */
2411 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2412                        struct btrfs_root *root,
2413                        u64 num_bytes, u64 min_alloc_size,
2414                        u64 root_objectid, u64 ref_generation,
2415                        u64 owner, u64 owner_offset,
2416                        u64 empty_size, u64 hint_byte,
2417                        u64 search_end, struct btrfs_key *ins, u64 data)
2418 {
2419         int ret;
2420
2421         maybe_lock_mutex(root);
2422
2423         ret = __btrfs_reserve_extent(trans, root, num_bytes,
2424                                      min_alloc_size, empty_size, hint_byte,
2425                                      search_end, ins, data);
2426         BUG_ON(ret);
2427         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2428                 ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2429                                                     ref_generation, owner,
2430                                                     owner_offset, ins);
2431                 BUG_ON(ret);
2432
2433         }
2434         maybe_unlock_mutex(root);
2435         return ret;
2436 }
2437
2438 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2439                                             struct btrfs_root *root,
2440                                             u64 bytenr, u32 blocksize)
2441 {
2442         struct extent_buffer *buf;
2443
2444         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2445         if (!buf)
2446                 return ERR_PTR(-ENOMEM);
2447         btrfs_set_header_generation(buf, trans->transid);
2448         btrfs_tree_lock(buf);
2449         clean_tree_block(trans, root, buf);
2450         btrfs_set_buffer_uptodate(buf);
2451         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2452                 set_extent_dirty(&root->dirty_log_pages, buf->start,
2453                          buf->start + buf->len - 1, GFP_NOFS);
2454         } else {
2455                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2456                          buf->start + buf->len - 1, GFP_NOFS);
2457         }
2458         trans->blocks_used++;
2459         return buf;
2460 }
2461
2462 /*
2463  * helper function to allocate a block for a given tree
2464  * returns the tree buffer or NULL.
2465  */
2466 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2467                                              struct btrfs_root *root,
2468                                              u32 blocksize,
2469                                              u64 root_objectid,
2470                                              u64 ref_generation,
2471                                              u64 first_objectid,
2472                                              int level,
2473                                              u64 hint,
2474                                              u64 empty_size)
2475 {
2476         struct btrfs_key ins;
2477         int ret;
2478         struct extent_buffer *buf;
2479
2480         ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2481                                  root_objectid, ref_generation,
2482                                  level, first_objectid, empty_size, hint,
2483                                  (u64)-1, &ins, 0);
2484         if (ret) {
2485                 BUG_ON(ret > 0);
2486                 return ERR_PTR(ret);
2487         }
2488
2489         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2490         return buf;
2491 }
2492
2493 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2494                         struct btrfs_root *root, struct extent_buffer *leaf)
2495 {
2496         u64 leaf_owner;
2497         u64 leaf_generation;
2498         struct btrfs_key key;
2499         struct btrfs_file_extent_item *fi;
2500         int i;
2501         int nritems;
2502         int ret;
2503
2504         BUG_ON(!btrfs_is_leaf(leaf));
2505         nritems = btrfs_header_nritems(leaf);
2506         leaf_owner = btrfs_header_owner(leaf);
2507         leaf_generation = btrfs_header_generation(leaf);
2508
2509         for (i = 0; i < nritems; i++) {
2510                 u64 disk_bytenr;
2511                 cond_resched();
2512
2513                 btrfs_item_key_to_cpu(leaf, &key, i);
2514                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2515                         continue;
2516                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2517                 if (btrfs_file_extent_type(leaf, fi) ==
2518                     BTRFS_FILE_EXTENT_INLINE)
2519                         continue;
2520                 /*
2521                  * FIXME make sure to insert a trans record that
2522                  * repeats the snapshot del on crash
2523                  */
2524                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2525                 if (disk_bytenr == 0)
2526                         continue;
2527
2528                 mutex_lock(&root->fs_info->alloc_mutex);
2529                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2530                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2531                                 leaf_owner, leaf_generation,
2532                                 key.objectid, key.offset, 0);
2533                 mutex_unlock(&root->fs_info->alloc_mutex);
2534
2535                 atomic_inc(&root->fs_info->throttle_gen);
2536                 wake_up(&root->fs_info->transaction_throttle);
2537                 cond_resched();
2538
2539                 BUG_ON(ret);
2540         }
2541         return 0;
2542 }
2543
2544 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2545                                         struct btrfs_root *root,
2546                                         struct btrfs_leaf_ref *ref)
2547 {
2548         int i;
2549         int ret;
2550         struct btrfs_extent_info *info = ref->extents;
2551
2552         for (i = 0; i < ref->nritems; i++) {
2553                 mutex_lock(&root->fs_info->alloc_mutex);
2554                 ret = __btrfs_free_extent(trans, root,
2555                                         info->bytenr, info->num_bytes,
2556                                         ref->owner, ref->generation,
2557                                         info->objectid, info->offset, 0);
2558                 mutex_unlock(&root->fs_info->alloc_mutex);
2559
2560                 atomic_inc(&root->fs_info->throttle_gen);
2561                 wake_up(&root->fs_info->transaction_throttle);
2562                 cond_resched();
2563
2564                 BUG_ON(ret);
2565                 info++;
2566         }
2567
2568         return 0;
2569 }
2570
2571 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2572                               u32 *refs)
2573 {
2574         int ret;
2575
2576         ret = lookup_extent_ref(NULL, root, start, len, refs);
2577         BUG_ON(ret);
2578
2579 #if 0 // some debugging code in case we see problems here
2580         /* if the refs count is one, it won't get increased again.  But
2581          * if the ref count is > 1, someone may be decreasing it at
2582          * the same time we are.
2583          */
2584         if (*refs != 1) {
2585                 struct extent_buffer *eb = NULL;
2586                 eb = btrfs_find_create_tree_block(root, start, len);
2587                 if (eb)
2588                         btrfs_tree_lock(eb);
2589
2590                 mutex_lock(&root->fs_info->alloc_mutex);
2591                 ret = lookup_extent_ref(NULL, root, start, len, refs);
2592                 BUG_ON(ret);
2593                 mutex_unlock(&root->fs_info->alloc_mutex);
2594
2595                 if (eb) {
2596                         btrfs_tree_unlock(eb);
2597                         free_extent_buffer(eb);
2598                 }
2599                 if (*refs == 1) {
2600                         printk("block %llu went down to one during drop_snap\n",
2601                                (unsigned long long)start);
2602                 }
2603
2604         }
2605 #endif
2606
2607         cond_resched();
2608         return ret;
2609 }
2610
2611 /*
2612  * helper function for drop_snapshot, this walks down the tree dropping ref
2613  * counts as it goes.
2614  */
2615 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2616                                    struct btrfs_root *root,
2617                                    struct btrfs_path *path, int *level)
2618 {
2619         u64 root_owner;
2620         u64 root_gen;
2621         u64 bytenr;
2622         u64 ptr_gen;
2623         struct extent_buffer *next;
2624         struct extent_buffer *cur;
2625         struct extent_buffer *parent;
2626         struct btrfs_leaf_ref *ref;
2627         u32 blocksize;
2628         int ret;
2629         u32 refs;
2630
2631         WARN_ON(*level < 0);
2632         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2633         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2634                                 path->nodes[*level]->len, &refs);
2635         BUG_ON(ret);
2636         if (refs > 1)
2637                 goto out;
2638
2639         /*
2640          * walk down to the last node level and free all the leaves
2641          */
2642         while(*level >= 0) {
2643                 WARN_ON(*level < 0);
2644                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2645                 cur = path->nodes[*level];
2646
2647                 if (btrfs_header_level(cur) != *level)
2648                         WARN_ON(1);
2649
2650                 if (path->slots[*level] >=
2651                     btrfs_header_nritems(cur))
2652                         break;
2653                 if (*level == 0) {
2654                         ret = btrfs_drop_leaf_ref(trans, root, cur);
2655                         BUG_ON(ret);
2656                         break;
2657                 }
2658                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2659                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2660                 blocksize = btrfs_level_size(root, *level - 1);
2661
2662                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2663                 BUG_ON(ret);
2664                 if (refs != 1) {
2665                         parent = path->nodes[*level];
2666                         root_owner = btrfs_header_owner(parent);
2667                         root_gen = btrfs_header_generation(parent);
2668                         path->slots[*level]++;
2669
2670                         mutex_lock(&root->fs_info->alloc_mutex);
2671                         ret = __btrfs_free_extent(trans, root, bytenr,
2672                                                 blocksize, root_owner,
2673                                                 root_gen, 0, 0, 1);
2674                         BUG_ON(ret);
2675                         mutex_unlock(&root->fs_info->alloc_mutex);
2676
2677                         atomic_inc(&root->fs_info->throttle_gen);
2678                         wake_up(&root->fs_info->transaction_throttle);
2679                         cond_resched();
2680
2681                         continue;
2682                 }
2683                 /*
2684                  * at this point, we have a single ref, and since the
2685                  * only place referencing this extent is a dead root
2686                  * the reference count should never go higher.
2687                  * So, we don't need to check it again
2688                  */
2689                 if (*level == 1) {
2690                         struct btrfs_key key;
2691                         btrfs_node_key_to_cpu(cur, &key, path->slots[*level]);
2692                         ref = btrfs_lookup_leaf_ref(root, bytenr);
2693                         if (ref) {
2694                                 ret = cache_drop_leaf_ref(trans, root, ref);
2695                                 BUG_ON(ret);
2696                                 btrfs_remove_leaf_ref(root, ref);
2697                                 btrfs_free_leaf_ref(root, ref);
2698                                 *level = 0;
2699                                 break;
2700                         }
2701                         if (printk_ratelimit())
2702                                 printk("leaf ref miss for bytenr %llu\n",
2703                                        (unsigned long long)bytenr);
2704                 }
2705                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2706                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2707                         free_extent_buffer(next);
2708
2709                         next = read_tree_block(root, bytenr, blocksize,
2710                                                ptr_gen);
2711                         cond_resched();
2712 #if 0
2713                         /*
2714                          * this is a debugging check and can go away
2715                          * the ref should never go all the way down to 1
2716                          * at this point
2717                          */
2718                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2719                                                 &refs);
2720                         BUG_ON(ret);
2721                         WARN_ON(refs != 1);
2722 #endif
2723                 }
2724                 WARN_ON(*level <= 0);
2725                 if (path->nodes[*level-1])
2726                         free_extent_buffer(path->nodes[*level-1]);
2727                 path->nodes[*level-1] = next;
2728                 *level = btrfs_header_level(next);
2729                 path->slots[*level] = 0;
2730                 cond_resched();
2731         }
2732 out:
2733         WARN_ON(*level < 0);
2734         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2735
2736         if (path->nodes[*level] == root->node) {
2737                 parent = path->nodes[*level];
2738                 bytenr = path->nodes[*level]->start;
2739         } else {
2740                 parent = path->nodes[*level + 1];
2741                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
2742         }
2743
2744         blocksize = btrfs_level_size(root, *level);
2745         root_owner = btrfs_header_owner(parent);
2746         root_gen = btrfs_header_generation(parent);
2747
2748         mutex_lock(&root->fs_info->alloc_mutex);
2749         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
2750                                   root_owner, root_gen, 0, 0, 1);
2751         free_extent_buffer(path->nodes[*level]);
2752         path->nodes[*level] = NULL;
2753         *level += 1;
2754         BUG_ON(ret);
2755         mutex_unlock(&root->fs_info->alloc_mutex);
2756
2757         cond_resched();
2758         return 0;
2759 }
2760
2761 /*
2762  * helper for dropping snapshots.  This walks back up the tree in the path
2763  * to find the first node higher up where we haven't yet gone through
2764  * all the slots
2765  */
2766 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2767                                  struct btrfs_root *root,
2768                                  struct btrfs_path *path, int *level)
2769 {
2770         u64 root_owner;
2771         u64 root_gen;
2772         struct btrfs_root_item *root_item = &root->root_item;
2773         int i;
2774         int slot;
2775         int ret;
2776
2777         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2778                 slot = path->slots[i];
2779                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2780                         struct extent_buffer *node;
2781                         struct btrfs_disk_key disk_key;
2782                         node = path->nodes[i];
2783                         path->slots[i]++;
2784                         *level = i;
2785                         WARN_ON(*level == 0);
2786                         btrfs_node_key(node, &disk_key, path->slots[i]);
2787                         memcpy(&root_item->drop_progress,
2788                                &disk_key, sizeof(disk_key));
2789                         root_item->drop_level = i;
2790                         return 0;
2791                 } else {
2792                         if (path->nodes[*level] == root->node) {
2793                                 root_owner = root->root_key.objectid;
2794                                 root_gen =
2795                                    btrfs_header_generation(path->nodes[*level]);
2796                         } else {
2797                                 struct extent_buffer *node;
2798                                 node = path->nodes[*level + 1];
2799                                 root_owner = btrfs_header_owner(node);
2800                                 root_gen = btrfs_header_generation(node);
2801                         }
2802                         ret = btrfs_free_extent(trans, root,
2803                                                 path->nodes[*level]->start,
2804                                                 path->nodes[*level]->len,
2805                                                 root_owner, root_gen, 0, 0, 1);
2806                         BUG_ON(ret);
2807                         free_extent_buffer(path->nodes[*level]);
2808                         path->nodes[*level] = NULL;
2809                         *level = i + 1;
2810                 }
2811         }
2812         return 1;
2813 }
2814
2815 /*
2816  * drop the reference count on the tree rooted at 'snap'.  This traverses
2817  * the tree freeing any blocks that have a ref count of zero after being
2818  * decremented.
2819  */
2820 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2821                         *root)
2822 {
2823         int ret = 0;
2824         int wret;
2825         int level;
2826         struct btrfs_path *path;
2827         int i;
2828         int orig_level;
2829         struct btrfs_root_item *root_item = &root->root_item;
2830
2831         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2832         path = btrfs_alloc_path();
2833         BUG_ON(!path);
2834
2835         level = btrfs_header_level(root->node);
2836         orig_level = level;
2837         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2838                 path->nodes[level] = root->node;
2839                 extent_buffer_get(root->node);
2840                 path->slots[level] = 0;
2841         } else {
2842                 struct btrfs_key key;
2843                 struct btrfs_disk_key found_key;
2844                 struct extent_buffer *node;
2845
2846                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2847                 level = root_item->drop_level;
2848                 path->lowest_level = level;
2849                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2850                 if (wret < 0) {
2851                         ret = wret;
2852                         goto out;
2853                 }
2854                 node = path->nodes[level];
2855                 btrfs_node_key(node, &found_key, path->slots[level]);
2856                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2857                                sizeof(found_key)));
2858                 /*
2859                  * unlock our path, this is safe because only this
2860                  * function is allowed to delete this snapshot
2861                  */
2862                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2863                         if (path->nodes[i] && path->locks[i]) {
2864                                 path->locks[i] = 0;
2865                                 btrfs_tree_unlock(path->nodes[i]);
2866                         }
2867                 }
2868         }
2869         while(1) {
2870                 wret = walk_down_tree(trans, root, path, &level);
2871                 if (wret > 0)
2872                         break;
2873                 if (wret < 0)
2874                         ret = wret;
2875
2876                 wret = walk_up_tree(trans, root, path, &level);
2877                 if (wret > 0)
2878                         break;
2879                 if (wret < 0)
2880                         ret = wret;
2881                 if (trans->transaction->in_commit) {
2882                         ret = -EAGAIN;
2883                         break;
2884                 }
2885                 atomic_inc(&root->fs_info->throttle_gen);
2886                 wake_up(&root->fs_info->transaction_throttle);
2887         }
2888         for (i = 0; i <= orig_level; i++) {
2889                 if (path->nodes[i]) {
2890                         free_extent_buffer(path->nodes[i]);
2891                         path->nodes[i] = NULL;
2892                 }
2893         }
2894 out:
2895         btrfs_free_path(path);
2896         return ret;
2897 }
2898
2899 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2900 {
2901         struct btrfs_block_group_cache *block_group;
2902         struct rb_node *n;
2903
2904         mutex_lock(&info->alloc_mutex);
2905         spin_lock(&info->block_group_cache_lock);
2906         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
2907                 block_group = rb_entry(n, struct btrfs_block_group_cache,
2908                                        cache_node);
2909
2910                 btrfs_remove_free_space_cache(block_group);
2911                 rb_erase(&block_group->cache_node,
2912                          &info->block_group_cache_tree);
2913                 spin_lock(&block_group->space_info->lock);
2914                 list_del(&block_group->list);
2915                 spin_unlock(&block_group->space_info->lock);
2916                 kfree(block_group);
2917         }
2918         spin_unlock(&info->block_group_cache_lock);
2919         mutex_unlock(&info->alloc_mutex);
2920         return 0;
2921 }
2922
2923 static unsigned long calc_ra(unsigned long start, unsigned long last,
2924                              unsigned long nr)
2925 {
2926         return min(last, start + nr - 1);
2927 }
2928
2929 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2930                                          u64 len)
2931 {
2932         u64 page_start;
2933         u64 page_end;
2934         unsigned long last_index;
2935         unsigned long i;
2936         struct page *page;
2937         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2938         struct file_ra_state *ra;
2939         unsigned long total_read = 0;
2940         unsigned long ra_pages;
2941         struct btrfs_ordered_extent *ordered;
2942         struct btrfs_trans_handle *trans;
2943
2944         ra = kzalloc(sizeof(*ra), GFP_NOFS);
2945
2946         mutex_lock(&inode->i_mutex);
2947         i = start >> PAGE_CACHE_SHIFT;
2948         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2949
2950         ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2951
2952         file_ra_state_init(ra, inode->i_mapping);
2953
2954         for (; i <= last_index; i++) {
2955                 if (total_read % ra_pages == 0) {
2956                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2957                                        calc_ra(i, last_index, ra_pages));
2958                 }
2959                 total_read++;
2960 again:
2961                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
2962                         goto truncate_racing;
2963                 page = grab_cache_page(inode->i_mapping, i);
2964                 if (!page) {
2965                         goto out_unlock;
2966                 }
2967                 if (!PageUptodate(page)) {
2968                         btrfs_readpage(NULL, page);
2969                         lock_page(page);
2970                         if (!PageUptodate(page)) {
2971                                 unlock_page(page);
2972                                 page_cache_release(page);
2973                                 goto out_unlock;
2974                         }
2975                 }
2976                 wait_on_page_writeback(page);
2977
2978                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2979                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2980                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2981
2982                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2983                 if (ordered) {
2984                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2985                         unlock_page(page);
2986                         page_cache_release(page);
2987                         btrfs_start_ordered_extent(inode, ordered, 1);
2988                         btrfs_put_ordered_extent(ordered);
2989                         goto again;
2990                 }
2991                 set_page_extent_mapped(page);
2992
2993                 /*
2994                  * make sure page_mkwrite is called for this page if userland
2995                  * wants to change it from mmap
2996                  */
2997                 clear_page_dirty_for_io(page);
2998
2999                 btrfs_set_extent_delalloc(inode, page_start, page_end);
3000                 set_page_dirty(page);
3001
3002                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3003                 unlock_page(page);
3004                 page_cache_release(page);
3005         }
3006
3007 out_unlock:
3008         /* we have to start the IO in order to get the ordered extents
3009          * instantiated.  This allows the relocation to code to wait
3010          * for all the ordered extents to hit the disk.
3011          *
3012          * Otherwise, it would constantly loop over the same extents
3013          * because the old ones don't get deleted  until the IO is
3014          * started
3015          */
3016         btrfs_fdatawrite_range(inode->i_mapping, start, start + len - 1,
3017                                WB_SYNC_NONE);
3018         kfree(ra);
3019         trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
3020         if (trans) {
3021                 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
3022                 mark_inode_dirty(inode);
3023         }
3024         mutex_unlock(&inode->i_mutex);
3025         return 0;
3026
3027 truncate_racing:
3028         vmtruncate(inode, inode->i_size);
3029         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
3030                                            total_read);
3031         goto out_unlock;
3032 }
3033
3034 /*
3035  * The back references tell us which tree holds a ref on a block,
3036  * but it is possible for the tree root field in the reference to
3037  * reflect the original root before a snapshot was made.  In this
3038  * case we should search through all the children of a given root
3039  * to find potential holders of references on a block.
3040  *
3041  * Instead, we do something a little less fancy and just search
3042  * all the roots for a given key/block combination.
3043  */
3044 static int find_root_for_ref(struct btrfs_root *root,
3045                              struct btrfs_path *path,
3046                              struct btrfs_key *key0,
3047                              int level,
3048                              int file_key,
3049                              struct btrfs_root **found_root,
3050                              u64 bytenr)
3051 {
3052         struct btrfs_key root_location;
3053         struct btrfs_root *cur_root = *found_root;
3054         struct btrfs_file_extent_item *file_extent;
3055         u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
3056         u64 found_bytenr;
3057         int ret;
3058
3059         root_location.offset = (u64)-1;
3060         root_location.type = BTRFS_ROOT_ITEM_KEY;
3061         path->lowest_level = level;
3062         path->reada = 0;
3063         while(1) {
3064                 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
3065                 found_bytenr = 0;
3066                 if (ret == 0 && file_key) {
3067                         struct extent_buffer *leaf = path->nodes[0];
3068                         file_extent = btrfs_item_ptr(leaf, path->slots[0],
3069                                              struct btrfs_file_extent_item);
3070                         if (btrfs_file_extent_type(leaf, file_extent) ==
3071                             BTRFS_FILE_EXTENT_REG) {
3072                                 found_bytenr =
3073                                         btrfs_file_extent_disk_bytenr(leaf,
3074                                                                file_extent);
3075                        }
3076                 } else if (!file_key) {
3077                         if (path->nodes[level])
3078                                 found_bytenr = path->nodes[level]->start;
3079                 }
3080
3081                 btrfs_release_path(cur_root, path);
3082
3083                 if (found_bytenr == bytenr) {
3084                         *found_root = cur_root;
3085                         ret = 0;
3086                         goto out;
3087                 }
3088                 ret = btrfs_search_root(root->fs_info->tree_root,
3089                                         root_search_start, &root_search_start);
3090                 if (ret)
3091                         break;
3092
3093                 root_location.objectid = root_search_start;
3094                 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
3095                                                       &root_location);
3096                 if (!cur_root) {
3097                         ret = 1;
3098                         break;
3099                 }
3100         }
3101 out:
3102         path->lowest_level = 0;
3103         return ret;
3104 }
3105
3106 /*
3107  * note, this releases the path
3108  */
3109 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
3110                                   struct btrfs_path *path,
3111                                   struct btrfs_key *extent_key,
3112                                   u64 *last_file_objectid,
3113                                   u64 *last_file_offset,
3114                                   u64 *last_file_root,
3115                                   u64 last_extent)
3116 {
3117         struct inode *inode;
3118         struct btrfs_root *found_root;
3119         struct btrfs_key root_location;
3120         struct btrfs_key found_key;
3121         struct btrfs_extent_ref *ref;
3122         u64 ref_root;
3123         u64 ref_gen;
3124         u64 ref_objectid;
3125         u64 ref_offset;
3126         int ret;
3127         int level;
3128
3129         WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
3130
3131         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
3132                              struct btrfs_extent_ref);
3133         ref_root = btrfs_ref_root(path->nodes[0], ref);
3134         ref_gen = btrfs_ref_generation(path->nodes[0], ref);
3135         ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
3136         ref_offset = btrfs_ref_offset(path->nodes[0], ref);
3137         btrfs_release_path(extent_root, path);
3138
3139         root_location.objectid = ref_root;
3140         if (ref_gen == 0)
3141                 root_location.offset = 0;
3142         else
3143                 root_location.offset = (u64)-1;
3144         root_location.type = BTRFS_ROOT_ITEM_KEY;
3145
3146         found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
3147                                                 &root_location);
3148         BUG_ON(!found_root);
3149         mutex_unlock(&extent_root->fs_info->alloc_mutex);
3150
3151         if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
3152                 found_key.objectid = ref_objectid;
3153                 found_key.type = BTRFS_EXTENT_DATA_KEY;
3154                 found_key.offset = ref_offset;
3155                 level = 0;
3156
3157                 if (last_extent == extent_key->objectid &&
3158                     *last_file_objectid == ref_objectid &&
3159                     *last_file_offset == ref_offset &&
3160                     *last_file_root == ref_root)
3161                         goto out;
3162
3163                 ret = find_root_for_ref(extent_root, path, &found_key,
3164                                         level, 1, &found_root,
3165                                         extent_key->objectid);
3166
3167                 if (ret)
3168                         goto out;
3169
3170                 if (last_extent == extent_key->objectid &&
3171                     *last_file_objectid == ref_objectid &&
3172                     *last_file_offset == ref_offset &&
3173                     *last_file_root == ref_root)
3174                         goto out;
3175
3176                 inode = btrfs_iget_locked(extent_root->fs_info->sb,
3177                                           ref_objectid, found_root);
3178                 if (inode->i_state & I_NEW) {
3179                         /* the inode and parent dir are two different roots */
3180                         BTRFS_I(inode)->root = found_root;
3181                         BTRFS_I(inode)->location.objectid = ref_objectid;
3182                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
3183                         BTRFS_I(inode)->location.offset = 0;
3184                         btrfs_read_locked_inode(inode);
3185                         unlock_new_inode(inode);
3186
3187                 }
3188                 /* this can happen if the reference is not against
3189                  * the latest version of the tree root
3190                  */
3191                 if (is_bad_inode(inode))
3192                         goto out;
3193
3194                 *last_file_objectid = inode->i_ino;
3195                 *last_file_root = found_root->root_key.objectid;
3196                 *last_file_offset = ref_offset;
3197
3198                 relocate_inode_pages(inode, ref_offset, extent_key->offset);
3199                 iput(inode);
3200         } else {
3201                 struct btrfs_trans_handle *trans;
3202                 struct extent_buffer *eb;
3203                 int needs_lock = 0;
3204
3205                 eb = read_tree_block(found_root, extent_key->objectid,
3206                                      extent_key->offset, 0);
3207                 btrfs_tree_lock(eb);
3208                 level = btrfs_header_level(eb);
3209
3210                 if (level == 0)
3211                         btrfs_item_key_to_cpu(eb, &found_key, 0);
3212                 else
3213                         btrfs_node_key_to_cpu(eb, &found_key, 0);
3214
3215                 btrfs_tree_unlock(eb);
3216                 free_extent_buffer(eb);
3217
3218                 ret = find_root_for_ref(extent_root, path, &found_key,
3219                                         level, 0, &found_root,
3220                                         extent_key->objectid);
3221
3222                 if (ret)
3223                         goto out;
3224
3225                 /*
3226                  * right here almost anything could happen to our key,
3227                  * but that's ok.  The cow below will either relocate it
3228                  * or someone else will have relocated it.  Either way,
3229                  * it is in a different spot than it was before and
3230                  * we're happy.
3231                  */
3232
3233                 trans = btrfs_start_transaction(found_root, 1);
3234
3235                 if (found_root == extent_root->fs_info->extent_root ||
3236                     found_root == extent_root->fs_info->chunk_root ||
3237                     found_root == extent_root->fs_info->dev_root) {
3238                         needs_lock = 1;
3239                         mutex_lock(&extent_root->fs_info->alloc_mutex);
3240                 }
3241
3242                 path->lowest_level = level;
3243                 path->reada = 2;
3244                 ret = btrfs_search_slot(trans, found_root, &found_key, path,
3245                                         0, 1);
3246                 path->lowest_level = 0;
3247                 btrfs_release_path(found_root, path);
3248
3249                 if (found_root == found_root->fs_info->extent_root)
3250                         btrfs_extent_post_op(trans, found_root);
3251                 if (needs_lock)
3252                         mutex_unlock(&extent_root->fs_info->alloc_mutex);
3253
3254                 btrfs_end_transaction(trans, found_root);
3255
3256         }
3257 out:
3258         mutex_lock(&extent_root->fs_info->alloc_mutex);
3259         return 0;
3260 }
3261
3262 static int noinline del_extent_zero(struct btrfs_root *extent_root,
3263                                     struct btrfs_path *path,
3264                                     struct btrfs_key *extent_key)
3265 {
3266         int ret;
3267         struct btrfs_trans_handle *trans;
3268
3269         trans = btrfs_start_transaction(extent_root, 1);
3270         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
3271         if (ret > 0) {
3272                 ret = -EIO;
3273                 goto out;
3274         }
3275         if (ret < 0)
3276                 goto out;
3277         ret = btrfs_del_item(trans, extent_root, path);
3278 out:
3279         btrfs_end_transaction(trans, extent_root);
3280         return ret;
3281 }
3282
3283 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
3284                                         struct btrfs_path *path,
3285                                         struct btrfs_key *extent_key)
3286 {
3287         struct btrfs_key key;
3288         struct btrfs_key found_key;
3289         struct extent_buffer *leaf;
3290         u64 last_file_objectid = 0;
3291         u64 last_file_root = 0;
3292         u64 last_file_offset = (u64)-1;
3293         u64 last_extent = 0;
3294         u32 nritems;
3295         u32 item_size;
3296         int ret = 0;
3297
3298         if (extent_key->objectid == 0) {
3299                 ret = del_extent_zero(extent_root, path, extent_key);
3300                 goto out;
3301         }
3302         key.objectid = extent_key->objectid;
3303         key.type = BTRFS_EXTENT_REF_KEY;
3304         key.offset = 0;
3305
3306         while(1) {
3307                 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3308
3309                 if (ret < 0)
3310                         goto out;
3311
3312                 ret = 0;
3313                 leaf = path->nodes[0];
3314                 nritems = btrfs_header_nritems(leaf);
3315                 if (path->slots[0] == nritems) {
3316                         ret = btrfs_next_leaf(extent_root, path);
3317                         if (ret > 0) {
3318                                 ret = 0;
3319                                 goto out;
3320                         }
3321                         if (ret < 0)
3322                                 goto out;
3323                         leaf = path->nodes[0];
3324                 }
3325
3326                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3327                 if (found_key.objectid != extent_key->objectid) {
3328                         break;
3329                 }
3330
3331                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
3332                         break;
3333                 }
3334
3335                 key.offset = found_key.offset + 1;
3336                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3337
3338                 ret = relocate_one_reference(extent_root, path, extent_key,
3339                                              &last_file_objectid,
3340                                              &last_file_offset,
3341                                              &last_file_root, last_extent);
3342                 if (ret)
3343                         goto out;
3344                 last_extent = extent_key->objectid;
3345         }
3346         ret = 0;
3347 out:
3348         btrfs_release_path(extent_root, path);
3349         return ret;
3350 }
3351
3352 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
3353 {
3354         u64 num_devices;
3355         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
3356                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
3357
3358         num_devices = root->fs_info->fs_devices->num_devices;
3359         if (num_devices == 1) {
3360                 stripped |= BTRFS_BLOCK_GROUP_DUP;
3361                 stripped = flags & ~stripped;
3362
3363                 /* turn raid0 into single device chunks */
3364                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
3365                         return stripped;
3366
3367                 /* turn mirroring into duplication */
3368                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3369                              BTRFS_BLOCK_GROUP_RAID10))
3370                         return stripped | BTRFS_BLOCK_GROUP_DUP;
3371                 return flags;
3372         } else {
3373                 /* they already had raid on here, just return */
3374                 if (flags & stripped)
3375                         return flags;
3376
3377                 stripped |= BTRFS_BLOCK_GROUP_DUP;
3378                 stripped = flags & ~stripped;
3379
3380                 /* switch duplicated blocks with raid1 */
3381                 if (flags & BTRFS_BLOCK_GROUP_DUP)
3382                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
3383
3384                 /* turn single device chunks into raid0 */
3385                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
3386         }
3387         return flags;
3388 }
3389
3390 int __alloc_chunk_for_shrink(struct btrfs_root *root,
3391                      struct btrfs_block_group_cache *shrink_block_group,
3392                      int force)
3393 {
3394         struct btrfs_trans_handle *trans;
3395         u64 new_alloc_flags;
3396         u64 calc;
3397
3398         spin_lock(&shrink_block_group->lock);
3399         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
3400                 spin_unlock(&shrink_block_group->lock);
3401                 mutex_unlock(&root->fs_info->alloc_mutex);
3402
3403                 trans = btrfs_start_transaction(root, 1);
3404                 mutex_lock(&root->fs_info->alloc_mutex);
3405                 spin_lock(&shrink_block_group->lock);
3406
3407                 new_alloc_flags = update_block_group_flags(root,
3408                                                    shrink_block_group->flags);
3409                 if (new_alloc_flags != shrink_block_group->flags) {
3410                         calc =
3411                              btrfs_block_group_used(&shrink_block_group->item);
3412                 } else {
3413                         calc = shrink_block_group->key.offset;
3414                 }
3415                 spin_unlock(&shrink_block_group->lock);
3416
3417                 do_chunk_alloc(trans, root->fs_info->extent_root,
3418                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
3419
3420                 mutex_unlock(&root->fs_info->alloc_mutex);
3421                 btrfs_end_transaction(trans, root);
3422                 mutex_lock(&root->fs_info->alloc_mutex);
3423         } else
3424                 spin_unlock(&shrink_block_group->lock);
3425         return 0;
3426 }
3427
3428 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
3429 {
3430         struct btrfs_trans_handle *trans;
3431         struct btrfs_root *tree_root = root->fs_info->tree_root;
3432         struct btrfs_path *path;
3433         u64 cur_byte;
3434         u64 total_found;
3435         u64 shrink_last_byte;
3436         struct btrfs_block_group_cache *shrink_block_group;
3437         struct btrfs_key key;
3438         struct btrfs_key found_key;
3439         struct extent_buffer *leaf;
3440         u32 nritems;
3441         int ret;
3442         int progress;
3443
3444         mutex_lock(&root->fs_info->alloc_mutex);
3445         shrink_block_group = btrfs_lookup_block_group(root->fs_info,
3446                                                       shrink_start);
3447         BUG_ON(!shrink_block_group);
3448
3449         shrink_last_byte = shrink_block_group->key.objectid +
3450                 shrink_block_group->key.offset;
3451
3452         shrink_block_group->space_info->total_bytes -=
3453                 shrink_block_group->key.offset;
3454         path = btrfs_alloc_path();
3455         root = root->fs_info->extent_root;
3456         path->reada = 2;
3457
3458         printk("btrfs relocating block group %llu flags %llu\n",
3459                (unsigned long long)shrink_start,
3460                (unsigned long long)shrink_block_group->flags);
3461
3462         __alloc_chunk_for_shrink(root, shrink_block_group, 1);
3463
3464 again:
3465
3466         shrink_block_group->ro = 1;
3467
3468         total_found = 0;
3469         progress = 0;
3470         key.objectid = shrink_start;
3471         key.offset = 0;
3472         key.type = 0;
3473         cur_byte = key.objectid;
3474
3475         mutex_unlock(&root->fs_info->alloc_mutex);
3476
3477         btrfs_start_delalloc_inodes(root);
3478         btrfs_wait_ordered_extents(tree_root, 0);
3479
3480         mutex_lock(&root->fs_info->alloc_mutex);
3481
3482         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3483         if (ret < 0)
3484                 goto out;
3485
3486         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
3487         if (ret < 0)
3488                 goto out;
3489
3490         if (ret == 0) {
3491                 leaf = path->nodes[0];
3492                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3493                 if (found_key.objectid + found_key.offset > shrink_start &&
3494                     found_key.objectid < shrink_last_byte) {
3495                         cur_byte = found_key.objectid;
3496                         key.objectid = cur_byte;
3497                 }
3498         }
3499         btrfs_release_path(root, path);
3500
3501         while(1) {
3502                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3503                 if (ret < 0)
3504                         goto out;
3505
3506 next:
3507                 leaf = path->nodes[0];
3508                 nritems = btrfs_header_nritems(leaf);
3509                 if (path->slots[0] >= nritems) {
3510                         ret = btrfs_next_leaf(root, path);
3511                         if (ret < 0)
3512                                 goto out;
3513                         if (ret == 1) {
3514                                 ret = 0;
3515                                 break;
3516                         }
3517                         leaf = path->nodes[0];
3518                         nritems = btrfs_header_nritems(leaf);
3519                 }
3520
3521                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3522
3523                 if (found_key.objectid >= shrink_last_byte)
3524                         break;
3525
3526                 if (progress && need_resched()) {
3527                         memcpy(&key, &found_key, sizeof(key));
3528                         cond_resched();
3529                         btrfs_release_path(root, path);
3530                         btrfs_search_slot(NULL, root, &key, path, 0, 0);
3531                         progress = 0;
3532                         goto next;
3533                 }
3534                 progress = 1;
3535
3536                 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3537                     found_key.objectid + found_key.offset <= cur_byte) {
3538                         memcpy(&key, &found_key, sizeof(key));
3539                         key.offset++;
3540                         path->slots[0]++;
3541                         goto next;
3542                 }
3543
3544                 total_found++;
3545                 cur_byte = found_key.objectid + found_key.offset;
3546                 key.objectid = cur_byte;
3547                 btrfs_release_path(root, path);
3548                 ret = relocate_one_extent(root, path, &found_key);
3549                 __alloc_chunk_for_shrink(root, shrink_block_group, 0);
3550         }
3551
3552         btrfs_release_path(root, path);
3553
3554         if (total_found > 0) {
3555                 printk("btrfs relocate found %llu last extent was %llu\n",
3556                        (unsigned long long)total_found,
3557                        (unsigned long long)found_key.objectid);
3558                 mutex_unlock(&root->fs_info->alloc_mutex);
3559                 trans = btrfs_start_transaction(tree_root, 1);
3560                 btrfs_commit_transaction(trans, tree_root);
3561
3562                 btrfs_clean_old_snapshots(tree_root);
3563
3564                 btrfs_start_delalloc_inodes(root);
3565                 btrfs_wait_ordered_extents(tree_root, 0);
3566
3567                 trans = btrfs_start_transaction(tree_root, 1);
3568                 btrfs_commit_transaction(trans, tree_root);
3569                 mutex_lock(&root->fs_info->alloc_mutex);
3570                 goto again;
3571         }
3572
3573         /*
3574          * we've freed all the extents, now remove the block
3575          * group item from the tree
3576          */
3577         mutex_unlock(&root->fs_info->alloc_mutex);
3578
3579         trans = btrfs_start_transaction(root, 1);
3580
3581         mutex_lock(&root->fs_info->alloc_mutex);
3582         memcpy(&key, &shrink_block_group->key, sizeof(key));
3583
3584         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3585         if (ret > 0)
3586                 ret = -EIO;
3587         if (ret < 0) {
3588                 btrfs_end_transaction(trans, root);
3589                 goto out;
3590         }
3591
3592         spin_lock(&root->fs_info->block_group_cache_lock);
3593         rb_erase(&shrink_block_group->cache_node,
3594                  &root->fs_info->block_group_cache_tree);
3595         spin_unlock(&root->fs_info->block_group_cache_lock);
3596
3597         ret = btrfs_remove_free_space(shrink_block_group, key.objectid,
3598                                       key.offset);
3599         if (ret) {
3600                 btrfs_end_transaction(trans, root);
3601                 goto out;
3602         }
3603         /*
3604         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3605         kfree(shrink_block_group);
3606         */
3607
3608         btrfs_del_item(trans, root, path);
3609         btrfs_release_path(root, path);
3610         mutex_unlock(&root->fs_info->alloc_mutex);
3611         btrfs_commit_transaction(trans, root);
3612
3613         mutex_lock(&root->fs_info->alloc_mutex);
3614
3615         /* the code to unpin extents might set a few bits in the free
3616          * space cache for this range again
3617          */
3618         /* XXX? */
3619         ret = btrfs_remove_free_space(shrink_block_group, key.objectid,
3620                                       key.offset);
3621 out:
3622         btrfs_free_path(path);
3623         mutex_unlock(&root->fs_info->alloc_mutex);
3624         return ret;
3625 }
3626
3627 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3628                            struct btrfs_key *key)
3629 {
3630         int ret = 0;
3631         struct btrfs_key found_key;
3632         struct extent_buffer *leaf;
3633         int slot;
3634
3635         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3636         if (ret < 0)
3637                 goto out;
3638
3639         while(1) {
3640                 slot = path->slots[0];
3641                 leaf = path->nodes[0];
3642                 if (slot >= btrfs_header_nritems(leaf)) {
3643                         ret = btrfs_next_leaf(root, path);
3644                         if (ret == 0)
3645                                 continue;
3646                         if (ret < 0)
3647                                 goto out;
3648                         break;
3649                 }
3650                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3651
3652                 if (found_key.objectid >= key->objectid &&
3653                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3654                         ret = 0;
3655                         goto out;
3656                 }
3657                 path->slots[0]++;
3658         }
3659         ret = -ENOENT;
3660 out:
3661         return ret;
3662 }
3663
3664 int btrfs_read_block_groups(struct btrfs_root *root)
3665 {
3666         struct btrfs_path *path;
3667         int ret;
3668         struct btrfs_block_group_cache *cache;
3669         struct btrfs_fs_info *info = root->fs_info;
3670         struct btrfs_space_info *space_info;
3671         struct btrfs_key key;
3672         struct btrfs_key found_key;
3673         struct extent_buffer *leaf;
3674
3675         root = info->extent_root;
3676         key.objectid = 0;
3677         key.offset = 0;
3678         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3679         path = btrfs_alloc_path();
3680         if (!path)
3681                 return -ENOMEM;
3682
3683         mutex_lock(&root->fs_info->alloc_mutex);
3684         while(1) {
3685                 ret = find_first_block_group(root, path, &key);
3686                 if (ret > 0) {
3687                         ret = 0;
3688                         goto error;
3689                 }
3690                 if (ret != 0)
3691                         goto error;
3692
3693                 leaf = path->nodes[0];
3694                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3695                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3696                 if (!cache) {
3697                         ret = -ENOMEM;
3698                         break;
3699                 }
3700
3701                 spin_lock_init(&cache->lock);
3702                 INIT_LIST_HEAD(&cache->list);
3703                 read_extent_buffer(leaf, &cache->item,
3704                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3705                                    sizeof(cache->item));
3706                 memcpy(&cache->key, &found_key, sizeof(found_key));
3707
3708                 key.objectid = found_key.objectid + found_key.offset;
3709                 btrfs_release_path(root, path);
3710                 cache->flags = btrfs_block_group_flags(&cache->item);
3711
3712                 ret = update_space_info(info, cache->flags, found_key.offset,
3713                                         btrfs_block_group_used(&cache->item),
3714                                         &space_info);
3715                 BUG_ON(ret);
3716                 cache->space_info = space_info;
3717                 spin_lock(&space_info->lock);
3718                 list_add(&cache->list, &space_info->block_groups);
3719                 spin_unlock(&space_info->lock);
3720
3721                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
3722                 BUG_ON(ret);
3723
3724                 if (key.objectid >=
3725                     btrfs_super_total_bytes(&info->super_copy))
3726                         break;
3727         }
3728         ret = 0;
3729 error:
3730         btrfs_free_path(path);
3731         mutex_unlock(&root->fs_info->alloc_mutex);
3732         return ret;
3733 }
3734
3735 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3736                            struct btrfs_root *root, u64 bytes_used,
3737                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3738                            u64 size)
3739 {
3740         int ret;
3741         struct btrfs_root *extent_root;
3742         struct btrfs_block_group_cache *cache;
3743
3744         WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
3745         extent_root = root->fs_info->extent_root;
3746
3747         root->fs_info->last_trans_new_blockgroup = trans->transid;
3748
3749         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3750         if (!cache)
3751                 return -ENOMEM;
3752
3753         cache->key.objectid = chunk_offset;
3754         cache->key.offset = size;
3755         spin_lock_init(&cache->lock);
3756         INIT_LIST_HEAD(&cache->list);
3757         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3758
3759         btrfs_set_block_group_used(&cache->item, bytes_used);
3760         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3761         cache->flags = type;
3762         btrfs_set_block_group_flags(&cache->item, type);
3763
3764         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3765                                 &cache->space_info);
3766         BUG_ON(ret);
3767         spin_lock(&cache->space_info->lock);
3768         list_add(&cache->list, &cache->space_info->block_groups);
3769         spin_unlock(&cache->space_info->lock);
3770
3771         ret = btrfs_add_block_group_cache(root->fs_info, cache);
3772         BUG_ON(ret);
3773
3774         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3775                                 sizeof(cache->item));
3776         BUG_ON(ret);
3777
3778         finish_current_insert(trans, extent_root);
3779         ret = del_pending_extents(trans, extent_root);
3780         BUG_ON(ret);
3781         set_avail_alloc_bits(extent_root->fs_info, type);
3782
3783         return 0;
3784 }