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 <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <linux/slab.h>
  26 #include "compat.h"
  27 #include "hash.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "print-tree.h"
  31 #include "transaction.h"
  32 #include "volumes.h"
  33 #include "locking.h"
  34 #include "free-space-cache.h"
  35
  36 /* control flags for do_chunk_alloc's force field
  37  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  38  * if we really need one.
  39  *
  40  * CHUNK_ALLOC_FORCE means it must try to allocate one
  41  *
  42  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  43  * if we have very few chunks already allocated.  This is
  44  * used as part of the clustering code to help make sure
  45  * we have a good pool of storage to cluster in, without
  46  * filling the FS with empty chunks
  47  *
  48  */
  49 enum {
  50         CHUNK_ALLOC_NO_FORCE = 0,
  51         CHUNK_ALLOC_FORCE = 1,
  52         CHUNK_ALLOC_LIMITED = 2,
  53 };
  54
  55 static int update_block_group(struct btrfs_trans_handle *trans,
  56                               struct btrfs_root *root,
  57                               u64 bytenr, u64 num_bytes, int alloc);
  58 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  59                                 struct btrfs_root *root,
  60                                 u64 bytenr, u64 num_bytes, u64 parent,
  61                                 u64 root_objectid, u64 owner_objectid,
  62                                 u64 owner_offset, int refs_to_drop,
  63                                 struct btrfs_delayed_extent_op *extra_op);
  64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  65                                     struct extent_buffer *leaf,
  66                                     struct btrfs_extent_item *ei);
  67 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  68                                       struct btrfs_root *root,
  69                                       u64 parent, u64 root_objectid,
  70                                       u64 flags, u64 owner, u64 offset,
  71                                       struct btrfs_key *ins, int ref_mod);
  72 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  73                                      struct btrfs_root *root,
  74                                      u64 parent, u64 root_objectid,
  75                                      u64 flags, struct btrfs_disk_key *key,
  76                                      int level, struct btrfs_key *ins);
  77 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  78                           struct btrfs_root *extent_root, u64 alloc_bytes,
  79                           u64 flags, int force);
  80 static int find_next_key(struct btrfs_path *path, int level,
  81                          struct btrfs_key *key);
  82 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  83                             int dump_block_groups);
  84
  85 static noinline int
  86 block_group_cache_done(struct btrfs_block_group_cache *cache)
  87 {
  88         smp_mb();
  89         return cache->cached == BTRFS_CACHE_FINISHED;
  90 }
  91
  92 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  93 {
  94         return (cache->flags & bits) == bits;
  95 }
  96
  97 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  98 {
  99         atomic_inc(&cache->count);
 100 }
 101
 102 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 103 {
 104         if (atomic_dec_and_test(&cache->count)) {
 105                 WARN_ON(cache->pinned > 0);
 106                 WARN_ON(cache->reserved > 0);
 107                 WARN_ON(cache->reserved_pinned > 0);
 108                 kfree(cache->free_space_ctl);
 109                 kfree(cache);
 110         }
 111 }
 112
 113 /*
 114  * this adds the block group to the fs_info rb tree for the block group
 115  * cache
 116  */
 117 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 118                                 struct btrfs_block_group_cache *block_group)
 119 {
 120         struct rb_node **p;
 121         struct rb_node *parent = NULL;
 122         struct btrfs_block_group_cache *cache;
 123
 124         spin_lock(&info->block_group_cache_lock);
 125         p = &info->block_group_cache_tree.rb_node;
 126
 127         while (*p) {
 128                 parent = *p;
 129                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 130                                  cache_node);
 131                 if (block_group->key.objectid < cache->key.objectid) {
 132                         p = &(*p)->rb_left;
 133                 } else if (block_group->key.objectid > cache->key.objectid) {
 134                         p = &(*p)->rb_right;
 135                 } else {
 136                         spin_unlock(&info->block_group_cache_lock);
 137                         return -EEXIST;
 138                 }
 139         }
 140
 141         rb_link_node(&block_group->cache_node, parent, p);
 142         rb_insert_color(&block_group->cache_node,
 143                         &info->block_group_cache_tree);
 144         spin_unlock(&info->block_group_cache_lock);
 145
 146         return 0;
 147 }
 148
 149 /*
 150  * This will return the block group at or after bytenr if contains is 0, else
 151  * it will return the block group that contains the bytenr
 152  */
 153 static struct btrfs_block_group_cache *
 154 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 155                               int contains)
 156 {
 157         struct btrfs_block_group_cache *cache, *ret = NULL;
 158         struct rb_node *n;
 159         u64 end, start;
 160
 161         spin_lock(&info->block_group_cache_lock);
 162         n = info->block_group_cache_tree.rb_node;
 163
 164         while (n) {
 165                 cache = rb_entry(n, struct btrfs_block_group_cache,
 166                                  cache_node);
 167                 end = cache->key.objectid + cache->key.offset - 1;
 168                 start = cache->key.objectid;
 169
 170                 if (bytenr < start) {
 171                         if (!contains && (!ret || start < ret->key.objectid))
 172                                 ret = cache;
 173                         n = n->rb_left;
 174                 } else if (bytenr > start) {
 175                         if (contains && bytenr <= end) {
 176                                 ret = cache;
 177                                 break;
 178                         }
 179                         n = n->rb_right;
 180                 } else {
 181                         ret = cache;
 182                         break;
 183                 }
 184         }
 185         if (ret)
 186                 btrfs_get_block_group(ret);
 187         spin_unlock(&info->block_group_cache_lock);
 188
 189         return ret;
 190 }
 191
 192 static int add_excluded_extent(struct btrfs_root *root,
 193                                u64 start, u64 num_bytes)
 194 {
 195         u64 end = start + num_bytes - 1;
 196         set_extent_bits(&root->fs_info->freed_extents[0],
 197                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 198         set_extent_bits(&root->fs_info->freed_extents[1],
 199                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 200         return 0;
 201 }
 202
 203 static void free_excluded_extents(struct btrfs_root *root,
 204                                   struct btrfs_block_group_cache *cache)
 205 {
 206         u64 start, end;
 207
 208         start = cache->key.objectid;
 209         end = start + cache->key.offset - 1;
 210
 211         clear_extent_bits(&root->fs_info->freed_extents[0],
 212                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 213         clear_extent_bits(&root->fs_info->freed_extents[1],
 214                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 215 }
 216
 217 static int exclude_super_stripes(struct btrfs_root *root,
 218                                  struct btrfs_block_group_cache *cache)
 219 {
 220         u64 bytenr;
 221         u64 *logical;
 222         int stripe_len;
 223         int i, nr, ret;
 224
 225         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 226                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 227                 cache->bytes_super += stripe_len;
 228                 ret = add_excluded_extent(root, cache->key.objectid,
 229                                           stripe_len);
 230                 BUG_ON(ret);
 231         }
 232
 233         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 234                 bytenr = btrfs_sb_offset(i);
 235                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 236                                        cache->key.objectid, bytenr,
 237                                        0, &logical, &nr, &stripe_len);
 238                 BUG_ON(ret);
 239
 240                 while (nr--) {
 241                         cache->bytes_super += stripe_len;
 242                         ret = add_excluded_extent(root, logical[nr],
 243                                                   stripe_len);
 244                         BUG_ON(ret);
 245                 }
 246
 247                 kfree(logical);
 248         }
 249         return 0;
 250 }
 251
 252 static struct btrfs_caching_control *
 253 get_caching_control(struct btrfs_block_group_cache *cache)
 254 {
 255         struct btrfs_caching_control *ctl;
 256
 257         spin_lock(&cache->lock);
 258         if (cache->cached != BTRFS_CACHE_STARTED) {
 259                 spin_unlock(&cache->lock);
 260                 return NULL;
 261         }
 262
 263         /* We're loading it the fast way, so we don't have a caching_ctl. */
 264         if (!cache->caching_ctl) {
 265                 spin_unlock(&cache->lock);
 266                 return NULL;
 267         }
 268
 269         ctl = cache->caching_ctl;
 270         atomic_inc(&ctl->count);
 271         spin_unlock(&cache->lock);
 272         return ctl;
 273 }
 274
 275 static void put_caching_control(struct btrfs_caching_control *ctl)
 276 {
 277         if (atomic_dec_and_test(&ctl->count))
 278                 kfree(ctl);
 279 }
 280
 281 /*
 282  * this is only called by cache_block_group, since we could have freed extents
 283  * we need to check the pinned_extents for any extents that can't be used yet
 284  * since their free space will be released as soon as the transaction commits.
 285  */
 286 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 287                               struct btrfs_fs_info *info, u64 start, u64 end)
 288 {
 289         u64 extent_start, extent_end, size, total_added = 0;
 290         int ret;
 291
 292         while (start < end) {
 293                 ret = find_first_extent_bit(info->pinned_extents, start,
 294                                             &extent_start, &extent_end,
 295                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 296                 if (ret)
 297                         break;
 298
 299                 if (extent_start <= start) {
 300                         start = extent_end + 1;
 301                 } else if (extent_start > start && extent_start < end) {
 302                         size = extent_start - start;
 303                         total_added += size;
 304                         ret = btrfs_add_free_space(block_group, start,
 305                                                    size);
 306                         BUG_ON(ret);
 307                         start = extent_end + 1;
 308                 } else {
 309                         break;
 310                 }
 311         }
 312
 313         if (start < end) {
 314                 size = end - start;
 315                 total_added += size;
 316                 ret = btrfs_add_free_space(block_group, start, size);
 317                 BUG_ON(ret);
 318         }
 319
 320         return total_added;
 321 }
 322
 323 static int caching_kthread(void *data)
 324 {
 325         struct btrfs_block_group_cache *block_group = data;
 326         struct btrfs_fs_info *fs_info = block_group->fs_info;
 327         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 328         struct btrfs_root *extent_root = fs_info->extent_root;
 329         struct btrfs_path *path;
 330         struct extent_buffer *leaf;
 331         struct btrfs_key key;
 332         u64 total_found = 0;
 333         u64 last = 0;
 334         u32 nritems;
 335         int ret = 0;
 336
 337         path = btrfs_alloc_path();
 338         if (!path)
 339                 return -ENOMEM;
 340
 341         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 342
 343         /*
 344          * We don't want to deadlock with somebody trying to allocate a new
 345          * extent for the extent root while also trying to search the extent
 346          * root to add free space.  So we skip locking and search the commit
 347          * root, since its read-only
 348          */
 349         path->skip_locking = 1;
 350         path->search_commit_root = 1;
 351         path->reada = 2;
 352
 353         key.objectid = last;
 354         key.offset = 0;
 355         key.type = BTRFS_EXTENT_ITEM_KEY;
 356 again:
 357         mutex_lock(&caching_ctl->mutex);
 358         /* need to make sure the commit_root doesn't disappear */
 359         down_read(&fs_info->extent_commit_sem);
 360
 361         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 362         if (ret < 0)
 363                 goto err;
 364
 365         leaf = path->nodes[0];
 366         nritems = btrfs_header_nritems(leaf);
 367
 368         while (1) {
 369                 smp_mb();
 370                 if (fs_info->closing > 1) {
 371                         last = (u64)-1;
 372                         break;
 373                 }
 374
 375                 if (path->slots[0] < nritems) {
 376                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 377                 } else {
 378                         ret = find_next_key(path, 0, &key);
 379                         if (ret)
 380                                 break;
 381
 382                         caching_ctl->progress = last;
 383                         btrfs_release_path(extent_root, path);
 384                         up_read(&fs_info->extent_commit_sem);
 385                         mutex_unlock(&caching_ctl->mutex);
 386                         if (btrfs_transaction_in_commit(fs_info))
 387                                 schedule_timeout(1);
 388                         else
 389                                 cond_resched();
 390                         goto again;
 391                 }
 392
 393                 if (key.objectid < block_group->key.objectid) {
 394                         path->slots[0]++;
 395                         continue;
 396                 }
 397
 398                 if (key.objectid >= block_group->key.objectid +
 399                     block_group->key.offset)
 400                         break;
 401
 402                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 403                         total_found += add_new_free_space(block_group,
 404                                                           fs_info, last,
 405                                                           key.objectid);
 406                         last = key.objectid + key.offset;
 407
 408                         if (total_found > (1024 * 1024 * 2)) {
 409                                 total_found = 0;
 410                                 wake_up(&caching_ctl->wait);
 411                         }
 412                 }
 413                 path->slots[0]++;
 414         }
 415         ret = 0;
 416
 417         total_found += add_new_free_space(block_group, fs_info, last,
 418                                           block_group->key.objectid +
 419                                           block_group->key.offset);
 420         caching_ctl->progress = (u64)-1;
 421
 422         spin_lock(&block_group->lock);
 423         block_group->caching_ctl = NULL;
 424         block_group->cached = BTRFS_CACHE_FINISHED;
 425         spin_unlock(&block_group->lock);
 426
 427 err:
 428         btrfs_free_path(path);
 429         up_read(&fs_info->extent_commit_sem);
 430
 431         free_excluded_extents(extent_root, block_group);
 432
 433         mutex_unlock(&caching_ctl->mutex);
 434         wake_up(&caching_ctl->wait);
 435
 436         put_caching_control(caching_ctl);
 437         atomic_dec(&block_group->space_info->caching_threads);
 438         btrfs_put_block_group(block_group);
 439
 440         return 0;
 441 }
 442
 443 static int cache_block_group(struct btrfs_block_group_cache *cache,
 444                              struct btrfs_trans_handle *trans,
 445                              struct btrfs_root *root,
 446                              int load_cache_only)
 447 {
 448         struct btrfs_fs_info *fs_info = cache->fs_info;
 449         struct btrfs_caching_control *caching_ctl;
 450         struct task_struct *tsk;
 451         int ret = 0;
 452
 453         smp_mb();
 454         if (cache->cached != BTRFS_CACHE_NO)
 455                 return 0;
 456
 457         /*
 458          * We can't do the read from on-disk cache during a commit since we need
 459          * to have the normal tree locking.  Also if we are currently trying to
 460          * allocate blocks for the tree root we can't do the fast caching since
 461          * we likely hold important locks.
 462          */
 463         if (trans && (!trans->transaction->in_commit) &&
 464             (root && root != root->fs_info->tree_root)) {
 465                 spin_lock(&cache->lock);
 466                 if (cache->cached != BTRFS_CACHE_NO) {
 467                         spin_unlock(&cache->lock);
 468                         return 0;
 469                 }
 470                 cache->cached = BTRFS_CACHE_STARTED;
 471                 spin_unlock(&cache->lock);
 472
 473                 ret = load_free_space_cache(fs_info, cache);
 474
 475                 spin_lock(&cache->lock);
 476                 if (ret == 1) {
 477                         cache->cached = BTRFS_CACHE_FINISHED;
 478                         cache->last_byte_to_unpin = (u64)-1;
 479                 } else {
 480                         cache->cached = BTRFS_CACHE_NO;
 481                 }
 482                 spin_unlock(&cache->lock);
 483                 if (ret == 1) {
 484                         free_excluded_extents(fs_info->extent_root, cache);
 485                         return 0;
 486                 }
 487         }
 488
 489         if (load_cache_only)
 490                 return 0;
 491
 492         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 493         BUG_ON(!caching_ctl);
 494
 495         INIT_LIST_HEAD(&caching_ctl->list);
 496         mutex_init(&caching_ctl->mutex);
 497         init_waitqueue_head(&caching_ctl->wait);
 498         caching_ctl->block_group = cache;
 499         caching_ctl->progress = cache->key.objectid;
 500         /* one for caching kthread, one for caching block group list */
 501         atomic_set(&caching_ctl->count, 2);
 502
 503         spin_lock(&cache->lock);
 504         if (cache->cached != BTRFS_CACHE_NO) {
 505                 spin_unlock(&cache->lock);
 506                 kfree(caching_ctl);
 507                 return 0;
 508         }
 509         cache->caching_ctl = caching_ctl;
 510         cache->cached = BTRFS_CACHE_STARTED;
 511         spin_unlock(&cache->lock);
 512
 513         down_write(&fs_info->extent_commit_sem);
 514         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 515         up_write(&fs_info->extent_commit_sem);
 516
 517         atomic_inc(&cache->space_info->caching_threads);
 518         btrfs_get_block_group(cache);
 519
 520         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 521                           cache->key.objectid);
 522         if (IS_ERR(tsk)) {
 523                 ret = PTR_ERR(tsk);
 524                 printk(KERN_ERR "error running thread %d\n", ret);
 525                 BUG();
 526         }
 527
 528         return ret;
 529 }
 530
 531 /*
 532  * return the block group that starts at or after bytenr
 533  */
 534 static struct btrfs_block_group_cache *
 535 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 536 {
 537         struct btrfs_block_group_cache *cache;
 538
 539         cache = block_group_cache_tree_search(info, bytenr, 0);
 540
 541         return cache;
 542 }
 543
 544 /*
 545  * return the block group that contains the given bytenr
 546  */
 547 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 548                                                  struct btrfs_fs_info *info,
 549                                                  u64 bytenr)
 550 {
 551         struct btrfs_block_group_cache *cache;
 552
 553         cache = block_group_cache_tree_search(info, bytenr, 1);
 554
 555         return cache;
 556 }
 557
 558 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 559                                                   u64 flags)
 560 {
 561         struct list_head *head = &info->space_info;
 562         struct btrfs_space_info *found;
 563
 564         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 565                  BTRFS_BLOCK_GROUP_METADATA;
 566
 567         rcu_read_lock();
 568         list_for_each_entry_rcu(found, head, list) {
 569                 if (found->flags & flags) {
 570                         rcu_read_unlock();
 571                         return found;
 572                 }
 573         }
 574         rcu_read_unlock();
 575         return NULL;
 576 }
 577
 578 /*
 579  * after adding space to the filesystem, we need to clear the full flags
 580  * on all the space infos.
 581  */
 582 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 583 {
 584         struct list_head *head = &info->space_info;
 585         struct btrfs_space_info *found;
 586
 587         rcu_read_lock();
 588         list_for_each_entry_rcu(found, head, list)
 589                 found->full = 0;
 590         rcu_read_unlock();
 591 }
 592
 593 static u64 div_factor(u64 num, int factor)
 594 {
 595         if (factor == 10)
 596                 return num;
 597         num *= factor;
 598         do_div(num, 10);
 599         return num;
 600 }
 601
 602 static u64 div_factor_fine(u64 num, int factor)
 603 {
 604         if (factor == 100)
 605                 return num;
 606         num *= factor;
 607         do_div(num, 100);
 608         return num;
 609 }
 610
 611 u64 btrfs_find_block_group(struct btrfs_root *root,
 612                            u64 search_start, u64 search_hint, int owner)
 613 {
 614         struct btrfs_block_group_cache *cache;
 615         u64 used;
 616         u64 last = max(search_hint, search_start);
 617         u64 group_start = 0;
 618         int full_search = 0;
 619         int factor = 9;
 620         int wrapped = 0;
 621 again:
 622         while (1) {
 623                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 624                 if (!cache)
 625                         break;
 626
 627                 spin_lock(&cache->lock);
 628                 last = cache->key.objectid + cache->key.offset;
 629                 used = btrfs_block_group_used(&cache->item);
 630
 631                 if ((full_search || !cache->ro) &&
 632                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 633                         if (used + cache->pinned + cache->reserved <
 634                             div_factor(cache->key.offset, factor)) {
 635                                 group_start = cache->key.objectid;
 636                                 spin_unlock(&cache->lock);
 637                                 btrfs_put_block_group(cache);
 638                                 goto found;
 639                         }
 640                 }
 641                 spin_unlock(&cache->lock);
 642                 btrfs_put_block_group(cache);
 643                 cond_resched();
 644         }
 645         if (!wrapped) {
 646                 last = search_start;
 647                 wrapped = 1;
 648                 goto again;
 649         }
 650         if (!full_search && factor < 10) {
 651                 last = search_start;
 652                 full_search = 1;
 653                 factor = 10;
 654                 goto again;
 655         }
 656 found:
 657         return group_start;
 658 }
 659
 660 /* simple helper to search for an existing extent at a given offset */
 661 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 662 {
 663         int ret;
 664         struct btrfs_key key;
 665         struct btrfs_path *path;
 666
 667         path = btrfs_alloc_path();
 668         BUG_ON(!path);
 669         key.objectid = start;
 670         key.offset = len;
 671         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 672         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 673                                 0, 0);
 674         btrfs_free_path(path);
 675         return ret;
 676 }
 677
 678 /*
 679  * helper function to lookup reference count and flags of extent.
 680  *
 681  * the head node for delayed ref is used to store the sum of all the
 682  * reference count modifications queued up in the rbtree. the head
 683  * node may also store the extent flags to set. This way you can check
 684  * to see what the reference count and extent flags would be if all of
 685  * the delayed refs are not processed.
 686  */
 687 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 688                              struct btrfs_root *root, u64 bytenr,
 689                              u64 num_bytes, u64 *refs, u64 *flags)
 690 {
 691         struct btrfs_delayed_ref_head *head;
 692         struct btrfs_delayed_ref_root *delayed_refs;
 693         struct btrfs_path *path;
 694         struct btrfs_extent_item *ei;
 695         struct extent_buffer *leaf;
 696         struct btrfs_key key;
 697         u32 item_size;
 698         u64 num_refs;
 699         u64 extent_flags;
 700         int ret;
 701
 702         path = btrfs_alloc_path();
 703         if (!path)
 704                 return -ENOMEM;
 705
 706         key.objectid = bytenr;
 707         key.type = BTRFS_EXTENT_ITEM_KEY;
 708         key.offset = num_bytes;
 709         if (!trans) {
 710                 path->skip_locking = 1;
 711                 path->search_commit_root = 1;
 712         }
 713 again:
 714         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 715                                 &key, path, 0, 0);
 716         if (ret < 0)
 717                 goto out_free;
 718
 719         if (ret == 0) {
 720                 leaf = path->nodes[0];
 721                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 722                 if (item_size >= sizeof(*ei)) {
 723                         ei = btrfs_item_ptr(leaf, path->slots[0],
 724                                             struct btrfs_extent_item);
 725                         num_refs = btrfs_extent_refs(leaf, ei);
 726                         extent_flags = btrfs_extent_flags(leaf, ei);
 727                 } else {
 728 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 729                         struct btrfs_extent_item_v0 *ei0;
 730                         BUG_ON(item_size != sizeof(*ei0));
 731                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 732                                              struct btrfs_extent_item_v0);
 733                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 734                         /* FIXME: this isn't correct for data */
 735                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 736 #else
 737                         BUG();
 738 #endif
 739                 }
 740                 BUG_ON(num_refs == 0);
 741         } else {
 742                 num_refs = 0;
 743                 extent_flags = 0;
 744                 ret = 0;
 745         }
 746
 747         if (!trans)
 748                 goto out;
 749
 750         delayed_refs = &trans->transaction->delayed_refs;
 751         spin_lock(&delayed_refs->lock);
 752         head = btrfs_find_delayed_ref_head(trans, bytenr);
 753         if (head) {
 754                 if (!mutex_trylock(&head->mutex)) {
 755                         atomic_inc(&head->node.refs);
 756                         spin_unlock(&delayed_refs->lock);
 757
 758                         btrfs_release_path(root->fs_info->extent_root, path);
 759
 760                         mutex_lock(&head->mutex);
 761                         mutex_unlock(&head->mutex);
 762                         btrfs_put_delayed_ref(&head->node);
 763                         goto again;
 764                 }
 765                 if (head->extent_op && head->extent_op->update_flags)
 766                         extent_flags |= head->extent_op->flags_to_set;
 767                 else
 768                         BUG_ON(num_refs == 0);
 769
 770                 num_refs += head->node.ref_mod;
 771                 mutex_unlock(&head->mutex);
 772         }
 773         spin_unlock(&delayed_refs->lock);
 774 out:
 775         WARN_ON(num_refs == 0);
 776         if (refs)
 777                 *refs = num_refs;
 778         if (flags)
 779                 *flags = extent_flags;
 780 out_free:
 781         btrfs_free_path(path);
 782         return ret;
 783 }
 784
 785 /*
 786  * Back reference rules.  Back refs have three main goals:
 787  *
 788  * 1) differentiate between all holders of references to an extent so that
 789  *    when a reference is dropped we can make sure it was a valid reference
 790  *    before freeing the extent.
 791  *
 792  * 2) Provide enough information to quickly find the holders of an extent
 793  *    if we notice a given block is corrupted or bad.
 794  *
 795  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 796  *    maintenance.  This is actually the same as #2, but with a slightly
 797  *    different use case.
 798  *
 799  * There are two kinds of back refs. The implicit back refs is optimized
 800  * for pointers in non-shared tree blocks. For a given pointer in a block,
 801  * back refs of this kind provide information about the block's owner tree
 802  * and the pointer's key. These information allow us to find the block by
 803  * b-tree searching. The full back refs is for pointers in tree blocks not
 804  * referenced by their owner trees. The location of tree block is recorded
 805  * in the back refs. Actually the full back refs is generic, and can be
 806  * used in all cases the implicit back refs is used. The major shortcoming
 807  * of the full back refs is its overhead. Every time a tree block gets
 808  * COWed, we have to update back refs entry for all pointers in it.
 809  *
 810  * For a newly allocated tree block, we use implicit back refs for
 811  * pointers in it. This means most tree related operations only involve
 812  * implicit back refs. For a tree block created in old transaction, the
 813  * only way to drop a reference to it is COW it. So we can detect the
 814  * event that tree block loses its owner tree's reference and do the
 815  * back refs conversion.
 816  *
 817  * When a tree block is COW'd through a tree, there are four cases:
 818  *
 819  * The reference count of the block is one and the tree is the block's
 820  * owner tree. Nothing to do in this case.
 821  *
 822  * The reference count of the block is one and the tree is not the
 823  * block's owner tree. In this case, full back refs is used for pointers
 824  * in the block. Remove these full back refs, add implicit back refs for
 825  * every pointers in the new block.
 826  *
 827  * The reference count of the block is greater than one and the tree is
 828  * the block's owner tree. In this case, implicit back refs is used for
 829  * pointers in the block. Add full back refs for every pointers in the
 830  * block, increase lower level extents' reference counts. The original
 831  * implicit back refs are entailed to the new block.
 832  *
 833  * The reference count of the block is greater than one and the tree is
 834  * not the block's owner tree. Add implicit back refs for every pointer in
 835  * the new block, increase lower level extents' reference count.
 836  *
 837  * Back Reference Key composing:
 838  *
 839  * The key objectid corresponds to the first byte in the extent,
 840  * The key type is used to differentiate between types of back refs.
 841  * There are different meanings of the key offset for different types
 842  * of back refs.
 843  *
 844  * File extents can be referenced by:
 845  *
 846  * - multiple snapshots, subvolumes, or different generations in one subvol
 847  * - different files inside a single subvolume
 848  * - different offsets inside a file (bookend extents in file.c)
 849  *
 850  * The extent ref structure for the implicit back refs has fields for:
 851  *
 852  * - Objectid of the subvolume root
 853  * - objectid of the file holding the reference
 854  * - original offset in the file
 855  * - how many bookend extents
 856  *
 857  * The key offset for the implicit back refs is hash of the first
 858  * three fields.
 859  *
 860  * The extent ref structure for the full back refs has field for:
 861  *
 862  * - number of pointers in the tree leaf
 863  *
 864  * The key offset for the implicit back refs is the first byte of
 865  * the tree leaf
 866  *
 867  * When a file extent is allocated, The implicit back refs is used.
 868  * the fields are filled in:
 869  *
 870  *     (root_key.objectid, inode objectid, offset in file, 1)
 871  *
 872  * When a file extent is removed file truncation, we find the
 873  * corresponding implicit back refs and check the following fields:
 874  *
 875  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 876  *
 877  * Btree extents can be referenced by:
 878  *
 879  * - Different subvolumes
 880  *
 881  * Both the implicit back refs and the full back refs for tree blocks
 882  * only consist of key. The key offset for the implicit back refs is
 883  * objectid of block's owner tree. The key offset for the full back refs
 884  * is the first byte of parent block.
 885  *
 886  * When implicit back refs is used, information about the lowest key and
 887  * level of the tree block are required. These information are stored in
 888  * tree block info structure.
 889  */
 890
 891 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 892 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 893                                   struct btrfs_root *root,
 894                                   struct btrfs_path *path,
 895                                   u64 owner, u32 extra_size)
 896 {
 897         struct btrfs_extent_item *item;
 898         struct btrfs_extent_item_v0 *ei0;
 899         struct btrfs_extent_ref_v0 *ref0;
 900         struct btrfs_tree_block_info *bi;
 901         struct extent_buffer *leaf;
 902         struct btrfs_key key;
 903         struct btrfs_key found_key;
 904         u32 new_size = sizeof(*item);
 905         u64 refs;
 906         int ret;
 907
 908         leaf = path->nodes[0];
 909         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 910
 911         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 912         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 913                              struct btrfs_extent_item_v0);
 914         refs = btrfs_extent_refs_v0(leaf, ei0);
 915
 916         if (owner == (u64)-1) {
 917                 while (1) {
 918                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 919                                 ret = btrfs_next_leaf(root, path);
 920                                 if (ret < 0)
 921                                         return ret;
 922                                 BUG_ON(ret > 0);
 923                                 leaf = path->nodes[0];
 924                         }
 925                         btrfs_item_key_to_cpu(leaf, &found_key,
 926                                               path->slots[0]);
 927                         BUG_ON(key.objectid != found_key.objectid);
 928                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 929                                 path->slots[0]++;
 930                                 continue;
 931                         }
 932                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 933                                               struct btrfs_extent_ref_v0);
 934                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 935                         break;
 936                 }
 937         }
 938         btrfs_release_path(root, path);
 939
 940         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 941                 new_size += sizeof(*bi);
 942
 943         new_size -= sizeof(*ei0);
 944         ret = btrfs_search_slot(trans, root, &key, path,
 945                                 new_size + extra_size, 1);
 946         if (ret < 0)
 947                 return ret;
 948         BUG_ON(ret);
 949
 950         ret = btrfs_extend_item(trans, root, path, new_size);
 951         BUG_ON(ret);
 952
 953         leaf = path->nodes[0];
 954         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 955         btrfs_set_extent_refs(leaf, item, refs);
 956         /* FIXME: get real generation */
 957         btrfs_set_extent_generation(leaf, item, 0);
 958         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 959                 btrfs_set_extent_flags(leaf, item,
 960                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 961                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 962                 bi = (struct btrfs_tree_block_info *)(item + 1);
 963                 /* FIXME: get first key of the block */
 964                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 965                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 966         } else {
 967                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 968         }
 969         btrfs_mark_buffer_dirty(leaf);
 970         return 0;
 971 }
 972 #endif
 973
 974 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 975 {
 976         u32 high_crc = ~(u32)0;
 977         u32 low_crc = ~(u32)0;
 978         __le64 lenum;
 979
 980         lenum = cpu_to_le64(root_objectid);
 981         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 982         lenum = cpu_to_le64(owner);
 983         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 984         lenum = cpu_to_le64(offset);
 985         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 986
 987         return ((u64)high_crc << 31) ^ (u64)low_crc;
 988 }
 989
 990 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 991                                      struct btrfs_extent_data_ref *ref)
 992 {
 993         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 994                                     btrfs_extent_data_ref_objectid(leaf, ref),
 995                                     btrfs_extent_data_ref_offset(leaf, ref));
 996 }
 997
 998 static int match_extent_data_ref(struct extent_buffer *leaf,
 999                                  struct btrfs_extent_data_ref *ref,
1000                                  u64 root_objectid, u64 owner, u64 offset)
1001 {
1002         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1003             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1004             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1005                 return 0;
1006         return 1;
1007 }
1008
1009 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1010                                            struct btrfs_root *root,
1011                                            struct btrfs_path *path,
1012                                            u64 bytenr, u64 parent,
1013                                            u64 root_objectid,
1014                                            u64 owner, u64 offset)
1015 {
1016         struct btrfs_key key;
1017         struct btrfs_extent_data_ref *ref;
1018         struct extent_buffer *leaf;
1019         u32 nritems;
1020         int ret;
1021         int recow;
1022         int err = -ENOENT;
1023
1024         key.objectid = bytenr;
1025         if (parent) {
1026                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1027                 key.offset = parent;
1028         } else {
1029                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1030                 key.offset = hash_extent_data_ref(root_objectid,
1031                                                   owner, offset);
1032         }
1033 again:
1034         recow = 0;
1035         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1036         if (ret < 0) {
1037                 err = ret;
1038                 goto fail;
1039         }
1040
1041         if (parent) {
1042                 if (!ret)
1043                         return 0;
1044 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1045                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1046                 btrfs_release_path(root, path);
1047                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1048                 if (ret < 0) {
1049                         err = ret;
1050                         goto fail;
1051                 }
1052                 if (!ret)
1053                         return 0;
1054 #endif
1055                 goto fail;
1056         }
1057
1058         leaf = path->nodes[0];
1059         nritems = btrfs_header_nritems(leaf);
1060         while (1) {
1061                 if (path->slots[0] >= nritems) {
1062                         ret = btrfs_next_leaf(root, path);
1063                         if (ret < 0)
1064                                 err = ret;
1065                         if (ret)
1066                                 goto fail;
1067
1068                         leaf = path->nodes[0];
1069                         nritems = btrfs_header_nritems(leaf);
1070                         recow = 1;
1071                 }
1072
1073                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1074                 if (key.objectid != bytenr ||
1075                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1076                         goto fail;
1077
1078                 ref = btrfs_item_ptr(leaf, path->slots[0],
1079                                      struct btrfs_extent_data_ref);
1080
1081                 if (match_extent_data_ref(leaf, ref, root_objectid,
1082                                           owner, offset)) {
1083                         if (recow) {
1084                                 btrfs_release_path(root, path);
1085                                 goto again;
1086                         }
1087                         err = 0;
1088                         break;
1089                 }
1090                 path->slots[0]++;
1091         }
1092 fail:
1093         return err;
1094 }
1095
1096 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1097                                            struct btrfs_root *root,
1098                                            struct btrfs_path *path,
1099                                            u64 bytenr, u64 parent,
1100                                            u64 root_objectid, u64 owner,
1101                                            u64 offset, int refs_to_add)
1102 {
1103         struct btrfs_key key;
1104         struct extent_buffer *leaf;
1105         u32 size;
1106         u32 num_refs;
1107         int ret;
1108
1109         key.objectid = bytenr;
1110         if (parent) {
1111                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1112                 key.offset = parent;
1113                 size = sizeof(struct btrfs_shared_data_ref);
1114         } else {
1115                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1116                 key.offset = hash_extent_data_ref(root_objectid,
1117                                                   owner, offset);
1118                 size = sizeof(struct btrfs_extent_data_ref);
1119         }
1120
1121         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1122         if (ret && ret != -EEXIST)
1123                 goto fail;
1124
1125         leaf = path->nodes[0];
1126         if (parent) {
1127                 struct btrfs_shared_data_ref *ref;
1128                 ref = btrfs_item_ptr(leaf, path->slots[0],
1129                                      struct btrfs_shared_data_ref);
1130                 if (ret == 0) {
1131                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1132                 } else {
1133                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1134                         num_refs += refs_to_add;
1135                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1136                 }
1137         } else {
1138                 struct btrfs_extent_data_ref *ref;
1139                 while (ret == -EEXIST) {
1140                         ref = btrfs_item_ptr(leaf, path->slots[0],
1141                                              struct btrfs_extent_data_ref);
1142                         if (match_extent_data_ref(leaf, ref, root_objectid,
1143                                                   owner, offset))
1144                                 break;
1145                         btrfs_release_path(root, path);
1146                         key.offset++;
1147                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1148                                                       size);
1149                         if (ret && ret != -EEXIST)
1150                                 goto fail;
1151
1152                         leaf = path->nodes[0];
1153                 }
1154                 ref = btrfs_item_ptr(leaf, path->slots[0],
1155                                      struct btrfs_extent_data_ref);
1156                 if (ret == 0) {
1157                         btrfs_set_extent_data_ref_root(leaf, ref,
1158                                                        root_objectid);
1159                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1160                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1161                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1162                 } else {
1163                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1164                         num_refs += refs_to_add;
1165                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1166                 }
1167         }
1168         btrfs_mark_buffer_dirty(leaf);
1169         ret = 0;
1170 fail:
1171         btrfs_release_path(root, path);
1172         return ret;
1173 }
1174
1175 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1176                                            struct btrfs_root *root,
1177                                            struct btrfs_path *path,
1178                                            int refs_to_drop)
1179 {
1180         struct btrfs_key key;
1181         struct btrfs_extent_data_ref *ref1 = NULL;
1182         struct btrfs_shared_data_ref *ref2 = NULL;
1183         struct extent_buffer *leaf;
1184         u32 num_refs = 0;
1185         int ret = 0;
1186
1187         leaf = path->nodes[0];
1188         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1189
1190         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1191                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1192                                       struct btrfs_extent_data_ref);
1193                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1194         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1195                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1196                                       struct btrfs_shared_data_ref);
1197                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1198 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1199         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1200                 struct btrfs_extent_ref_v0 *ref0;
1201                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1202                                       struct btrfs_extent_ref_v0);
1203                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1204 #endif
1205         } else {
1206                 BUG();
1207         }
1208
1209         BUG_ON(num_refs < refs_to_drop);
1210         num_refs -= refs_to_drop;
1211
1212         if (num_refs == 0) {
1213                 ret = btrfs_del_item(trans, root, path);
1214         } else {
1215                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1216                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1217                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1218                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1219 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1220                 else {
1221                         struct btrfs_extent_ref_v0 *ref0;
1222                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1223                                         struct btrfs_extent_ref_v0);
1224                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1225                 }
1226 #endif
1227                 btrfs_mark_buffer_dirty(leaf);
1228         }
1229         return ret;
1230 }
1231
1232 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1233                                           struct btrfs_path *path,
1234                                           struct btrfs_extent_inline_ref *iref)
1235 {
1236         struct btrfs_key key;
1237         struct extent_buffer *leaf;
1238         struct btrfs_extent_data_ref *ref1;
1239         struct btrfs_shared_data_ref *ref2;
1240         u32 num_refs = 0;
1241
1242         leaf = path->nodes[0];
1243         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1244         if (iref) {
1245                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1246                     BTRFS_EXTENT_DATA_REF_KEY) {
1247                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1248                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1249                 } else {
1250                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1251                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1252                 }
1253         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1254                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1255                                       struct btrfs_extent_data_ref);
1256                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1257         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1258                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1259                                       struct btrfs_shared_data_ref);
1260                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1261 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1262         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1263                 struct btrfs_extent_ref_v0 *ref0;
1264                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1265                                       struct btrfs_extent_ref_v0);
1266                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1267 #endif
1268         } else {
1269                 WARN_ON(1);
1270         }
1271         return num_refs;
1272 }
1273
1274 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1275                                           struct btrfs_root *root,
1276                                           struct btrfs_path *path,
1277                                           u64 bytenr, u64 parent,
1278                                           u64 root_objectid)
1279 {
1280         struct btrfs_key key;
1281         int ret;
1282
1283         key.objectid = bytenr;
1284         if (parent) {
1285                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1286                 key.offset = parent;
1287         } else {
1288                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1289                 key.offset = root_objectid;
1290         }
1291
1292         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1293         if (ret > 0)
1294                 ret = -ENOENT;
1295 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1296         if (ret == -ENOENT && parent) {
1297                 btrfs_release_path(root, path);
1298                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1299                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1300                 if (ret > 0)
1301                         ret = -ENOENT;
1302         }
1303 #endif
1304         return ret;
1305 }
1306
1307 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1308                                           struct btrfs_root *root,
1309                                           struct btrfs_path *path,
1310                                           u64 bytenr, u64 parent,
1311                                           u64 root_objectid)
1312 {
1313         struct btrfs_key key;
1314         int ret;
1315
1316         key.objectid = bytenr;
1317         if (parent) {
1318                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1319                 key.offset = parent;
1320         } else {
1321                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1322                 key.offset = root_objectid;
1323         }
1324
1325         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1326         btrfs_release_path(root, path);
1327         return ret;
1328 }
1329
1330 static inline int extent_ref_type(u64 parent, u64 owner)
1331 {
1332         int type;
1333         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1334                 if (parent > 0)
1335                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1336                 else
1337                         type = BTRFS_TREE_BLOCK_REF_KEY;
1338         } else {
1339                 if (parent > 0)
1340                         type = BTRFS_SHARED_DATA_REF_KEY;
1341                 else
1342                         type = BTRFS_EXTENT_DATA_REF_KEY;
1343         }
1344         return type;
1345 }
1346
1347 static int find_next_key(struct btrfs_path *path, int level,
1348                          struct btrfs_key *key)
1349
1350 {
1351         for (; level < BTRFS_MAX_LEVEL; level++) {
1352                 if (!path->nodes[level])
1353                         break;
1354                 if (path->slots[level] + 1 >=
1355                     btrfs_header_nritems(path->nodes[level]))
1356                         continue;
1357                 if (level == 0)
1358                         btrfs_item_key_to_cpu(path->nodes[level], key,
1359                                               path->slots[level] + 1);
1360                 else
1361                         btrfs_node_key_to_cpu(path->nodes[level], key,
1362                                               path->slots[level] + 1);
1363                 return 0;
1364         }
1365         return 1;
1366 }
1367
1368 /*
1369  * look for inline back ref. if back ref is found, *ref_ret is set
1370  * to the address of inline back ref, and 0 is returned.
1371  *
1372  * if back ref isn't found, *ref_ret is set to the address where it
1373  * should be inserted, and -ENOENT is returned.
1374  *
1375  * if insert is true and there are too many inline back refs, the path
1376  * points to the extent item, and -EAGAIN is returned.
1377  *
1378  * NOTE: inline back refs are ordered in the same way that back ref
1379  *       items in the tree are ordered.
1380  */
1381 static noinline_for_stack
1382 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1383                                  struct btrfs_root *root,
1384                                  struct btrfs_path *path,
1385                                  struct btrfs_extent_inline_ref **ref_ret,
1386                                  u64 bytenr, u64 num_bytes,
1387                                  u64 parent, u64 root_objectid,
1388                                  u64 owner, u64 offset, int insert)
1389 {
1390         struct btrfs_key key;
1391         struct extent_buffer *leaf;
1392         struct btrfs_extent_item *ei;
1393         struct btrfs_extent_inline_ref *iref;
1394         u64 flags;
1395         u64 item_size;
1396         unsigned long ptr;
1397         unsigned long end;
1398         int extra_size;
1399         int type;
1400         int want;
1401         int ret;
1402         int err = 0;
1403
1404         key.objectid = bytenr;
1405         key.type = BTRFS_EXTENT_ITEM_KEY;
1406         key.offset = num_bytes;
1407
1408         want = extent_ref_type(parent, owner);
1409         if (insert) {
1410                 extra_size = btrfs_extent_inline_ref_size(want);
1411                 path->keep_locks = 1;
1412         } else
1413                 extra_size = -1;
1414         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1415         if (ret < 0) {
1416                 err = ret;
1417                 goto out;
1418         }
1419         BUG_ON(ret);
1420
1421         leaf = path->nodes[0];
1422         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1423 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1424         if (item_size < sizeof(*ei)) {
1425                 if (!insert) {
1426                         err = -ENOENT;
1427                         goto out;
1428                 }
1429                 ret = convert_extent_item_v0(trans, root, path, owner,
1430                                              extra_size);
1431                 if (ret < 0) {
1432                         err = ret;
1433                         goto out;
1434                 }
1435                 leaf = path->nodes[0];
1436                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1437         }
1438 #endif
1439         BUG_ON(item_size < sizeof(*ei));
1440
1441         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1442         flags = btrfs_extent_flags(leaf, ei);
1443
1444         ptr = (unsigned long)(ei + 1);
1445         end = (unsigned long)ei + item_size;
1446
1447         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1448                 ptr += sizeof(struct btrfs_tree_block_info);
1449                 BUG_ON(ptr > end);
1450         } else {
1451                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1452         }
1453
1454         err = -ENOENT;
1455         while (1) {
1456                 if (ptr >= end) {
1457                         WARN_ON(ptr > end);
1458                         break;
1459                 }
1460                 iref = (struct btrfs_extent_inline_ref *)ptr;
1461                 type = btrfs_extent_inline_ref_type(leaf, iref);
1462                 if (want < type)
1463                         break;
1464                 if (want > type) {
1465                         ptr += btrfs_extent_inline_ref_size(type);
1466                         continue;
1467                 }
1468
1469                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1470                         struct btrfs_extent_data_ref *dref;
1471                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1472                         if (match_extent_data_ref(leaf, dref, root_objectid,
1473                                                   owner, offset)) {
1474                                 err = 0;
1475                                 break;
1476                         }
1477                         if (hash_extent_data_ref_item(leaf, dref) <
1478                             hash_extent_data_ref(root_objectid, owner, offset))
1479                                 break;
1480                 } else {
1481                         u64 ref_offset;
1482                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1483                         if (parent > 0) {
1484                                 if (parent == ref_offset) {
1485                                         err = 0;
1486                                         break;
1487                                 }
1488                                 if (ref_offset < parent)
1489                                         break;
1490                         } else {
1491                                 if (root_objectid == ref_offset) {
1492                                         err = 0;
1493                                         break;
1494                                 }
1495                                 if (ref_offset < root_objectid)
1496                                         break;
1497                         }
1498                 }
1499                 ptr += btrfs_extent_inline_ref_size(type);
1500         }
1501         if (err == -ENOENT && insert) {
1502                 if (item_size + extra_size >=
1503                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1504                         err = -EAGAIN;
1505                         goto out;
1506                 }
1507                 /*
1508                  * To add new inline back ref, we have to make sure
1509                  * there is no corresponding back ref item.
1510                  * For simplicity, we just do not add new inline back
1511                  * ref if there is any kind of item for this block
1512                  */
1513                 if (find_next_key(path, 0, &key) == 0 &&
1514                     key.objectid == bytenr &&
1515                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1516                         err = -EAGAIN;
1517                         goto out;
1518                 }
1519         }
1520         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1521 out:
1522         if (insert) {
1523                 path->keep_locks = 0;
1524                 btrfs_unlock_up_safe(path, 1);
1525         }
1526         return err;
1527 }
1528
1529 /*
1530  * helper to add new inline back ref
1531  */
1532 static noinline_for_stack
1533 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1534                                 struct btrfs_root *root,
1535                                 struct btrfs_path *path,
1536                                 struct btrfs_extent_inline_ref *iref,
1537                                 u64 parent, u64 root_objectid,
1538                                 u64 owner, u64 offset, int refs_to_add,
1539                                 struct btrfs_delayed_extent_op *extent_op)
1540 {
1541         struct extent_buffer *leaf;
1542         struct btrfs_extent_item *ei;
1543         unsigned long ptr;
1544         unsigned long end;
1545         unsigned long item_offset;
1546         u64 refs;
1547         int size;
1548         int type;
1549         int ret;
1550
1551         leaf = path->nodes[0];
1552         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1553         item_offset = (unsigned long)iref - (unsigned long)ei;
1554
1555         type = extent_ref_type(parent, owner);
1556         size = btrfs_extent_inline_ref_size(type);
1557
1558         ret = btrfs_extend_item(trans, root, path, size);
1559         BUG_ON(ret);
1560
1561         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1562         refs = btrfs_extent_refs(leaf, ei);
1563         refs += refs_to_add;
1564         btrfs_set_extent_refs(leaf, ei, refs);
1565         if (extent_op)
1566                 __run_delayed_extent_op(extent_op, leaf, ei);
1567
1568         ptr = (unsigned long)ei + item_offset;
1569         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1570         if (ptr < end - size)
1571                 memmove_extent_buffer(leaf, ptr + size, ptr,
1572                                       end - size - ptr);
1573
1574         iref = (struct btrfs_extent_inline_ref *)ptr;
1575         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1576         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1577                 struct btrfs_extent_data_ref *dref;
1578                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1579                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1580                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1581                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1582                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1583         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1584                 struct btrfs_shared_data_ref *sref;
1585                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1586                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1587                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1588         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1589                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1590         } else {
1591                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1592         }
1593         btrfs_mark_buffer_dirty(leaf);
1594         return 0;
1595 }
1596
1597 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1598                                  struct btrfs_root *root,
1599                                  struct btrfs_path *path,
1600                                  struct btrfs_extent_inline_ref **ref_ret,
1601                                  u64 bytenr, u64 num_bytes, u64 parent,
1602                                  u64 root_objectid, u64 owner, u64 offset)
1603 {
1604         int ret;
1605
1606         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1607                                            bytenr, num_bytes, parent,
1608                                            root_objectid, owner, offset, 0);
1609         if (ret != -ENOENT)
1610                 return ret;
1611
1612         btrfs_release_path(root, path);
1613         *ref_ret = NULL;
1614
1615         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1616                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1617                                             root_objectid);
1618         } else {
1619                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1620                                              root_objectid, owner, offset);
1621         }
1622         return ret;
1623 }
1624
1625 /*
1626  * helper to update/remove inline back ref
1627  */
1628 static noinline_for_stack
1629 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1630                                  struct btrfs_root *root,
1631                                  struct btrfs_path *path,
1632                                  struct btrfs_extent_inline_ref *iref,
1633                                  int refs_to_mod,
1634                                  struct btrfs_delayed_extent_op *extent_op)
1635 {
1636         struct extent_buffer *leaf;
1637         struct btrfs_extent_item *ei;
1638         struct btrfs_extent_data_ref *dref = NULL;
1639         struct btrfs_shared_data_ref *sref = NULL;
1640         unsigned long ptr;
1641         unsigned long end;
1642         u32 item_size;
1643         int size;
1644         int type;
1645         int ret;
1646         u64 refs;
1647
1648         leaf = path->nodes[0];
1649         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1650         refs = btrfs_extent_refs(leaf, ei);
1651         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1652         refs += refs_to_mod;
1653         btrfs_set_extent_refs(leaf, ei, refs);
1654         if (extent_op)
1655                 __run_delayed_extent_op(extent_op, leaf, ei);
1656
1657         type = btrfs_extent_inline_ref_type(leaf, iref);
1658
1659         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1660                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1661                 refs = btrfs_extent_data_ref_count(leaf, dref);
1662         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1663                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1664                 refs = btrfs_shared_data_ref_count(leaf, sref);
1665         } else {
1666                 refs = 1;
1667                 BUG_ON(refs_to_mod != -1);
1668         }
1669
1670         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1671         refs += refs_to_mod;
1672
1673         if (refs > 0) {
1674                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1675                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1676                 else
1677                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1678         } else {
1679                 size =  btrfs_extent_inline_ref_size(type);
1680                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1681                 ptr = (unsigned long)iref;
1682                 end = (unsigned long)ei + item_size;
1683                 if (ptr + size < end)
1684                         memmove_extent_buffer(leaf, ptr, ptr + size,
1685                                               end - ptr - size);
1686                 item_size -= size;
1687                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1688                 BUG_ON(ret);
1689         }
1690         btrfs_mark_buffer_dirty(leaf);
1691         return 0;
1692 }
1693
1694 static noinline_for_stack
1695 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1696                                  struct btrfs_root *root,
1697                                  struct btrfs_path *path,
1698                                  u64 bytenr, u64 num_bytes, u64 parent,
1699                                  u64 root_objectid, u64 owner,
1700                                  u64 offset, int refs_to_add,
1701                                  struct btrfs_delayed_extent_op *extent_op)
1702 {
1703         struct btrfs_extent_inline_ref *iref;
1704         int ret;
1705
1706         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1707                                            bytenr, num_bytes, parent,
1708                                            root_objectid, owner, offset, 1);
1709         if (ret == 0) {
1710                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1711                 ret = update_inline_extent_backref(trans, root, path, iref,
1712                                                    refs_to_add, extent_op);
1713         } else if (ret == -ENOENT) {
1714                 ret = setup_inline_extent_backref(trans, root, path, iref,
1715                                                   parent, root_objectid,
1716                                                   owner, offset, refs_to_add,
1717                                                   extent_op);
1718         }
1719         return ret;
1720 }
1721
1722 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1723                                  struct btrfs_root *root,
1724                                  struct btrfs_path *path,
1725                                  u64 bytenr, u64 parent, u64 root_objectid,
1726                                  u64 owner, u64 offset, int refs_to_add)
1727 {
1728         int ret;
1729         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1730                 BUG_ON(refs_to_add != 1);
1731                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1732                                             parent, root_objectid);
1733         } else {
1734                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1735                                              parent, root_objectid,
1736                                              owner, offset, refs_to_add);
1737         }
1738         return ret;
1739 }
1740
1741 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1742                                  struct btrfs_root *root,
1743                                  struct btrfs_path *path,
1744                                  struct btrfs_extent_inline_ref *iref,
1745                                  int refs_to_drop, int is_data)
1746 {
1747         int ret;
1748
1749         BUG_ON(!is_data && refs_to_drop != 1);
1750         if (iref) {
1751                 ret = update_inline_extent_backref(trans, root, path, iref,
1752                                                    -refs_to_drop, NULL);
1753         } else if (is_data) {
1754                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1755         } else {
1756                 ret = btrfs_del_item(trans, root, path);
1757         }
1758         return ret;
1759 }
1760
1761 static int btrfs_issue_discard(struct block_device *bdev,
1762                                 u64 start, u64 len)
1763 {
1764         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1765 }
1766
1767 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1768                                 u64 num_bytes, u64 *actual_bytes)
1769 {
1770         int ret;
1771         u64 discarded_bytes = 0;
1772         struct btrfs_multi_bio *multi = NULL;
1773
1774
1775         /* Tell the block device(s) that the sectors can be discarded */
1776         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1777                               bytenr, &num_bytes, &multi, 0);
1778         if (!ret) {
1779                 struct btrfs_bio_stripe *stripe = multi->stripes;
1780                 int i;
1781
1782
1783                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1784                         ret = btrfs_issue_discard(stripe->dev->bdev,
1785                                                   stripe->physical,
1786                                                   stripe->length);
1787                         if (!ret)
1788                                 discarded_bytes += stripe->length;
1789                         else if (ret != -EOPNOTSUPP)
1790                                 break;
1791                 }
1792                 kfree(multi);
1793         }
1794         if (discarded_bytes && ret == -EOPNOTSUPP)
1795                 ret = 0;
1796
1797         if (actual_bytes)
1798                 *actual_bytes = discarded_bytes;
1799
1800
1801         return ret;
1802 }
1803
1804 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1805                          struct btrfs_root *root,
1806                          u64 bytenr, u64 num_bytes, u64 parent,
1807                          u64 root_objectid, u64 owner, u64 offset)
1808 {
1809         int ret;
1810         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1811                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1812
1813         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1814                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1815                                         parent, root_objectid, (int)owner,
1816                                         BTRFS_ADD_DELAYED_REF, NULL);
1817         } else {
1818                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1819                                         parent, root_objectid, owner, offset,
1820                                         BTRFS_ADD_DELAYED_REF, NULL);
1821         }
1822         return ret;
1823 }
1824
1825 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1826                                   struct btrfs_root *root,
1827                                   u64 bytenr, u64 num_bytes,
1828                                   u64 parent, u64 root_objectid,
1829                                   u64 owner, u64 offset, int refs_to_add,
1830                                   struct btrfs_delayed_extent_op *extent_op)
1831 {
1832         struct btrfs_path *path;
1833         struct extent_buffer *leaf;
1834         struct btrfs_extent_item *item;
1835         u64 refs;
1836         int ret;
1837         int err = 0;
1838
1839         path = btrfs_alloc_path();
1840         if (!path)
1841                 return -ENOMEM;
1842
1843         path->reada = 1;
1844         path->leave_spinning = 1;
1845         /* this will setup the path even if it fails to insert the back ref */
1846         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1847                                            path, bytenr, num_bytes, parent,
1848                                            root_objectid, owner, offset,
1849                                            refs_to_add, extent_op);
1850         if (ret == 0)
1851                 goto out;
1852
1853         if (ret != -EAGAIN) {
1854                 err = ret;
1855                 goto out;
1856         }
1857
1858         leaf = path->nodes[0];
1859         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1860         refs = btrfs_extent_refs(leaf, item);
1861         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1862         if (extent_op)
1863                 __run_delayed_extent_op(extent_op, leaf, item);
1864
1865         btrfs_mark_buffer_dirty(leaf);
1866         btrfs_release_path(root->fs_info->extent_root, path);
1867
1868         path->reada = 1;
1869         path->leave_spinning = 1;
1870
1871         /* now insert the actual backref */
1872         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1873                                     path, bytenr, parent, root_objectid,
1874                                     owner, offset, refs_to_add);
1875         BUG_ON(ret);
1876 out:
1877         btrfs_free_path(path);
1878         return err;
1879 }
1880
1881 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1882                                 struct btrfs_root *root,
1883                                 struct btrfs_delayed_ref_node *node,
1884                                 struct btrfs_delayed_extent_op *extent_op,
1885                                 int insert_reserved)
1886 {
1887         int ret = 0;
1888         struct btrfs_delayed_data_ref *ref;
1889         struct btrfs_key ins;
1890         u64 parent = 0;
1891         u64 ref_root = 0;
1892         u64 flags = 0;
1893
1894         ins.objectid = node->bytenr;
1895         ins.offset = node->num_bytes;
1896         ins.type = BTRFS_EXTENT_ITEM_KEY;
1897
1898         ref = btrfs_delayed_node_to_data_ref(node);
1899         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1900                 parent = ref->parent;
1901         else
1902                 ref_root = ref->root;
1903
1904         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1905                 if (extent_op) {
1906                         BUG_ON(extent_op->update_key);
1907                         flags |= extent_op->flags_to_set;
1908                 }
1909                 ret = alloc_reserved_file_extent(trans, root,
1910                                                  parent, ref_root, flags,
1911                                                  ref->objectid, ref->offset,
1912                                                  &ins, node->ref_mod);
1913         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1914                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1915                                              node->num_bytes, parent,
1916                                              ref_root, ref->objectid,
1917                                              ref->offset, node->ref_mod,
1918                                              extent_op);
1919         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1920                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1921                                           node->num_bytes, parent,
1922                                           ref_root, ref->objectid,
1923                                           ref->offset, node->ref_mod,
1924                                           extent_op);
1925         } else {
1926                 BUG();
1927         }
1928         return ret;
1929 }
1930
1931 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1932                                     struct extent_buffer *leaf,
1933                                     struct btrfs_extent_item *ei)
1934 {
1935         u64 flags = btrfs_extent_flags(leaf, ei);
1936         if (extent_op->update_flags) {
1937                 flags |= extent_op->flags_to_set;
1938                 btrfs_set_extent_flags(leaf, ei, flags);
1939         }
1940
1941         if (extent_op->update_key) {
1942                 struct btrfs_tree_block_info *bi;
1943                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1944                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1945                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1946         }
1947 }
1948
1949 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1950                                  struct btrfs_root *root,
1951                                  struct btrfs_delayed_ref_node *node,
1952                                  struct btrfs_delayed_extent_op *extent_op)
1953 {
1954         struct btrfs_key key;
1955         struct btrfs_path *path;
1956         struct btrfs_extent_item *ei;
1957         struct extent_buffer *leaf;
1958         u32 item_size;
1959         int ret;
1960         int err = 0;
1961
1962         path = btrfs_alloc_path();
1963         if (!path)
1964                 return -ENOMEM;
1965
1966         key.objectid = node->bytenr;
1967         key.type = BTRFS_EXTENT_ITEM_KEY;
1968         key.offset = node->num_bytes;
1969
1970         path->reada = 1;
1971         path->leave_spinning = 1;
1972         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1973                                 path, 0, 1);
1974         if (ret < 0) {
1975                 err = ret;
1976                 goto out;
1977         }
1978         if (ret > 0) {
1979                 err = -EIO;
1980                 goto out;
1981         }
1982
1983         leaf = path->nodes[0];
1984         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1985 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1986         if (item_size < sizeof(*ei)) {
1987                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1988                                              path, (u64)-1, 0);
1989                 if (ret < 0) {
1990                         err = ret;
1991                         goto out;
1992                 }
1993                 leaf = path->nodes[0];
1994                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1995         }
1996 #endif
1997         BUG_ON(item_size < sizeof(*ei));
1998         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1999         __run_delayed_extent_op(extent_op, leaf, ei);
2000
2001         btrfs_mark_buffer_dirty(leaf);
2002 out:
2003         btrfs_free_path(path);
2004         return err;
2005 }
2006
2007 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2008                                 struct btrfs_root *root,
2009                                 struct btrfs_delayed_ref_node *node,
2010                                 struct btrfs_delayed_extent_op *extent_op,
2011                                 int insert_reserved)
2012 {
2013         int ret = 0;
2014         struct btrfs_delayed_tree_ref *ref;
2015         struct btrfs_key ins;
2016         u64 parent = 0;
2017         u64 ref_root = 0;
2018
2019         ins.objectid = node->bytenr;
2020         ins.offset = node->num_bytes;
2021         ins.type = BTRFS_EXTENT_ITEM_KEY;
2022
2023         ref = btrfs_delayed_node_to_tree_ref(node);
2024         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2025                 parent = ref->parent;
2026         else
2027                 ref_root = ref->root;
2028
2029         BUG_ON(node->ref_mod != 1);
2030         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2031                 BUG_ON(!extent_op || !extent_op->update_flags ||
2032                        !extent_op->update_key);
2033                 ret = alloc_reserved_tree_block(trans, root,
2034                                                 parent, ref_root,
2035                                                 extent_op->flags_to_set,
2036                                                 &extent_op->key,
2037                                                 ref->level, &ins);
2038         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2039                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2040                                              node->num_bytes, parent, ref_root,
2041                                              ref->level, 0, 1, extent_op);
2042         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2043                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2044                                           node->num_bytes, parent, ref_root,
2045                                           ref->level, 0, 1, extent_op);
2046         } else {
2047                 BUG();
2048         }
2049         return ret;
2050 }
2051
2052 /* helper function to actually process a single delayed ref entry */
2053 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2054                                struct btrfs_root *root,
2055                                struct btrfs_delayed_ref_node *node,
2056                                struct btrfs_delayed_extent_op *extent_op,
2057                                int insert_reserved)
2058 {
2059         int ret;
2060         if (btrfs_delayed_ref_is_head(node)) {
2061                 struct btrfs_delayed_ref_head *head;
2062                 /*
2063                  * we've hit the end of the chain and we were supposed
2064                  * to insert this extent into the tree.  But, it got
2065                  * deleted before we ever needed to insert it, so all
2066                  * we have to do is clean up the accounting
2067                  */
2068                 BUG_ON(extent_op);
2069                 head = btrfs_delayed_node_to_head(node);
2070                 if (insert_reserved) {
2071                         btrfs_pin_extent(root, node->bytenr,
2072                                          node->num_bytes, 1);
2073                         if (head->is_data) {
2074                                 ret = btrfs_del_csums(trans, root,
2075                                                       node->bytenr,
2076                                                       node->num_bytes);
2077                                 BUG_ON(ret);
2078                         }
2079                 }
2080                 mutex_unlock(&head->mutex);
2081                 return 0;
2082         }
2083
2084         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2085             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2086                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2087                                            insert_reserved);
2088         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2089                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2090                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2091                                            insert_reserved);
2092         else
2093                 BUG();
2094         return ret;
2095 }
2096
2097 static noinline struct btrfs_delayed_ref_node *
2098 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2099 {
2100         struct rb_node *node;
2101         struct btrfs_delayed_ref_node *ref;
2102         int action = BTRFS_ADD_DELAYED_REF;
2103 again:
2104         /*
2105          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2106          * this prevents ref count from going down to zero when
2107          * there still are pending delayed ref.
2108          */
2109         node = rb_prev(&head->node.rb_node);
2110         while (1) {
2111                 if (!node)
2112                         break;
2113                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2114                                 rb_node);
2115                 if (ref->bytenr != head->node.bytenr)
2116                         break;
2117                 if (ref->action == action)
2118                         return ref;
2119                 node = rb_prev(node);
2120         }
2121         if (action == BTRFS_ADD_DELAYED_REF) {
2122                 action = BTRFS_DROP_DELAYED_REF;
2123                 goto again;
2124         }
2125         return NULL;
2126 }
2127
2128 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2129                                        struct btrfs_root *root,
2130                                        struct list_head *cluster)
2131 {
2132         struct btrfs_delayed_ref_root *delayed_refs;
2133         struct btrfs_delayed_ref_node *ref;
2134         struct btrfs_delayed_ref_head *locked_ref = NULL;
2135         struct btrfs_delayed_extent_op *extent_op;
2136         int ret;
2137         int count = 0;
2138         int must_insert_reserved = 0;
2139
2140         delayed_refs = &trans->transaction->delayed_refs;
2141         while (1) {
2142                 if (!locked_ref) {
2143                         /* pick a new head ref from the cluster list */
2144                         if (list_empty(cluster))
2145                                 break;
2146
2147                         locked_ref = list_entry(cluster->next,
2148                                      struct btrfs_delayed_ref_head, cluster);
2149
2150                         /* grab the lock that says we are going to process
2151                          * all the refs for this head */
2152                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2153
2154                         /*
2155                          * we may have dropped the spin lock to get the head
2156                          * mutex lock, and that might have given someone else
2157                          * time to free the head.  If that's true, it has been
2158                          * removed from our list and we can move on.
2159                          */
2160                         if (ret == -EAGAIN) {
2161                                 locked_ref = NULL;
2162                                 count++;
2163                                 continue;
2164                         }
2165                 }
2166
2167                 /*
2168                  * record the must insert reserved flag before we
2169                  * drop the spin lock.
2170                  */
2171                 must_insert_reserved = locked_ref->must_insert_reserved;
2172                 locked_ref->must_insert_reserved = 0;
2173
2174                 extent_op = locked_ref->extent_op;
2175                 locked_ref->extent_op = NULL;
2176
2177                 /*
2178                  * locked_ref is the head node, so we have to go one
2179                  * node back for any delayed ref updates
2180                  */
2181                 ref = select_delayed_ref(locked_ref);
2182                 if (!ref) {
2183                         /* All delayed refs have been processed, Go ahead
2184                          * and send the head node to run_one_delayed_ref,
2185                          * so that any accounting fixes can happen
2186                          */
2187                         ref = &locked_ref->node;
2188
2189                         if (extent_op && must_insert_reserved) {
2190                                 kfree(extent_op);
2191                                 extent_op = NULL;
2192                         }
2193
2194                         if (extent_op) {
2195                                 spin_unlock(&delayed_refs->lock);
2196
2197                                 ret = run_delayed_extent_op(trans, root,
2198                                                             ref, extent_op);
2199                                 BUG_ON(ret);
2200                                 kfree(extent_op);
2201
2202                                 cond_resched();
2203                                 spin_lock(&delayed_refs->lock);
2204                                 continue;
2205                         }
2206
2207                         list_del_init(&locked_ref->cluster);
2208                         locked_ref = NULL;
2209                 }
2210
2211                 ref->in_tree = 0;
2212                 rb_erase(&ref->rb_node, &delayed_refs->root);
2213                 delayed_refs->num_entries--;
2214
2215                 spin_unlock(&delayed_refs->lock);
2216
2217                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2218                                           must_insert_reserved);
2219                 BUG_ON(ret);
2220
2221                 btrfs_put_delayed_ref(ref);
2222                 kfree(extent_op);
2223                 count++;
2224
2225                 cond_resched();
2226                 spin_lock(&delayed_refs->lock);
2227         }
2228         return count;
2229 }
2230
2231 /*
2232  * this starts processing the delayed reference count updates and
2233  * extent insertions we have queued up so far.  count can be
2234  * 0, which means to process everything in the tree at the start
2235  * of the run (but not newly added entries), or it can be some target
2236  * number you'd like to process.
2237  */
2238 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2239                            struct btrfs_root *root, unsigned long count)
2240 {
2241         struct rb_node *node;
2242         struct btrfs_delayed_ref_root *delayed_refs;
2243         struct btrfs_delayed_ref_node *ref;
2244         struct list_head cluster;
2245         int ret;
2246         int run_all = count == (unsigned long)-1;
2247         int run_most = 0;
2248
2249         if (root == root->fs_info->extent_root)
2250                 root = root->fs_info->tree_root;
2251
2252         delayed_refs = &trans->transaction->delayed_refs;
2253         INIT_LIST_HEAD(&cluster);
2254 again:
2255         spin_lock(&delayed_refs->lock);
2256         if (count == 0) {
2257                 count = delayed_refs->num_entries * 2;
2258                 run_most = 1;
2259         }
2260         while (1) {
2261                 if (!(run_all || run_most) &&
2262                     delayed_refs->num_heads_ready < 64)
2263                         break;
2264
2265                 /*
2266                  * go find something we can process in the rbtree.  We start at
2267                  * the beginning of the tree, and then build a cluster
2268                  * of refs to process starting at the first one we are able to
2269                  * lock
2270                  */
2271                 ret = btrfs_find_ref_cluster(trans, &cluster,
2272                                              delayed_refs->run_delayed_start);
2273                 if (ret)
2274                         break;
2275
2276                 ret = run_clustered_refs(trans, root, &cluster);
2277                 BUG_ON(ret < 0);
2278
2279                 count -= min_t(unsigned long, ret, count);
2280
2281                 if (count == 0)
2282                         break;
2283         }
2284
2285         if (run_all) {
2286                 node = rb_first(&delayed_refs->root);
2287                 if (!node)
2288                         goto out;
2289                 count = (unsigned long)-1;
2290
2291                 while (node) {
2292                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2293                                        rb_node);
2294                         if (btrfs_delayed_ref_is_head(ref)) {
2295                                 struct btrfs_delayed_ref_head *head;
2296
2297                                 head = btrfs_delayed_node_to_head(ref);
2298                                 atomic_inc(&ref->refs);
2299
2300                                 spin_unlock(&delayed_refs->lock);
2301                                 mutex_lock(&head->mutex);
2302                                 mutex_unlock(&head->mutex);
2303
2304                                 btrfs_put_delayed_ref(ref);
2305                                 cond_resched();
2306                                 goto again;
2307                         }
2308                         node = rb_next(node);
2309                 }
2310                 spin_unlock(&delayed_refs->lock);
2311                 schedule_timeout(1);
2312                 goto again;
2313         }
2314 out:
2315         spin_unlock(&delayed_refs->lock);
2316         return 0;
2317 }
2318
2319 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2320                                 struct btrfs_root *root,
2321                                 u64 bytenr, u64 num_bytes, u64 flags,
2322                                 int is_data)
2323 {
2324         struct btrfs_delayed_extent_op *extent_op;
2325         int ret;
2326
2327         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2328         if (!extent_op)
2329                 return -ENOMEM;
2330
2331         extent_op->flags_to_set = flags;
2332         extent_op->update_flags = 1;
2333         extent_op->update_key = 0;
2334         extent_op->is_data = is_data ? 1 : 0;
2335
2336         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2337         if (ret)
2338                 kfree(extent_op);
2339         return ret;
2340 }
2341
2342 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2343                                       struct btrfs_root *root,
2344                                       struct btrfs_path *path,
2345                                       u64 objectid, u64 offset, u64 bytenr)
2346 {
2347         struct btrfs_delayed_ref_head *head;
2348         struct btrfs_delayed_ref_node *ref;
2349         struct btrfs_delayed_data_ref *data_ref;
2350         struct btrfs_delayed_ref_root *delayed_refs;
2351         struct rb_node *node;
2352         int ret = 0;
2353
2354         ret = -ENOENT;
2355         delayed_refs = &trans->transaction->delayed_refs;
2356         spin_lock(&delayed_refs->lock);
2357         head = btrfs_find_delayed_ref_head(trans, bytenr);
2358         if (!head)
2359                 goto out;
2360
2361         if (!mutex_trylock(&head->mutex)) {
2362                 atomic_inc(&head->node.refs);
2363                 spin_unlock(&delayed_refs->lock);
2364
2365                 btrfs_release_path(root->fs_info->extent_root, path);
2366
2367                 mutex_lock(&head->mutex);
2368                 mutex_unlock(&head->mutex);
2369                 btrfs_put_delayed_ref(&head->node);
2370                 return -EAGAIN;
2371         }
2372
2373         node = rb_prev(&head->node.rb_node);
2374         if (!node)
2375                 goto out_unlock;
2376
2377         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2378
2379         if (ref->bytenr != bytenr)
2380                 goto out_unlock;
2381
2382         ret = 1;
2383         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2384                 goto out_unlock;
2385
2386         data_ref = btrfs_delayed_node_to_data_ref(ref);
2387
2388         node = rb_prev(node);
2389         if (node) {
2390                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2391                 if (ref->bytenr == bytenr)
2392                         goto out_unlock;
2393         }
2394
2395         if (data_ref->root != root->root_key.objectid ||
2396             data_ref->objectid != objectid || data_ref->offset != offset)
2397                 goto out_unlock;
2398
2399         ret = 0;
2400 out_unlock:
2401         mutex_unlock(&head->mutex);
2402 out:
2403         spin_unlock(&delayed_refs->lock);
2404         return ret;
2405 }
2406
2407 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2408                                         struct btrfs_root *root,
2409                                         struct btrfs_path *path,
2410                                         u64 objectid, u64 offset, u64 bytenr)
2411 {
2412         struct btrfs_root *extent_root = root->fs_info->extent_root;
2413         struct extent_buffer *leaf;
2414         struct btrfs_extent_data_ref *ref;
2415         struct btrfs_extent_inline_ref *iref;
2416         struct btrfs_extent_item *ei;
2417         struct btrfs_key key;
2418         u32 item_size;
2419         int ret;
2420
2421         key.objectid = bytenr;
2422         key.offset = (u64)-1;
2423         key.type = BTRFS_EXTENT_ITEM_KEY;
2424
2425         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2426         if (ret < 0)
2427                 goto out;
2428         BUG_ON(ret == 0);
2429
2430         ret = -ENOENT;
2431         if (path->slots[0] == 0)
2432                 goto out;
2433
2434         path->slots[0]--;
2435         leaf = path->nodes[0];
2436         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2437
2438         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2439                 goto out;
2440
2441         ret = 1;
2442         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2443 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2444         if (item_size < sizeof(*ei)) {
2445                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2446                 goto out;
2447         }
2448 #endif
2449         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2450
2451         if (item_size != sizeof(*ei) +
2452             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2453                 goto out;
2454
2455         if (btrfs_extent_generation(leaf, ei) <=
2456             btrfs_root_last_snapshot(&root->root_item))
2457                 goto out;
2458
2459         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2460         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2461             BTRFS_EXTENT_DATA_REF_KEY)
2462                 goto out;
2463
2464         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2465         if (btrfs_extent_refs(leaf, ei) !=
2466             btrfs_extent_data_ref_count(leaf, ref) ||
2467             btrfs_extent_data_ref_root(leaf, ref) !=
2468             root->root_key.objectid ||
2469             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2470             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2471                 goto out;
2472
2473         ret = 0;
2474 out:
2475         return ret;
2476 }
2477
2478 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2479                           struct btrfs_root *root,
2480                           u64 objectid, u64 offset, u64 bytenr)
2481 {
2482         struct btrfs_path *path;
2483         int ret;
2484         int ret2;
2485
2486         path = btrfs_alloc_path();
2487         if (!path)
2488                 return -ENOENT;
2489
2490         do {
2491                 ret = check_committed_ref(trans, root, path, objectid,
2492                                           offset, bytenr);
2493                 if (ret && ret != -ENOENT)
2494                         goto out;
2495
2496                 ret2 = check_delayed_ref(trans, root, path, objectid,
2497                                          offset, bytenr);
2498         } while (ret2 == -EAGAIN);
2499
2500         if (ret2 && ret2 != -ENOENT) {
2501                 ret = ret2;
2502                 goto out;
2503         }
2504
2505         if (ret != -ENOENT || ret2 != -ENOENT)
2506                 ret = 0;
2507 out:
2508         btrfs_free_path(path);
2509         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2510                 WARN_ON(ret > 0);
2511         return ret;
2512 }
2513
2514 #if 0
2515 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2516                     struct extent_buffer *buf, u32 nr_extents)
2517 {
2518         struct btrfs_key key;
2519         struct btrfs_file_extent_item *fi;
2520         u64 root_gen;
2521         u32 nritems;
2522         int i;
2523         int level;
2524         int ret = 0;
2525         int shared = 0;
2526
2527         if (!root->ref_cows)
2528                 return 0;
2529
2530         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2531                 shared = 0;
2532                 root_gen = root->root_key.offset;
2533         } else {
2534                 shared = 1;
2535                 root_gen = trans->transid - 1;
2536         }
2537
2538         level = btrfs_header_level(buf);
2539         nritems = btrfs_header_nritems(buf);
2540
2541         if (level == 0) {
2542                 struct btrfs_leaf_ref *ref;
2543                 struct btrfs_extent_info *info;
2544
2545                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2546                 if (!ref) {
2547                         ret = -ENOMEM;
2548                         goto out;
2549                 }
2550
2551                 ref->root_gen = root_gen;
2552                 ref->bytenr = buf->start;
2553                 ref->owner = btrfs_header_owner(buf);
2554                 ref->generation = btrfs_header_generation(buf);
2555                 ref->nritems = nr_extents;
2556                 info = ref->extents;
2557
2558                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2559                         u64 disk_bytenr;
2560                         btrfs_item_key_to_cpu(buf, &key, i);
2561                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2562                                 continue;
2563                         fi = btrfs_item_ptr(buf, i,
2564                                             struct btrfs_file_extent_item);
2565                         if (btrfs_file_extent_type(buf, fi) ==
2566                             BTRFS_FILE_EXTENT_INLINE)
2567                                 continue;
2568                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2569                         if (disk_bytenr == 0)
2570                                 continue;
2571
2572                         info->bytenr = disk_bytenr;
2573                         info->num_bytes =
2574                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2575                         info->objectid = key.objectid;
2576                         info->offset = key.offset;
2577                         info++;
2578                 }
2579
2580                 ret = btrfs_add_leaf_ref(root, ref, shared);
2581                 if (ret == -EEXIST && shared) {
2582                         struct btrfs_leaf_ref *old;
2583                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2584                         BUG_ON(!old);
2585                         btrfs_remove_leaf_ref(root, old);
2586                         btrfs_free_leaf_ref(root, old);
2587                         ret = btrfs_add_leaf_ref(root, ref, shared);
2588                 }
2589                 WARN_ON(ret);
2590                 btrfs_free_leaf_ref(root, ref);
2591         }
2592 out:
2593         return ret;
2594 }
2595
2596 /* when a block goes through cow, we update the reference counts of
2597  * everything that block points to.  The internal pointers of the block
2598  * can be in just about any order, and it is likely to have clusters of
2599  * things that are close together and clusters of things that are not.
2600  *
2601  * To help reduce the seeks that come with updating all of these reference
2602  * counts, sort them by byte number before actual updates are done.
2603  *
2604  * struct refsort is used to match byte number to slot in the btree block.
2605  * we sort based on the byte number and then use the slot to actually
2606  * find the item.
2607  *
2608  * struct refsort is smaller than strcut btrfs_item and smaller than
2609  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2610  * for a btree block, there's no way for a kmalloc of refsorts for a
2611  * single node to be bigger than a page.
2612  */
2613 struct refsort {
2614         u64 bytenr;
2615         u32 slot;
2616 };
2617
2618 /*
2619  * for passing into sort()
2620  */
2621 static int refsort_cmp(const void *a_void, const void *b_void)
2622 {
2623         const struct refsort *a = a_void;
2624         const struct refsort *b = b_void;
2625
2626         if (a->bytenr < b->bytenr)
2627                 return -1;
2628         if (a->bytenr > b->bytenr)
2629                 return 1;
2630         return 0;
2631 }
2632 #endif
2633
2634 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2635                            struct btrfs_root *root,
2636                            struct extent_buffer *buf,
2637                            int full_backref, int inc)
2638 {
2639         u64 bytenr;
2640         u64 num_bytes;
2641         u64 parent;
2642         u64 ref_root;
2643         u32 nritems;
2644         struct btrfs_key key;
2645         struct btrfs_file_extent_item *fi;
2646         int i;
2647         int level;
2648         int ret = 0;
2649         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2650                             u64, u64, u64, u64, u64, u64);
2651
2652         ref_root = btrfs_header_owner(buf);
2653         nritems = btrfs_header_nritems(buf);
2654         level = btrfs_header_level(buf);
2655
2656         if (!root->ref_cows && level == 0)
2657                 return 0;
2658
2659         if (inc)
2660                 process_func = btrfs_inc_extent_ref;
2661         else
2662                 process_func = btrfs_free_extent;
2663
2664         if (full_backref)
2665                 parent = buf->start;
2666         else
2667                 parent = 0;
2668
2669         for (i = 0; i < nritems; i++) {
2670                 if (level == 0) {
2671                         btrfs_item_key_to_cpu(buf, &key, i);
2672                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2673                                 continue;
2674                         fi = btrfs_item_ptr(buf, i,
2675                                             struct btrfs_file_extent_item);
2676                         if (btrfs_file_extent_type(buf, fi) ==
2677                             BTRFS_FILE_EXTENT_INLINE)
2678                                 continue;
2679                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2680                         if (bytenr == 0)
2681                                 continue;
2682
2683                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2684                         key.offset -= btrfs_file_extent_offset(buf, fi);
2685                         ret = process_func(trans, root, bytenr, num_bytes,
2686                                            parent, ref_root, key.objectid,
2687                                            key.offset);
2688                         if (ret)
2689                                 goto fail;
2690                 } else {
2691                         bytenr = btrfs_node_blockptr(buf, i);
2692                         num_bytes = btrfs_level_size(root, level - 1);
2693                         ret = process_func(trans, root, bytenr, num_bytes,
2694                                            parent, ref_root, level - 1, 0);
2695                         if (ret)
2696                                 goto fail;
2697                 }
2698         }
2699         return 0;
2700 fail:
2701         BUG();
2702         return ret;
2703 }
2704
2705 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2706                   struct extent_buffer *buf, int full_backref)
2707 {
2708         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2709 }
2710
2711 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2712                   struct extent_buffer *buf, int full_backref)
2713 {
2714         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2715 }
2716
2717 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2718                                  struct btrfs_root *root,
2719                                  struct btrfs_path *path,
2720                                  struct btrfs_block_group_cache *cache)
2721 {
2722         int ret;
2723         struct btrfs_root *extent_root = root->fs_info->extent_root;
2724         unsigned long bi;
2725         struct extent_buffer *leaf;
2726
2727         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2728         if (ret < 0)
2729                 goto fail;
2730         BUG_ON(ret);
2731
2732         leaf = path->nodes[0];
2733         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2734         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2735         btrfs_mark_buffer_dirty(leaf);
2736         btrfs_release_path(extent_root, path);
2737 fail:
2738         if (ret)
2739                 return ret;
2740         return 0;
2741
2742 }
2743
2744 static struct btrfs_block_group_cache *
2745 next_block_group(struct btrfs_root *root,
2746                  struct btrfs_block_group_cache *cache)
2747 {
2748         struct rb_node *node;
2749         spin_lock(&root->fs_info->block_group_cache_lock);
2750         node = rb_next(&cache->cache_node);
2751         btrfs_put_block_group(cache);
2752         if (node) {
2753                 cache = rb_entry(node, struct btrfs_block_group_cache,
2754                                  cache_node);
2755                 btrfs_get_block_group(cache);
2756         } else
2757                 cache = NULL;
2758         spin_unlock(&root->fs_info->block_group_cache_lock);
2759         return cache;
2760 }
2761
2762 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2763                             struct btrfs_trans_handle *trans,
2764                             struct btrfs_path *path)
2765 {
2766         struct btrfs_root *root = block_group->fs_info->tree_root;
2767         struct inode *inode = NULL;
2768         u64 alloc_hint = 0;
2769         int dcs = BTRFS_DC_ERROR;
2770         int num_pages = 0;
2771         int retries = 0;
2772         int ret = 0;
2773
2774         /*
2775          * If this block group is smaller than 100 megs don't bother caching the
2776          * block group.
2777          */
2778         if (block_group->key.offset < (100 * 1024 * 1024)) {
2779                 spin_lock(&block_group->lock);
2780                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2781                 spin_unlock(&block_group->lock);
2782                 return 0;
2783         }
2784
2785 again:
2786         inode = lookup_free_space_inode(root, block_group, path);
2787         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2788                 ret = PTR_ERR(inode);
2789                 btrfs_release_path(root, path);
2790                 goto out;
2791         }
2792
2793         if (IS_ERR(inode)) {
2794                 BUG_ON(retries);
2795                 retries++;
2796
2797                 if (block_group->ro)
2798                         goto out_free;
2799
2800                 ret = create_free_space_inode(root, trans, block_group, path);
2801                 if (ret)
2802                         goto out_free;
2803                 goto again;
2804         }
2805
2806         /*
2807          * We want to set the generation to 0, that way if anything goes wrong
2808          * from here on out we know not to trust this cache when we load up next
2809          * time.
2810          */
2811         BTRFS_I(inode)->generation = 0;
2812         ret = btrfs_update_inode(trans, root, inode);
2813         WARN_ON(ret);
2814
2815         if (i_size_read(inode) > 0) {
2816                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2817                                                       inode);
2818                 if (ret)
2819                         goto out_put;
2820         }
2821
2822         spin_lock(&block_group->lock);
2823         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2824                 /* We're not cached, don't bother trying to write stuff out */
2825                 dcs = BTRFS_DC_WRITTEN;
2826                 spin_unlock(&block_group->lock);
2827                 goto out_put;
2828         }
2829         spin_unlock(&block_group->lock);
2830
2831         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2832         if (!num_pages)
2833                 num_pages = 1;
2834
2835         /*
2836          * Just to make absolutely sure we have enough space, we're going to
2837          * preallocate 12 pages worth of space for each block group.  In
2838          * practice we ought to use at most 8, but we need extra space so we can
2839          * add our header and have a terminator between the extents and the
2840          * bitmaps.
2841          */
2842         num_pages *= 16;
2843         num_pages *= PAGE_CACHE_SIZE;
2844
2845         ret = btrfs_check_data_free_space(inode, num_pages);
2846         if (ret)
2847                 goto out_put;
2848
2849         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2850                                               num_pages, num_pages,
2851                                               &alloc_hint);
2852         if (!ret)
2853                 dcs = BTRFS_DC_SETUP;
2854         btrfs_free_reserved_data_space(inode, num_pages);
2855 out_put:
2856         iput(inode);
2857 out_free:
2858         btrfs_release_path(root, path);
2859 out:
2860         spin_lock(&block_group->lock);
2861         block_group->disk_cache_state = dcs;
2862         spin_unlock(&block_group->lock);
2863
2864         return ret;
2865 }
2866
2867 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2868                                    struct btrfs_root *root)
2869 {
2870         struct btrfs_block_group_cache *cache;
2871         int err = 0;
2872         struct btrfs_path *path;
2873         u64 last = 0;
2874
2875         path = btrfs_alloc_path();
2876         if (!path)
2877                 return -ENOMEM;
2878
2879 again:
2880         while (1) {
2881                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2882                 while (cache) {
2883                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2884                                 break;
2885                         cache = next_block_group(root, cache);
2886                 }
2887                 if (!cache) {
2888                         if (last == 0)
2889                                 break;
2890                         last = 0;
2891                         continue;
2892                 }
2893                 err = cache_save_setup(cache, trans, path);
2894                 last = cache->key.objectid + cache->key.offset;
2895                 btrfs_put_block_group(cache);
2896         }
2897
2898         while (1) {
2899                 if (last == 0) {
2900                         err = btrfs_run_delayed_refs(trans, root,
2901                                                      (unsigned long)-1);
2902                         BUG_ON(err);
2903                 }
2904
2905                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2906                 while (cache) {
2907                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2908                                 btrfs_put_block_group(cache);
2909                                 goto again;
2910                         }
2911
2912                         if (cache->dirty)
2913                                 break;
2914                         cache = next_block_group(root, cache);
2915                 }
2916                 if (!cache) {
2917                         if (last == 0)
2918                                 break;
2919                         last = 0;
2920                         continue;
2921                 }
2922
2923                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2924                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2925                 cache->dirty = 0;
2926                 last = cache->key.objectid + cache->key.offset;
2927
2928                 err = write_one_cache_group(trans, root, path, cache);
2929                 BUG_ON(err);
2930                 btrfs_put_block_group(cache);
2931         }
2932
2933         while (1) {
2934                 /*
2935                  * I don't think this is needed since we're just marking our
2936                  * preallocated extent as written, but just in case it can't
2937                  * hurt.
2938                  */
2939                 if (last == 0) {
2940                         err = btrfs_run_delayed_refs(trans, root,
2941                                                      (unsigned long)-1);
2942                         BUG_ON(err);
2943                 }
2944
2945                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2946                 while (cache) {
2947                         /*
2948                          * Really this shouldn't happen, but it could if we
2949                          * couldn't write the entire preallocated extent and
2950                          * splitting the extent resulted in a new block.
2951                          */
2952                         if (cache->dirty) {
2953                                 btrfs_put_block_group(cache);
2954                                 goto again;
2955                         }
2956                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2957                                 break;
2958                         cache = next_block_group(root, cache);
2959                 }
2960                 if (!cache) {
2961                         if (last == 0)
2962                                 break;
2963                         last = 0;
2964                         continue;
2965                 }
2966
2967                 btrfs_write_out_cache(root, trans, cache, path);
2968
2969                 /*
2970                  * If we didn't have an error then the cache state is still
2971                  * NEED_WRITE, so we can set it to WRITTEN.
2972                  */
2973                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2974                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2975                 last = cache->key.objectid + cache->key.offset;
2976                 btrfs_put_block_group(cache);
2977         }
2978
2979         btrfs_free_path(path);
2980         return 0;
2981 }
2982
2983 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2984 {
2985         struct btrfs_block_group_cache *block_group;
2986         int readonly = 0;
2987
2988         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2989         if (!block_group || block_group->ro)
2990                 readonly = 1;
2991         if (block_group)
2992                 btrfs_put_block_group(block_group);
2993         return readonly;
2994 }
2995
2996 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2997                              u64 total_bytes, u64 bytes_used,
2998                              struct btrfs_space_info **space_info)
2999 {
3000         struct btrfs_space_info *found;
3001         int i;
3002         int factor;
3003
3004         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3005                      BTRFS_BLOCK_GROUP_RAID10))
3006                 factor = 2;
3007         else
3008                 factor = 1;
3009
3010         found = __find_space_info(info, flags);
3011         if (found) {
3012                 spin_lock(&found->lock);
3013                 found->total_bytes += total_bytes;
3014                 found->disk_total += total_bytes * factor;
3015                 found->bytes_used += bytes_used;
3016                 found->disk_used += bytes_used * factor;
3017                 found->full = 0;
3018                 spin_unlock(&found->lock);
3019                 *space_info = found;
3020                 return 0;
3021         }
3022         found = kzalloc(sizeof(*found), GFP_NOFS);
3023         if (!found)
3024                 return -ENOMEM;
3025
3026         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3027                 INIT_LIST_HEAD(&found->block_groups[i]);
3028         init_rwsem(&found->groups_sem);
3029         spin_lock_init(&found->lock);
3030         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3031                                 BTRFS_BLOCK_GROUP_SYSTEM |
3032                                 BTRFS_BLOCK_GROUP_METADATA);
3033         found->total_bytes = total_bytes;
3034         found->disk_total = total_bytes * factor;
3035         found->bytes_used = bytes_used;
3036         found->disk_used = bytes_used * factor;
3037         found->bytes_pinned = 0;
3038         found->bytes_reserved = 0;
3039         found->bytes_readonly = 0;
3040         found->bytes_may_use = 0;
3041         found->full = 0;
3042         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3043         found->chunk_alloc = 0;
3044         *space_info = found;
3045         list_add_rcu(&found->list, &info->space_info);
3046         atomic_set(&found->caching_threads, 0);
3047         return 0;
3048 }
3049
3050 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3051 {
3052         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3053                                    BTRFS_BLOCK_GROUP_RAID1 |
3054                                    BTRFS_BLOCK_GROUP_RAID10 |
3055                                    BTRFS_BLOCK_GROUP_DUP);
3056         if (extra_flags) {
3057                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3058                         fs_info->avail_data_alloc_bits |= extra_flags;
3059                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3060                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3061                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3062                         fs_info->avail_system_alloc_bits |= extra_flags;
3063         }
3064 }
3065
3066 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3067 {
3068         /*
3069          * we add in the count of missing devices because we want
3070          * to make sure that any RAID levels on a degraded FS
3071          * continue to be honored.
3072          */
3073         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3074                 root->fs_info->fs_devices->missing_devices;
3075
3076         if (num_devices == 1)
3077                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3078         if (num_devices < 4)
3079                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3080
3081         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3082             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3083                       BTRFS_BLOCK_GROUP_RAID10))) {
3084                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3085         }
3086
3087         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3088             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3089                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3090         }
3091
3092         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3093             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3094              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3095              (flags & BTRFS_BLOCK_GROUP_DUP)))
3096                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3097         return flags;
3098 }
3099
3100 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3101 {
3102         if (flags & BTRFS_BLOCK_GROUP_DATA)
3103                 flags |= root->fs_info->avail_data_alloc_bits &
3104                          root->fs_info->data_alloc_profile;
3105         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3106                 flags |= root->fs_info->avail_system_alloc_bits &
3107                          root->fs_info->system_alloc_profile;
3108         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3109                 flags |= root->fs_info->avail_metadata_alloc_bits &
3110                          root->fs_info->metadata_alloc_profile;
3111         return btrfs_reduce_alloc_profile(root, flags);
3112 }
3113
3114 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3115 {
3116         u64 flags;
3117
3118         if (data)
3119                 flags = BTRFS_BLOCK_GROUP_DATA;
3120         else if (root == root->fs_info->chunk_root)
3121                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3122         else
3123                 flags = BTRFS_BLOCK_GROUP_METADATA;
3124
3125         return get_alloc_profile(root, flags);
3126 }
3127
3128 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3129 {
3130         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3131                                                        BTRFS_BLOCK_GROUP_DATA);
3132 }
3133
3134 /*
3135  * This will check the space that the inode allocates from to make sure we have
3136  * enough space for bytes.
3137  */
3138 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3139 {
3140         struct btrfs_space_info *data_sinfo;
3141         struct btrfs_root *root = BTRFS_I(inode)->root;
3142         u64 used;
3143         int ret = 0, committed = 0, alloc_chunk = 1;
3144
3145         /* make sure bytes are sectorsize aligned */
3146         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3147
3148         if (root == root->fs_info->tree_root ||
3149             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3150                 alloc_chunk = 0;
3151                 committed = 1;
3152         }
3153
3154         data_sinfo = BTRFS_I(inode)->space_info;
3155         if (!data_sinfo)
3156                 goto alloc;
3157
3158 again:
3159         /* make sure we have enough space to handle the data first */
3160         spin_lock(&data_sinfo->lock);
3161         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3162                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3163                 data_sinfo->bytes_may_use;
3164
3165         if (used + bytes > data_sinfo->total_bytes) {
3166                 struct btrfs_trans_handle *trans;
3167
3168                 /*
3169                  * if we don't have enough free bytes in this space then we need
3170                  * to alloc a new chunk.
3171                  */
3172                 if (!data_sinfo->full && alloc_chunk) {
3173                         u64 alloc_target;
3174
3175                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3176                         spin_unlock(&data_sinfo->lock);
3177 alloc:
3178                         alloc_target = btrfs_get_alloc_profile(root, 1);
3179                         trans = btrfs_join_transaction(root, 1);
3180                         if (IS_ERR(trans))
3181                                 return PTR_ERR(trans);
3182
3183                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3184                                              bytes + 2 * 1024 * 1024,
3185                                              alloc_target,
3186                                              CHUNK_ALLOC_NO_FORCE);
3187                         btrfs_end_transaction(trans, root);
3188                         if (ret < 0) {
3189                                 if (ret != -ENOSPC)
3190                                         return ret;
3191                                 else
3192                                         goto commit_trans;
3193                         }
3194
3195                         if (!data_sinfo) {
3196                                 btrfs_set_inode_space_info(root, inode);
3197                                 data_sinfo = BTRFS_I(inode)->space_info;
3198                         }
3199                         goto again;
3200                 }
3201                 spin_unlock(&data_sinfo->lock);
3202
3203                 /* commit the current transaction and try again */
3204 commit_trans:
3205                 if (!committed && !root->fs_info->open_ioctl_trans) {
3206                         committed = 1;
3207                         trans = btrfs_join_transaction(root, 1);
3208                         if (IS_ERR(trans))
3209                                 return PTR_ERR(trans);
3210                         ret = btrfs_commit_transaction(trans, root);
3211                         if (ret)
3212                                 return ret;
3213                         goto again;
3214                 }
3215
3216 #if 0 /* I hope we never need this code again, just in case */
3217                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3218                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3219                        "%llu bytes_readonly, %llu may use %llu total\n",
3220                        (unsigned long long)bytes,
3221                        (unsigned long long)data_sinfo->bytes_used,
3222                        (unsigned long long)data_sinfo->bytes_reserved,
3223                        (unsigned long long)data_sinfo->bytes_pinned,
3224                        (unsigned long long)data_sinfo->bytes_readonly,
3225                        (unsigned long long)data_sinfo->bytes_may_use,
3226                        (unsigned long long)data_sinfo->total_bytes);
3227 #endif
3228                 return -ENOSPC;
3229         }
3230         data_sinfo->bytes_may_use += bytes;
3231         BTRFS_I(inode)->reserved_bytes += bytes;
3232         spin_unlock(&data_sinfo->lock);
3233
3234         return 0;
3235 }
3236
3237 /*
3238  * called when we are clearing an delalloc extent from the
3239  * inode's io_tree or there was an error for whatever reason
3240  * after calling btrfs_check_data_free_space
3241  */
3242 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3243 {
3244         struct btrfs_root *root = BTRFS_I(inode)->root;
3245         struct btrfs_space_info *data_sinfo;
3246
3247         /* make sure bytes are sectorsize aligned */
3248         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3249
3250         data_sinfo = BTRFS_I(inode)->space_info;
3251         spin_lock(&data_sinfo->lock);
3252         data_sinfo->bytes_may_use -= bytes;
3253         BTRFS_I(inode)->reserved_bytes -= bytes;
3254         spin_unlock(&data_sinfo->lock);
3255 }
3256
3257 static void force_metadata_allocation(struct btrfs_fs_info *info)
3258 {
3259         struct list_head *head = &info->space_info;
3260         struct btrfs_space_info *found;
3261
3262         rcu_read_lock();
3263         list_for_each_entry_rcu(found, head, list) {
3264                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3265                         found->force_alloc = CHUNK_ALLOC_FORCE;
3266         }
3267         rcu_read_unlock();
3268 }
3269
3270 static int should_alloc_chunk(struct btrfs_root *root,
3271                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3272                               int force)
3273 {
3274         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3275         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3276         u64 thresh;
3277
3278         if (force == CHUNK_ALLOC_FORCE)
3279                 return 1;
3280
3281         /*
3282          * in limited mode, we want to have some free space up to
3283          * about 1% of the FS size.
3284          */
3285         if (force == CHUNK_ALLOC_LIMITED) {
3286                 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3287                 thresh = max_t(u64, 64 * 1024 * 1024,
3288                                div_factor_fine(thresh, 1));
3289
3290                 if (num_bytes - num_allocated < thresh)
3291                         return 1;
3292         }
3293
3294         /*
3295          * we have two similar checks here, one based on percentage
3296          * and once based on a hard number of 256MB.  The idea
3297          * is that if we have a good amount of free
3298          * room, don't allocate a chunk.  A good mount is
3299          * less than 80% utilized of the chunks we have allocated,
3300          * or more than 256MB free
3301          */
3302         if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3303                 return 0;
3304
3305         if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3306                 return 0;
3307
3308         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3309
3310         /* 256MB or 5% of the FS */
3311         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3312
3313         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3314                 return 0;
3315         return 1;
3316 }
3317
3318 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3319                           struct btrfs_root *extent_root, u64 alloc_bytes,
3320                           u64 flags, int force)
3321 {
3322         struct btrfs_space_info *space_info;
3323         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3324         int wait_for_alloc = 0;
3325         int ret = 0;
3326
3327         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3328
3329         space_info = __find_space_info(extent_root->fs_info, flags);
3330         if (!space_info) {
3331                 ret = update_space_info(extent_root->fs_info, flags,
3332                                         0, 0, &space_info);
3333                 BUG_ON(ret);
3334         }
3335         BUG_ON(!space_info);
3336
3337 again:
3338         spin_lock(&space_info->lock);
3339         if (space_info->force_alloc)
3340                 force = space_info->force_alloc;
3341         if (space_info->full) {
3342                 spin_unlock(&space_info->lock);
3343                 return 0;
3344         }
3345
3346         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3347                 spin_unlock(&space_info->lock);
3348                 return 0;
3349         } else if (space_info->chunk_alloc) {
3350                 wait_for_alloc = 1;
3351         } else {
3352                 space_info->chunk_alloc = 1;
3353         }
3354
3355         spin_unlock(&space_info->lock);
3356
3357         mutex_lock(&fs_info->chunk_mutex);
3358
3359         /*
3360          * The chunk_mutex is held throughout the entirety of a chunk
3361          * allocation, so once we've acquired the chunk_mutex we know that the
3362          * other guy is done and we need to recheck and see if we should
3363          * allocate.
3364          */
3365         if (wait_for_alloc) {
3366                 mutex_unlock(&fs_info->chunk_mutex);
3367                 wait_for_alloc = 0;
3368                 goto again;
3369         }
3370
3371         /*
3372          * If we have mixed data/metadata chunks we want to make sure we keep
3373          * allocating mixed chunks instead of individual chunks.
3374          */
3375         if (btrfs_mixed_space_info(space_info))
3376                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3377
3378         /*
3379          * if we're doing a data chunk, go ahead and make sure that
3380          * we keep a reasonable number of metadata chunks allocated in the
3381          * FS as well.
3382          */
3383         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3384                 fs_info->data_chunk_allocations++;
3385                 if (!(fs_info->data_chunk_allocations %
3386                       fs_info->metadata_ratio))
3387                         force_metadata_allocation(fs_info);
3388         }
3389
3390         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3391         spin_lock(&space_info->lock);
3392         if (ret)
3393                 space_info->full = 1;
3394         else
3395                 ret = 1;
3396
3397         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3398         space_info->chunk_alloc = 0;
3399         spin_unlock(&space_info->lock);
3400         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3401         return ret;
3402 }
3403
3404 /*
3405  * shrink metadata reservation for delalloc
3406  */
3407 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3408                            struct btrfs_root *root, u64 to_reclaim, int sync)
3409 {
3410         struct btrfs_block_rsv *block_rsv;
3411         struct btrfs_space_info *space_info;
3412         u64 reserved;
3413         u64 max_reclaim;
3414         u64 reclaimed = 0;
3415         long time_left;
3416         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3417         int loops = 0;
3418         unsigned long progress;
3419
3420         block_rsv = &root->fs_info->delalloc_block_rsv;
3421         space_info = block_rsv->space_info;
3422
3423         smp_mb();
3424         reserved = space_info->bytes_reserved;
3425         progress = space_info->reservation_progress;
3426
3427         if (reserved == 0)
3428                 return 0;
3429
3430         max_reclaim = min(reserved, to_reclaim);
3431
3432         while (loops < 1024) {
3433                 /* have the flusher threads jump in and do some IO */
3434                 smp_mb();
3435                 nr_pages = min_t(unsigned long, nr_pages,
3436                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3437                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3438
3439                 spin_lock(&space_info->lock);
3440                 if (reserved > space_info->bytes_reserved)
3441                         reclaimed += reserved - space_info->bytes_reserved;
3442                 reserved = space_info->bytes_reserved;
3443                 spin_unlock(&space_info->lock);
3444
3445                 loops++;
3446
3447                 if (reserved == 0 || reclaimed >= max_reclaim)
3448                         break;
3449
3450                 if (trans && trans->transaction->blocked)
3451                         return -EAGAIN;
3452
3453                 time_left = schedule_timeout_interruptible(1);
3454
3455                 /* We were interrupted, exit */
3456                 if (time_left)
3457                         break;
3458
3459                 /* we've kicked the IO a few times, if anything has been freed,
3460                  * exit.  There is no sense in looping here for a long time
3461                  * when we really need to commit the transaction, or there are
3462                  * just too many writers without enough free space
3463                  */
3464
3465                 if (loops > 3) {
3466                         smp_mb();
3467                         if (progress != space_info->reservation_progress)
3468                                 break;
3469                 }
3470
3471         }
3472         return reclaimed >= to_reclaim;
3473 }
3474
3475 /*
3476  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3477  * idea is if this is the first call (retries == 0) then we will add to our
3478  * reserved count if we can't make the allocation in order to hold our place
3479  * while we go and try and free up space.  That way for retries > 1 we don't try
3480  * and add space, we just check to see if the amount of unused space is >= the
3481  * total space, meaning that our reservation is valid.
3482  *
3483  * However if we don't intend to retry this reservation, pass -1 as retries so
3484  * that it short circuits this logic.
3485  */
3486 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3487                                   struct btrfs_root *root,
3488                                   struct btrfs_block_rsv *block_rsv,
3489                                   u64 orig_bytes, int flush)
3490 {
3491         struct btrfs_space_info *space_info = block_rsv->space_info;
3492         u64 unused;
3493         u64 num_bytes = orig_bytes;
3494         int retries = 0;
3495         int ret = 0;
3496         bool reserved = false;
3497         bool committed = false;
3498
3499 again:
3500         ret = -ENOSPC;
3501         if (reserved)
3502                 num_bytes = 0;
3503
3504         spin_lock(&space_info->lock);
3505         unused = space_info->bytes_used + space_info->bytes_reserved +
3506                  space_info->bytes_pinned + space_info->bytes_readonly +
3507                  space_info->bytes_may_use;
3508
3509         /*
3510          * The idea here is that we've not already over-reserved the block group
3511          * then we can go ahead and save our reservation first and then start
3512          * flushing if we need to.  Otherwise if we've already overcommitted
3513          * lets start flushing stuff first and then come back and try to make
3514          * our reservation.
3515          */
3516         if (unused <= space_info->total_bytes) {
3517                 unused = space_info->total_bytes - unused;
3518                 if (unused >= num_bytes) {
3519                         if (!reserved)
3520                                 space_info->bytes_reserved += orig_bytes;
3521                         ret = 0;
3522                 } else {
3523                         /*
3524                          * Ok set num_bytes to orig_bytes since we aren't
3525                          * overocmmitted, this way we only try and reclaim what
3526                          * we need.
3527                          */
3528                         num_bytes = orig_bytes;
3529                 }
3530         } else {
3531                 /*
3532                  * Ok we're over committed, set num_bytes to the overcommitted
3533                  * amount plus the amount of bytes that we need for this
3534                  * reservation.
3535                  */
3536                 num_bytes = unused - space_info->total_bytes +
3537                         (orig_bytes * (retries + 1));
3538         }
3539
3540         /*
3541          * Couldn't make our reservation, save our place so while we're trying
3542          * to reclaim space we can actually use it instead of somebody else
3543          * stealing it from us.
3544          */
3545         if (ret && !reserved) {
3546                 space_info->bytes_reserved += orig_bytes;
3547                 reserved = true;
3548         }
3549
3550         spin_unlock(&space_info->lock);
3551
3552         if (!ret)
3553                 return 0;
3554
3555         if (!flush)
3556                 goto out;
3557
3558         /*
3559          * We do synchronous shrinking since we don't actually unreserve
3560          * metadata until after the IO is completed.
3561          */
3562         ret = shrink_delalloc(trans, root, num_bytes, 1);
3563         if (ret > 0)
3564                 return 0;
3565         else if (ret < 0)
3566                 goto out;
3567
3568         /*
3569          * So if we were overcommitted it's possible that somebody else flushed
3570          * out enough space and we simply didn't have enough space to reclaim,
3571          * so go back around and try again.
3572          */
3573         if (retries < 2) {
3574                 retries++;
3575                 goto again;
3576         }
3577
3578         spin_lock(&space_info->lock);
3579         /*
3580          * Not enough space to be reclaimed, don't bother committing the
3581          * transaction.
3582          */
3583         if (space_info->bytes_pinned < orig_bytes)
3584                 ret = -ENOSPC;
3585         spin_unlock(&space_info->lock);
3586         if (ret)
3587                 goto out;
3588
3589         ret = -EAGAIN;
3590         if (trans || committed)
3591                 goto out;
3592
3593         ret = -ENOSPC;
3594         trans = btrfs_join_transaction(root, 1);
3595         if (IS_ERR(trans))
3596                 goto out;
3597         ret = btrfs_commit_transaction(trans, root);
3598         if (!ret) {
3599                 trans = NULL;
3600                 committed = true;
3601                 goto again;
3602         }
3603
3604 out:
3605         if (reserved) {
3606                 spin_lock(&space_info->lock);
3607                 space_info->bytes_reserved -= orig_bytes;
3608                 spin_unlock(&space_info->lock);
3609         }
3610
3611         return ret;
3612 }
3613
3614 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3615                                              struct btrfs_root *root)
3616 {
3617         struct btrfs_block_rsv *block_rsv;
3618         if (root->ref_cows)
3619                 block_rsv = trans->block_rsv;
3620         else
3621                 block_rsv = root->block_rsv;
3622
3623         if (!block_rsv)
3624                 block_rsv = &root->fs_info->empty_block_rsv;
3625
3626         return block_rsv;
3627 }
3628
3629 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3630                                u64 num_bytes)
3631 {
3632         int ret = -ENOSPC;
3633         spin_lock(&block_rsv->lock);
3634         if (block_rsv->reserved >= num_bytes) {
3635                 block_rsv->reserved -= num_bytes;
3636                 if (block_rsv->reserved < block_rsv->size)
3637                         block_rsv->full = 0;
3638                 ret = 0;
3639         }
3640         spin_unlock(&block_rsv->lock);
3641         return ret;
3642 }
3643
3644 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3645                                 u64 num_bytes, int update_size)
3646 {
3647         spin_lock(&block_rsv->lock);
3648         block_rsv->reserved += num_bytes;
3649         if (update_size)
3650                 block_rsv->size += num_bytes;
3651         else if (block_rsv->reserved >= block_rsv->size)
3652                 block_rsv->full = 1;
3653         spin_unlock(&block_rsv->lock);
3654 }
3655
3656 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3657                              struct btrfs_block_rsv *dest, u64 num_bytes)
3658 {
3659         struct btrfs_space_info *space_info = block_rsv->space_info;
3660
3661         spin_lock(&block_rsv->lock);
3662         if (num_bytes == (u64)-1)
3663                 num_bytes = block_rsv->size;
3664         block_rsv->size -= num_bytes;
3665         if (block_rsv->reserved >= block_rsv->size) {
3666                 num_bytes = block_rsv->reserved - block_rsv->size;
3667                 block_rsv->reserved = block_rsv->size;
3668                 block_rsv->full = 1;
3669         } else {
3670                 num_bytes = 0;
3671         }
3672         spin_unlock(&block_rsv->lock);
3673
3674         if (num_bytes > 0) {
3675                 if (dest) {
3676                         spin_lock(&dest->lock);
3677                         if (!dest->full) {
3678                                 u64 bytes_to_add;
3679
3680                                 bytes_to_add = dest->size - dest->reserved;
3681                                 bytes_to_add = min(num_bytes, bytes_to_add);
3682                                 dest->reserved += bytes_to_add;
3683                                 if (dest->reserved >= dest->size)
3684                                         dest->full = 1;
3685                                 num_bytes -= bytes_to_add;
3686                         }
3687                         spin_unlock(&dest->lock);
3688                 }
3689                 if (num_bytes) {
3690                         spin_lock(&space_info->lock);
3691                         space_info->bytes_reserved -= num_bytes;
3692                         space_info->reservation_progress++;
3693                         spin_unlock(&space_info->lock);
3694                 }
3695         }
3696 }
3697
3698 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3699                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3700 {
3701         int ret;
3702
3703         ret = block_rsv_use_bytes(src, num_bytes);
3704         if (ret)
3705                 return ret;
3706
3707         block_rsv_add_bytes(dst, num_bytes, 1);
3708         return 0;
3709 }
3710
3711 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3712 {
3713         memset(rsv, 0, sizeof(*rsv));
3714         spin_lock_init(&rsv->lock);
3715         atomic_set(&rsv->usage, 1);
3716         rsv->priority = 6;
3717         INIT_LIST_HEAD(&rsv->list);
3718 }
3719
3720 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3721 {
3722         struct btrfs_block_rsv *block_rsv;
3723         struct btrfs_fs_info *fs_info = root->fs_info;
3724
3725         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3726         if (!block_rsv)
3727                 return NULL;
3728
3729         btrfs_init_block_rsv(block_rsv);
3730         block_rsv->space_info = __find_space_info(fs_info,
3731                                                   BTRFS_BLOCK_GROUP_METADATA);
3732         return block_rsv;
3733 }
3734
3735 void btrfs_free_block_rsv(struct btrfs_root *root,
3736                           struct btrfs_block_rsv *rsv)
3737 {
3738         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3739                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3740                 if (!rsv->durable)
3741                         kfree(rsv);
3742         }
3743 }
3744
3745 /*
3746  * make the block_rsv struct be able to capture freed space.
3747  * the captured space will re-add to the the block_rsv struct
3748  * after transaction commit
3749  */
3750 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3751                                  struct btrfs_block_rsv *block_rsv)
3752 {
3753         block_rsv->durable = 1;
3754         mutex_lock(&fs_info->durable_block_rsv_mutex);
3755         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3756         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3757 }
3758
3759 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3760                         struct btrfs_root *root,
3761                         struct btrfs_block_rsv *block_rsv,
3762                         u64 num_bytes)
3763 {
3764         int ret;
3765
3766         if (num_bytes == 0)
3767                 return 0;
3768
3769         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3770         if (!ret) {
3771                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3772                 return 0;
3773         }
3774
3775         return ret;
3776 }
3777
3778 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3779                           struct btrfs_root *root,
3780                           struct btrfs_block_rsv *block_rsv,
3781                           u64 min_reserved, int min_factor)
3782 {
3783         u64 num_bytes = 0;
3784         int commit_trans = 0;
3785         int ret = -ENOSPC;
3786
3787         if (!block_rsv)
3788                 return 0;
3789
3790         spin_lock(&block_rsv->lock);
3791         if (min_factor > 0)
3792                 num_bytes = div_factor(block_rsv->size, min_factor);
3793         if (min_reserved > num_bytes)
3794                 num_bytes = min_reserved;
3795
3796         if (block_rsv->reserved >= num_bytes) {
3797                 ret = 0;
3798         } else {
3799                 num_bytes -= block_rsv->reserved;
3800                 if (block_rsv->durable &&
3801                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3802                         commit_trans = 1;
3803         }
3804         spin_unlock(&block_rsv->lock);
3805         if (!ret)
3806                 return 0;
3807
3808         if (block_rsv->refill_used) {
3809                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3810                                              num_bytes, 0);
3811                 if (!ret) {
3812                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3813                         return 0;
3814                 }
3815         }
3816
3817         if (commit_trans) {
3818                 if (trans)
3819                         return -EAGAIN;
3820
3821                 trans = btrfs_join_transaction(root, 1);
3822                 BUG_ON(IS_ERR(trans));
3823                 ret = btrfs_commit_transaction(trans, root);
3824                 return 0;
3825         }
3826
3827         return -ENOSPC;
3828 }
3829
3830 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3831                             struct btrfs_block_rsv *dst_rsv,
3832                             u64 num_bytes)
3833 {
3834         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3835 }
3836
3837 void btrfs_block_rsv_release(struct btrfs_root *root,
3838                              struct btrfs_block_rsv *block_rsv,
3839                              u64 num_bytes)
3840 {
3841         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3842         if (global_rsv->full || global_rsv == block_rsv ||
3843             block_rsv->space_info != global_rsv->space_info)
3844                 global_rsv = NULL;
3845         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3846 }
3847
3848 /*
3849  * helper to calculate size of global block reservation.
3850  * the desired value is sum of space used by extent tree,
3851  * checksum tree and root tree
3852  */
3853 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3854 {
3855         struct btrfs_space_info *sinfo;
3856         u64 num_bytes;
3857         u64 meta_used;
3858         u64 data_used;
3859         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3860 #if 0
3861         /*
3862          * per tree used space accounting can be inaccuracy, so we
3863          * can't rely on it.
3864          */
3865         spin_lock(&fs_info->extent_root->accounting_lock);
3866         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3867         spin_unlock(&fs_info->extent_root->accounting_lock);
3868
3869         spin_lock(&fs_info->csum_root->accounting_lock);
3870         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3871         spin_unlock(&fs_info->csum_root->accounting_lock);
3872
3873         spin_lock(&fs_info->tree_root->accounting_lock);
3874         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3875         spin_unlock(&fs_info->tree_root->accounting_lock);
3876 #endif
3877         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3878         spin_lock(&sinfo->lock);
3879         data_used = sinfo->bytes_used;
3880         spin_unlock(&sinfo->lock);
3881
3882         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3883         spin_lock(&sinfo->lock);
3884         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3885                 data_used = 0;
3886         meta_used = sinfo->bytes_used;
3887         spin_unlock(&sinfo->lock);
3888
3889         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3890                     csum_size * 2;
3891         num_bytes += div64_u64(data_used + meta_used, 50);
3892
3893         if (num_bytes * 3 > meta_used)
3894                 num_bytes = div64_u64(meta_used, 3);
3895
3896         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3897 }
3898
3899 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3900 {
3901         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3902         struct btrfs_space_info *sinfo = block_rsv->space_info;
3903         u64 num_bytes;
3904
3905         num_bytes = calc_global_metadata_size(fs_info);
3906
3907         spin_lock(&block_rsv->lock);
3908         spin_lock(&sinfo->lock);
3909
3910         block_rsv->size = num_bytes;
3911
3912         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3913                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3914                     sinfo->bytes_may_use;
3915
3916         if (sinfo->total_bytes > num_bytes) {
3917                 num_bytes = sinfo->total_bytes - num_bytes;
3918                 block_rsv->reserved += num_bytes;
3919                 sinfo->bytes_reserved += num_bytes;
3920         }
3921
3922         if (block_rsv->reserved >= block_rsv->size) {
3923                 num_bytes = block_rsv->reserved - block_rsv->size;
3924                 sinfo->bytes_reserved -= num_bytes;
3925                 sinfo->reservation_progress++;
3926                 block_rsv->reserved = block_rsv->size;
3927                 block_rsv->full = 1;
3928         }
3929 #if 0
3930         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3931                 block_rsv->size, block_rsv->reserved);
3932 #endif
3933         spin_unlock(&sinfo->lock);
3934         spin_unlock(&block_rsv->lock);
3935 }
3936
3937 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3938 {
3939         struct btrfs_space_info *space_info;
3940
3941         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3942         fs_info->chunk_block_rsv.space_info = space_info;
3943         fs_info->chunk_block_rsv.priority = 10;
3944
3945         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3946         fs_info->global_block_rsv.space_info = space_info;
3947         fs_info->global_block_rsv.priority = 10;
3948         fs_info->global_block_rsv.refill_used = 1;
3949         fs_info->delalloc_block_rsv.space_info = space_info;
3950         fs_info->trans_block_rsv.space_info = space_info;
3951         fs_info->empty_block_rsv.space_info = space_info;
3952         fs_info->empty_block_rsv.priority = 10;
3953
3954         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3955         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3956         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3957         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3958         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3959
3960         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3961
3962         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3963
3964         update_global_block_rsv(fs_info);
3965 }
3966
3967 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3968 {
3969         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3970         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3971         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3972         WARN_ON(fs_info->trans_block_rsv.size > 0);
3973         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3974         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3975         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3976 }
3977
3978 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3979 {
3980         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3981                 3 * num_items;
3982 }
3983
3984 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3985                                  struct btrfs_root *root,
3986                                  int num_items)
3987 {
3988         u64 num_bytes;
3989         int ret;
3990
3991         if (num_items == 0 || root->fs_info->chunk_root == root)
3992                 return 0;
3993
3994         num_bytes = calc_trans_metadata_size(root, num_items);
3995         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3996                                   num_bytes);
3997         if (!ret) {
3998                 trans->bytes_reserved += num_bytes;
3999                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4000         }
4001         return ret;
4002 }
4003
4004 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4005                                   struct btrfs_root *root)
4006 {
4007         if (!trans->bytes_reserved)
4008                 return;
4009
4010         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
4011         btrfs_block_rsv_release(root, trans->block_rsv,
4012                                 trans->bytes_reserved);
4013         trans->bytes_reserved = 0;
4014 }
4015
4016 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4017                                   struct inode *inode)
4018 {
4019         struct btrfs_root *root = BTRFS_I(inode)->root;
4020         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4021         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4022
4023         /*
4024          * one for deleting orphan item, one for updating inode and
4025          * two for calling btrfs_truncate_inode_items.
4026          *
4027          * btrfs_truncate_inode_items is a delete operation, it frees
4028          * more space than it uses in most cases. So two units of
4029          * metadata space should be enough for calling it many times.
4030          * If all of the metadata space is used, we can commit
4031          * transaction and use space it freed.
4032          */
4033         u64 num_bytes = calc_trans_metadata_size(root, 4);
4034         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4035 }
4036
4037 void btrfs_orphan_release_metadata(struct inode *inode)
4038 {
4039         struct btrfs_root *root = BTRFS_I(inode)->root;
4040         u64 num_bytes = calc_trans_metadata_size(root, 4);
4041         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4042 }
4043
4044 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4045                                 struct btrfs_pending_snapshot *pending)
4046 {
4047         struct btrfs_root *root = pending->root;
4048         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4049         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4050         /*
4051          * two for root back/forward refs, two for directory entries
4052          * and one for root of the snapshot.
4053          */
4054         u64 num_bytes = calc_trans_metadata_size(root, 5);
4055         dst_rsv->space_info = src_rsv->space_info;
4056         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4057 }
4058
4059 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
4060 {
4061         return num_bytes >>= 3;
4062 }
4063
4064 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4065 {
4066         struct btrfs_root *root = BTRFS_I(inode)->root;
4067         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4068         u64 to_reserve;
4069         int nr_extents;
4070         int reserved_extents;
4071         int ret;
4072
4073         if (btrfs_transaction_in_commit(root->fs_info))
4074                 schedule_timeout(1);
4075
4076         num_bytes = ALIGN(num_bytes, root->sectorsize);
4077
4078         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4079         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4080
4081         if (nr_extents > reserved_extents) {
4082                 nr_extents -= reserved_extents;
4083                 to_reserve = calc_trans_metadata_size(root, nr_extents);
4084         } else {
4085                 nr_extents = 0;
4086                 to_reserve = 0;
4087         }
4088
4089         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4090         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4091         if (ret)
4092                 return ret;
4093
4094         atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
4095         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4096
4097         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4098
4099         if (block_rsv->size > 512 * 1024 * 1024)
4100                 shrink_delalloc(NULL, root, to_reserve, 0);
4101
4102         return 0;
4103 }
4104
4105 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4106 {
4107         struct btrfs_root *root = BTRFS_I(inode)->root;
4108         u64 to_free;
4109         int nr_extents;
4110         int reserved_extents;
4111
4112         num_bytes = ALIGN(num_bytes, root->sectorsize);
4113         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4114         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4115
4116         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4117         do {
4118                 int old, new;
4119
4120                 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4121                 if (nr_extents >= reserved_extents) {
4122                         nr_extents = 0;
4123                         break;
4124                 }
4125                 old = reserved_extents;
4126                 nr_extents = reserved_extents - nr_extents;
4127                 new = reserved_extents - nr_extents;
4128                 old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
4129                                      reserved_extents, new);
4130                 if (likely(old == reserved_extents))
4131                         break;
4132                 reserved_extents = old;
4133         } while (1);
4134
4135         to_free = calc_csum_metadata_size(inode, num_bytes);
4136         if (nr_extents > 0)
4137                 to_free += calc_trans_metadata_size(root, nr_extents);
4138
4139         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4140                                 to_free);
4141 }
4142
4143 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4144 {
4145         int ret;
4146
4147         ret = btrfs_check_data_free_space(inode, num_bytes);
4148         if (ret)
4149                 return ret;
4150
4151         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4152         if (ret) {
4153                 btrfs_free_reserved_data_space(inode, num_bytes);
4154                 return ret;
4155         }
4156
4157         return 0;
4158 }
4159
4160 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4161 {
4162         btrfs_delalloc_release_metadata(inode, num_bytes);
4163         btrfs_free_reserved_data_space(inode, num_bytes);
4164 }
4165
4166 static int update_block_group(struct btrfs_trans_handle *trans,
4167                               struct btrfs_root *root,
4168                               u64 bytenr, u64 num_bytes, int alloc)
4169 {
4170         struct btrfs_block_group_cache *cache = NULL;
4171         struct btrfs_fs_info *info = root->fs_info;
4172         u64 total = num_bytes;
4173         u64 old_val;
4174         u64 byte_in_group;
4175         int factor;
4176
4177         /* block accounting for super block */
4178         spin_lock(&info->delalloc_lock);
4179         old_val = btrfs_super_bytes_used(&info->super_copy);
4180         if (alloc)
4181                 old_val += num_bytes;
4182         else
4183                 old_val -= num_bytes;
4184         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4185         spin_unlock(&info->delalloc_lock);
4186
4187         while (total) {
4188                 cache = btrfs_lookup_block_group(info, bytenr);
4189                 if (!cache)
4190                         return -1;
4191                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4192                                     BTRFS_BLOCK_GROUP_RAID1 |
4193                                     BTRFS_BLOCK_GROUP_RAID10))
4194                         factor = 2;
4195                 else
4196                         factor = 1;
4197                 /*
4198                  * If this block group has free space cache written out, we
4199                  * need to make sure to load it if we are removing space.  This
4200                  * is because we need the unpinning stage to actually add the
4201                  * space back to the block group, otherwise we will leak space.
4202                  */
4203                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4204                         cache_block_group(cache, trans, NULL, 1);
4205
4206                 byte_in_group = bytenr - cache->key.objectid;
4207                 WARN_ON(byte_in_group > cache->key.offset);
4208
4209                 spin_lock(&cache->space_info->lock);
4210                 spin_lock(&cache->lock);
4211
4212                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4213                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4214                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4215
4216                 cache->dirty = 1;
4217                 old_val = btrfs_block_group_used(&cache->item);
4218                 num_bytes = min(total, cache->key.offset - byte_in_group);
4219                 if (alloc) {
4220                         old_val += num_bytes;
4221                         btrfs_set_block_group_used(&cache->item, old_val);
4222                         cache->reserved -= num_bytes;
4223                         cache->space_info->bytes_reserved -= num_bytes;
4224                         cache->space_info->reservation_progress++;
4225                         cache->space_info->bytes_used += num_bytes;
4226                         cache->space_info->disk_used += num_bytes * factor;
4227                         spin_unlock(&cache->lock);
4228                         spin_unlock(&cache->space_info->lock);
4229                 } else {
4230                         old_val -= num_bytes;
4231                         btrfs_set_block_group_used(&cache->item, old_val);
4232                         cache->pinned += num_bytes;
4233                         cache->space_info->bytes_pinned += num_bytes;
4234                         cache->space_info->bytes_used -= num_bytes;
4235                         cache->space_info->disk_used -= num_bytes * factor;
4236                         spin_unlock(&cache->lock);
4237                         spin_unlock(&cache->space_info->lock);
4238
4239                         set_extent_dirty(info->pinned_extents,
4240                                          bytenr, bytenr + num_bytes - 1,
4241                                          GFP_NOFS | __GFP_NOFAIL);
4242                 }
4243                 btrfs_put_block_group(cache);
4244                 total -= num_bytes;
4245                 bytenr += num_bytes;
4246         }
4247         return 0;
4248 }
4249
4250 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4251 {
4252         struct btrfs_block_group_cache *cache;
4253         u64 bytenr;
4254
4255         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4256         if (!cache)
4257                 return 0;
4258
4259         bytenr = cache->key.objectid;
4260         btrfs_put_block_group(cache);
4261
4262         return bytenr;
4263 }
4264
4265 static int pin_down_extent(struct btrfs_root *root,
4266                            struct btrfs_block_group_cache *cache,
4267                            u64 bytenr, u64 num_bytes, int reserved)
4268 {
4269         spin_lock(&cache->space_info->lock);
4270         spin_lock(&cache->lock);
4271         cache->pinned += num_bytes;
4272         cache->space_info->bytes_pinned += num_bytes;
4273         if (reserved) {
4274                 cache->reserved -= num_bytes;
4275                 cache->space_info->bytes_reserved -= num_bytes;
4276                 cache->space_info->reservation_progress++;
4277         }
4278         spin_unlock(&cache->lock);
4279         spin_unlock(&cache->space_info->lock);
4280
4281         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4282                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4283         return 0;
4284 }
4285
4286 /*
4287  * this function must be called within transaction
4288  */
4289 int btrfs_pin_extent(struct btrfs_root *root,
4290                      u64 bytenr, u64 num_bytes, int reserved)
4291 {
4292         struct btrfs_block_group_cache *cache;
4293
4294         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4295         BUG_ON(!cache);
4296
4297         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4298
4299         btrfs_put_block_group(cache);
4300         return 0;
4301 }
4302
4303 /*
4304  * update size of reserved extents. this function may return -EAGAIN
4305  * if 'reserve' is true or 'sinfo' is false.
4306  */
4307 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4308                                 u64 num_bytes, int reserve, int sinfo)
4309 {
4310         int ret = 0;
4311         if (sinfo) {
4312                 struct btrfs_space_info *space_info = cache->space_info;
4313                 spin_lock(&space_info->lock);
4314                 spin_lock(&cache->lock);
4315                 if (reserve) {
4316                         if (cache->ro) {
4317                                 ret = -EAGAIN;
4318                         } else {
4319                                 cache->reserved += num_bytes;
4320                                 space_info->bytes_reserved += num_bytes;
4321                         }
4322                 } else {
4323                         if (cache->ro)
4324                                 space_info->bytes_readonly += num_bytes;
4325                         cache->reserved -= num_bytes;
4326                         space_info->bytes_reserved -= num_bytes;
4327                         space_info->reservation_progress++;
4328                 }
4329                 spin_unlock(&cache->lock);
4330                 spin_unlock(&space_info->lock);
4331         } else {
4332                 spin_lock(&cache->lock);
4333                 if (cache->ro) {
4334                         ret = -EAGAIN;
4335                 } else {
4336                         if (reserve)
4337                                 cache->reserved += num_bytes;
4338                         else
4339                                 cache->reserved -= num_bytes;
4340                 }
4341                 spin_unlock(&cache->lock);
4342         }
4343         return ret;
4344 }
4345
4346 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4347                                 struct btrfs_root *root)
4348 {
4349         struct btrfs_fs_info *fs_info = root->fs_info;
4350         struct btrfs_caching_control *next;
4351         struct btrfs_caching_control *caching_ctl;
4352         struct btrfs_block_group_cache *cache;
4353
4354         down_write(&fs_info->extent_commit_sem);
4355
4356         list_for_each_entry_safe(caching_ctl, next,
4357                                  &fs_info->caching_block_groups, list) {
4358                 cache = caching_ctl->block_group;
4359                 if (block_group_cache_done(cache)) {
4360                         cache->last_byte_to_unpin = (u64)-1;
4361                         list_del_init(&caching_ctl->list);
4362                         put_caching_control(caching_ctl);
4363                 } else {
4364                         cache->last_byte_to_unpin = caching_ctl->progress;
4365                 }
4366         }
4367
4368         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4369                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4370         else
4371                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4372
4373         up_write(&fs_info->extent_commit_sem);
4374
4375         update_global_block_rsv(fs_info);
4376         return 0;
4377 }
4378
4379 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4380 {
4381         struct btrfs_fs_info *fs_info = root->fs_info;
4382         struct btrfs_block_group_cache *cache = NULL;
4383         u64 len;
4384
4385         while (start <= end) {
4386                 if (!cache ||
4387                     start >= cache->key.objectid + cache->key.offset) {
4388                         if (cache)
4389                                 btrfs_put_block_group(cache);
4390                         cache = btrfs_lookup_block_group(fs_info, start);
4391                         BUG_ON(!cache);
4392                 }
4393
4394                 len = cache->key.objectid + cache->key.offset - start;
4395                 len = min(len, end + 1 - start);
4396
4397                 if (start < cache->last_byte_to_unpin) {
4398                         len = min(len, cache->last_byte_to_unpin - start);
4399                         btrfs_add_free_space(cache, start, len);
4400                 }
4401
4402                 start += len;
4403
4404                 spin_lock(&cache->space_info->lock);
4405                 spin_lock(&cache->lock);
4406                 cache->pinned -= len;
4407                 cache->space_info->bytes_pinned -= len;
4408                 if (cache->ro) {
4409                         cache->space_info->bytes_readonly += len;
4410                 } else if (cache->reserved_pinned > 0) {
4411                         len = min(len, cache->reserved_pinned);
4412                         cache->reserved_pinned -= len;
4413                         cache->space_info->bytes_reserved += len;
4414                 }
4415                 spin_unlock(&cache->lock);
4416                 spin_unlock(&cache->space_info->lock);
4417         }
4418
4419         if (cache)
4420                 btrfs_put_block_group(cache);
4421         return 0;
4422 }
4423
4424 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4425                                struct btrfs_root *root)
4426 {
4427         struct btrfs_fs_info *fs_info = root->fs_info;
4428         struct extent_io_tree *unpin;
4429         struct btrfs_block_rsv *block_rsv;
4430         struct btrfs_block_rsv *next_rsv;
4431         u64 start;
4432         u64 end;
4433         int idx;
4434         int ret;
4435
4436         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4437                 unpin = &fs_info->freed_extents[1];
4438         else
4439                 unpin = &fs_info->freed_extents[0];
4440
4441         while (1) {
4442                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4443                                             EXTENT_DIRTY);
4444                 if (ret)
4445                         break;
4446
4447                 if (btrfs_test_opt(root, DISCARD))
4448                         ret = btrfs_discard_extent(root, start,
4449                                                    end + 1 - start, NULL);
4450
4451                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4452                 unpin_extent_range(root, start, end);
4453                 cond_resched();
4454         }
4455
4456         mutex_lock(&fs_info->durable_block_rsv_mutex);
4457         list_for_each_entry_safe(block_rsv, next_rsv,
4458                                  &fs_info->durable_block_rsv_list, list) {
4459
4460                 idx = trans->transid & 0x1;
4461                 if (block_rsv->freed[idx] > 0) {
4462                         block_rsv_add_bytes(block_rsv,
4463                                             block_rsv->freed[idx], 0);
4464                         block_rsv->freed[idx] = 0;
4465                 }
4466                 if (atomic_read(&block_rsv->usage) == 0) {
4467                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4468
4469                         if (block_rsv->freed[0] == 0 &&
4470                             block_rsv->freed[1] == 0) {
4471                                 list_del_init(&block_rsv->list);
4472                                 kfree(block_rsv);
4473                         }
4474                 } else {
4475                         btrfs_block_rsv_release(root, block_rsv, 0);
4476                 }
4477         }
4478         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4479
4480         return 0;
4481 }
4482
4483 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4484                                 struct btrfs_root *root,
4485                                 u64 bytenr, u64 num_bytes, u64 parent,
4486                                 u64 root_objectid, u64 owner_objectid,
4487                                 u64 owner_offset, int refs_to_drop,
4488                                 struct btrfs_delayed_extent_op *extent_op)
4489 {
4490         struct btrfs_key key;
4491         struct btrfs_path *path;
4492         struct btrfs_fs_info *info = root->fs_info;
4493         struct btrfs_root *extent_root = info->extent_root;
4494         struct extent_buffer *leaf;
4495         struct btrfs_extent_item *ei;
4496         struct btrfs_extent_inline_ref *iref;
4497         int ret;
4498         int is_data;
4499         int extent_slot = 0;
4500         int found_extent = 0;
4501         int num_to_del = 1;
4502         u32 item_size;
4503         u64 refs;
4504
4505         path = btrfs_alloc_path();
4506         if (!path)
4507                 return -ENOMEM;
4508
4509         path->reada = 1;
4510         path->leave_spinning = 1;
4511
4512         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4513         BUG_ON(!is_data && refs_to_drop != 1);
4514
4515         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4516                                     bytenr, num_bytes, parent,
4517                                     root_objectid, owner_objectid,
4518                                     owner_offset);
4519         if (ret == 0) {
4520                 extent_slot = path->slots[0];
4521                 while (extent_slot >= 0) {
4522                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4523                                               extent_slot);
4524                         if (key.objectid != bytenr)
4525                                 break;
4526                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4527                             key.offset == num_bytes) {
4528                                 found_extent = 1;
4529                                 break;
4530                         }
4531                         if (path->slots[0] - extent_slot > 5)
4532                                 break;
4533                         extent_slot--;
4534                 }
4535 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4536                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4537                 if (found_extent && item_size < sizeof(*ei))
4538                         found_extent = 0;
4539 #endif
4540                 if (!found_extent) {
4541                         BUG_ON(iref);
4542                         ret = remove_extent_backref(trans, extent_root, path,
4543                                                     NULL, refs_to_drop,
4544                                                     is_data);
4545                         BUG_ON(ret);
4546                         btrfs_release_path(extent_root, path);
4547                         path->leave_spinning = 1;
4548
4549                         key.objectid = bytenr;
4550                         key.type = BTRFS_EXTENT_ITEM_KEY;
4551                         key.offset = num_bytes;
4552
4553                         ret = btrfs_search_slot(trans, extent_root,
4554                                                 &key, path, -1, 1);
4555                         if (ret) {
4556                                 printk(KERN_ERR "umm, got %d back from search"
4557                                        ", was looking for %llu\n", ret,
4558                                        (unsigned long long)bytenr);
4559                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4560                         }
4561                         BUG_ON(ret);
4562                         extent_slot = path->slots[0];
4563                 }
4564         } else {
4565                 btrfs_print_leaf(extent_root, path->nodes[0]);
4566                 WARN_ON(1);
4567                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4568                        "parent %llu root %llu  owner %llu offset %llu\n",
4569                        (unsigned long long)bytenr,
4570                        (unsigned long long)parent,
4571                        (unsigned long long)root_objectid,
4572                        (unsigned long long)owner_objectid,
4573                        (unsigned long long)owner_offset);
4574         }
4575
4576         leaf = path->nodes[0];
4577         item_size = btrfs_item_size_nr(leaf, extent_slot);
4578 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4579         if (item_size < sizeof(*ei)) {
4580                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4581                 ret = convert_extent_item_v0(trans, extent_root, path,
4582                                              owner_objectid, 0);
4583                 BUG_ON(ret < 0);
4584
4585                 btrfs_release_path(extent_root, path);
4586                 path->leave_spinning = 1;
4587
4588                 key.objectid = bytenr;
4589                 key.type = BTRFS_EXTENT_ITEM_KEY;
4590                 key.offset = num_bytes;
4591
4592                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4593                                         -1, 1);
4594                 if (ret) {
4595                         printk(KERN_ERR "umm, got %d back from search"
4596                                ", was looking for %llu\n", ret,
4597                                (unsigned long long)bytenr);
4598                         btrfs_print_leaf(extent_root, path->nodes[0]);
4599                 }
4600                 BUG_ON(ret);
4601                 extent_slot = path->slots[0];
4602                 leaf = path->nodes[0];
4603                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4604         }
4605 #endif
4606         BUG_ON(item_size < sizeof(*ei));
4607         ei = btrfs_item_ptr(leaf, extent_slot,
4608                             struct btrfs_extent_item);
4609         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4610                 struct btrfs_tree_block_info *bi;
4611                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4612                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4613                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4614         }
4615
4616         refs = btrfs_extent_refs(leaf, ei);
4617         BUG_ON(refs < refs_to_drop);
4618         refs -= refs_to_drop;
4619
4620         if (refs > 0) {
4621                 if (extent_op)
4622                         __run_delayed_extent_op(extent_op, leaf, ei);
4623                 /*
4624                  * In the case of inline back ref, reference count will
4625                  * be updated by remove_extent_backref
4626                  */
4627                 if (iref) {
4628                         BUG_ON(!found_extent);
4629                 } else {
4630                         btrfs_set_extent_refs(leaf, ei, refs);
4631                         btrfs_mark_buffer_dirty(leaf);
4632                 }
4633                 if (found_extent) {
4634                         ret = remove_extent_backref(trans, extent_root, path,
4635                                                     iref, refs_to_drop,
4636                                                     is_data);
4637                         BUG_ON(ret);
4638                 }
4639         } else {
4640                 if (found_extent) {
4641                         BUG_ON(is_data && refs_to_drop !=
4642                                extent_data_ref_count(root, path, iref));
4643                         if (iref) {
4644                                 BUG_ON(path->slots[0] != extent_slot);
4645                         } else {
4646                                 BUG_ON(path->slots[0] != extent_slot + 1);
4647                                 path->slots[0] = extent_slot;
4648                                 num_to_del = 2;
4649                         }
4650                 }
4651
4652                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4653                                       num_to_del);
4654                 BUG_ON(ret);
4655                 btrfs_release_path(extent_root, path);
4656
4657                 if (is_data) {
4658                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4659                         BUG_ON(ret);
4660                 } else {
4661                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4662                              bytenr >> PAGE_CACHE_SHIFT,
4663                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4664                 }
4665
4666                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4667                 BUG_ON(ret);
4668         }
4669         btrfs_free_path(path);
4670         return ret;
4671 }
4672
4673 /*
4674  * when we free an block, it is possible (and likely) that we free the last
4675  * delayed ref for that extent as well.  This searches the delayed ref tree for
4676  * a given extent, and if there are no other delayed refs to be processed, it
4677  * removes it from the tree.
4678  */
4679 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4680                                       struct btrfs_root *root, u64 bytenr)
4681 {
4682         struct btrfs_delayed_ref_head *head;
4683         struct btrfs_delayed_ref_root *delayed_refs;
4684         struct btrfs_delayed_ref_node *ref;
4685         struct rb_node *node;
4686         int ret = 0;
4687
4688         delayed_refs = &trans->transaction->delayed_refs;
4689         spin_lock(&delayed_refs->lock);
4690         head = btrfs_find_delayed_ref_head(trans, bytenr);
4691         if (!head)
4692                 goto out;
4693
4694         node = rb_prev(&head->node.rb_node);
4695         if (!node)
4696                 goto out;
4697
4698         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4699
4700         /* there are still entries for this ref, we can't drop it */
4701         if (ref->bytenr == bytenr)
4702                 goto out;
4703
4704         if (head->extent_op) {
4705                 if (!head->must_insert_reserved)
4706                         goto out;
4707                 kfree(head->extent_op);
4708                 head->extent_op = NULL;
4709         }
4710
4711         /*
4712          * waiting for the lock here would deadlock.  If someone else has it
4713          * locked they are already in the process of dropping it anyway
4714          */
4715         if (!mutex_trylock(&head->mutex))
4716                 goto out;
4717
4718         /*
4719          * at this point we have a head with no other entries.  Go
4720          * ahead and process it.
4721          */
4722         head->node.in_tree = 0;
4723         rb_erase(&head->node.rb_node, &delayed_refs->root);
4724
4725         delayed_refs->num_entries--;
4726
4727         /*
4728          * we don't take a ref on the node because we're removing it from the
4729          * tree, so we just steal the ref the tree was holding.
4730          */
4731         delayed_refs->num_heads--;
4732         if (list_empty(&head->cluster))
4733                 delayed_refs->num_heads_ready--;
4734
4735         list_del_init(&head->cluster);
4736         spin_unlock(&delayed_refs->lock);
4737
4738         BUG_ON(head->extent_op);
4739         if (head->must_insert_reserved)
4740                 ret = 1;
4741
4742         mutex_unlock(&head->mutex);
4743         btrfs_put_delayed_ref(&head->node);
4744         return ret;
4745 out:
4746         spin_unlock(&delayed_refs->lock);
4747         return 0;
4748 }
4749
4750 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4751                            struct btrfs_root *root,
4752                            struct extent_buffer *buf,
4753                            u64 parent, int last_ref)
4754 {
4755         struct btrfs_block_rsv *block_rsv;
4756         struct btrfs_block_group_cache *cache = NULL;
4757         int ret;
4758
4759         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4760                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4761                                                 parent, root->root_key.objectid,
4762                                                 btrfs_header_level(buf),
4763                                                 BTRFS_DROP_DELAYED_REF, NULL);
4764                 BUG_ON(ret);
4765         }
4766
4767         if (!last_ref)
4768                 return;
4769
4770         block_rsv = get_block_rsv(trans, root);
4771         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4772         if (block_rsv->space_info != cache->space_info)
4773                 goto out;
4774
4775         if (btrfs_header_generation(buf) == trans->transid) {
4776                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4777                         ret = check_ref_cleanup(trans, root, buf->start);
4778                         if (!ret)
4779                                 goto pin;
4780                 }
4781
4782                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4783                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4784                         goto pin;
4785                 }
4786
4787                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4788
4789                 btrfs_add_free_space(cache, buf->start, buf->len);
4790                 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
4791                 if (ret == -EAGAIN) {
4792                         /* block group became read-only */
4793                         btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
4794                         goto out;
4795                 }
4796
4797                 ret = 1;
4798                 spin_lock(&block_rsv->lock);
4799                 if (block_rsv->reserved < block_rsv->size) {
4800                         block_rsv->reserved += buf->len;
4801                         ret = 0;
4802                 }
4803                 spin_unlock(&block_rsv->lock);
4804
4805                 if (ret) {
4806                         spin_lock(&cache->space_info->lock);
4807                         cache->space_info->bytes_reserved -= buf->len;
4808                         cache->space_info->reservation_progress++;
4809                         spin_unlock(&cache->space_info->lock);
4810                 }
4811                 goto out;
4812         }
4813 pin:
4814         if (block_rsv->durable && !cache->ro) {
4815                 ret = 0;
4816                 spin_lock(&cache->lock);
4817                 if (!cache->ro) {
4818                         cache->reserved_pinned += buf->len;
4819                         ret = 1;
4820                 }
4821                 spin_unlock(&cache->lock);
4822
4823                 if (ret) {
4824                         spin_lock(&block_rsv->lock);
4825                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4826                         spin_unlock(&block_rsv->lock);
4827                 }
4828         }
4829 out:
4830         /*
4831          * Deleting the buffer, clear the corrupt flag since it doesn't matter
4832          * anymore.
4833          */
4834         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4835         btrfs_put_block_group(cache);
4836 }
4837
4838 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4839                       struct btrfs_root *root,
4840                       u64 bytenr, u64 num_bytes, u64 parent,
4841                       u64 root_objectid, u64 owner, u64 offset)
4842 {
4843         int ret;
4844
4845         /*
4846          * tree log blocks never actually go into the extent allocation
4847          * tree, just update pinning info and exit early.
4848          */
4849         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4850                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4851                 /* unlocks the pinned mutex */
4852                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4853                 ret = 0;
4854         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4855                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4856                                         parent, root_objectid, (int)owner,
4857                                         BTRFS_DROP_DELAYED_REF, NULL);
4858                 BUG_ON(ret);
4859         } else {
4860                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4861                                         parent, root_objectid, owner,
4862                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4863                 BUG_ON(ret);
4864         }
4865         return ret;
4866 }
4867
4868 static u64 stripe_align(struct btrfs_root *root, u64 val)
4869 {
4870         u64 mask = ((u64)root->stripesize - 1);
4871         u64 ret = (val + mask) & ~mask;
4872         return ret;
4873 }
4874
4875 /*
4876  * when we wait for progress in the block group caching, its because
4877  * our allocation attempt failed at least once.  So, we must sleep
4878  * and let some progress happen before we try again.
4879  *
4880  * This function will sleep at least once waiting for new free space to
4881  * show up, and then it will check the block group free space numbers
4882  * for our min num_bytes.  Another option is to have it go ahead
4883  * and look in the rbtree for a free extent of a given size, but this
4884  * is a good start.
4885  */
4886 static noinline int
4887 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4888                                 u64 num_bytes)
4889 {
4890         struct btrfs_caching_control *caching_ctl;
4891         DEFINE_WAIT(wait);
4892
4893         caching_ctl = get_caching_control(cache);
4894         if (!caching_ctl)
4895                 return 0;
4896
4897         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4898                    (cache->free_space_ctl->free_space >= num_bytes));
4899
4900         put_caching_control(caching_ctl);
4901         return 0;
4902 }
4903
4904 static noinline int
4905 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4906 {
4907         struct btrfs_caching_control *caching_ctl;
4908         DEFINE_WAIT(wait);
4909
4910         caching_ctl = get_caching_control(cache);
4911         if (!caching_ctl)
4912                 return 0;
4913
4914         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4915
4916         put_caching_control(caching_ctl);
4917         return 0;
4918 }
4919
4920 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4921 {
4922         int index;
4923         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4924                 index = 0;
4925         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4926                 index = 1;
4927         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4928                 index = 2;
4929         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4930                 index = 3;
4931         else
4932                 index = 4;
4933         return index;
4934 }
4935
4936 enum btrfs_loop_type {
4937         LOOP_FIND_IDEAL = 0,
4938         LOOP_CACHING_NOWAIT = 1,
4939         LOOP_CACHING_WAIT = 2,
4940         LOOP_ALLOC_CHUNK = 3,
4941         LOOP_NO_EMPTY_SIZE = 4,
4942 };
4943
4944 /*
4945  * walks the btree of allocated extents and find a hole of a given size.
4946  * The key ins is changed to record the hole:
4947  * ins->objectid == block start
4948  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4949  * ins->offset == number of blocks
4950  * Any available blocks before search_start are skipped.
4951  */
4952 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4953                                      struct btrfs_root *orig_root,
4954                                      u64 num_bytes, u64 empty_size,
4955                                      u64 search_start, u64 search_end,
4956                                      u64 hint_byte, struct btrfs_key *ins,
4957                                      int data)
4958 {
4959         int ret = 0;
4960         struct btrfs_root *root = orig_root->fs_info->extent_root;
4961         struct btrfs_free_cluster *last_ptr = NULL;
4962         struct btrfs_block_group_cache *block_group = NULL;
4963         int empty_cluster = 2 * 1024 * 1024;
4964         int allowed_chunk_alloc = 0;
4965         int done_chunk_alloc = 0;
4966         struct btrfs_space_info *space_info;
4967         int last_ptr_loop = 0;
4968         int loop = 0;
4969         int index = 0;
4970         bool found_uncached_bg = false;
4971         bool failed_cluster_refill = false;
4972         bool failed_alloc = false;
4973         bool use_cluster = true;
4974         u64 ideal_cache_percent = 0;
4975         u64 ideal_cache_offset = 0;
4976
4977         WARN_ON(num_bytes < root->sectorsize);
4978         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4979         ins->objectid = 0;
4980         ins->offset = 0;
4981
4982         space_info = __find_space_info(root->fs_info, data);
4983         if (!space_info) {
4984                 printk(KERN_ERR "No space info for %d\n", data);
4985                 return -ENOSPC;
4986         }
4987
4988         /*
4989          * If the space info is for both data and metadata it means we have a
4990          * small filesystem and we can't use the clustering stuff.
4991          */
4992         if (btrfs_mixed_space_info(space_info))
4993                 use_cluster = false;
4994
4995         if (orig_root->ref_cows || empty_size)
4996                 allowed_chunk_alloc = 1;
4997
4998         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4999                 last_ptr = &root->fs_info->meta_alloc_cluster;
5000                 if (!btrfs_test_opt(root, SSD))
5001                         empty_cluster = 64 * 1024;
5002         }
5003
5004         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5005             btrfs_test_opt(root, SSD)) {
5006                 last_ptr = &root->fs_info->data_alloc_cluster;
5007         }
5008
5009         if (last_ptr) {
5010                 spin_lock(&last_ptr->lock);
5011                 if (last_ptr->block_group)
5012                         hint_byte = last_ptr->window_start;
5013                 spin_unlock(&last_ptr->lock);
5014         }
5015
5016         search_start = max(search_start, first_logical_byte(root, 0));
5017         search_start = max(search_start, hint_byte);
5018
5019         if (!last_ptr)
5020                 empty_cluster = 0;
5021
5022         if (search_start == hint_byte) {
5023 ideal_cache:
5024                 block_group = btrfs_lookup_block_group(root->fs_info,
5025                                                        search_start);
5026                 /*
5027                  * we don't want to use the block group if it doesn't match our
5028                  * allocation bits, or if its not cached.
5029                  *
5030                  * However if we are re-searching with an ideal block group
5031                  * picked out then we don't care that the block group is cached.
5032                  */
5033                 if (block_group && block_group_bits(block_group, data) &&
5034                     (block_group->cached != BTRFS_CACHE_NO ||
5035                      search_start == ideal_cache_offset)) {
5036                         down_read(&space_info->groups_sem);
5037                         if (list_empty(&block_group->list) ||
5038                             block_group->ro) {
5039                                 /*
5040                                  * someone is removing this block group,
5041                                  * we can't jump into the have_block_group
5042                                  * target because our list pointers are not
5043                                  * valid
5044                                  */
5045                                 btrfs_put_block_group(block_group);
5046                                 up_read(&space_info->groups_sem);
5047                         } else {
5048                                 index = get_block_group_index(block_group);
5049                                 goto have_block_group;
5050                         }
5051                 } else if (block_group) {
5052                         btrfs_put_block_group(block_group);
5053                 }
5054         }
5055 search:
5056         down_read(&space_info->groups_sem);
5057         list_for_each_entry(block_group, &space_info->block_groups[index],
5058                             list) {
5059                 u64 offset;
5060                 int cached;
5061
5062                 btrfs_get_block_group(block_group);
5063                 search_start = block_group->key.objectid;
5064
5065                 /*
5066                  * this can happen if we end up cycling through all the
5067                  * raid types, but we want to make sure we only allocate
5068                  * for the proper type.
5069                  */
5070                 if (!block_group_bits(block_group, data)) {
5071                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5072                                 BTRFS_BLOCK_GROUP_RAID1 |
5073                                 BTRFS_BLOCK_GROUP_RAID10;
5074
5075                         /*
5076                          * if they asked for extra copies and this block group
5077                          * doesn't provide them, bail.  This does allow us to
5078                          * fill raid0 from raid1.
5079                          */
5080                         if ((data & extra) && !(block_group->flags & extra))
5081                                 goto loop;
5082                 }
5083
5084 have_block_group:
5085                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5086                         u64 free_percent;
5087
5088                         ret = cache_block_group(block_group, trans,
5089                                                 orig_root, 1);
5090                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5091                                 goto have_block_group;
5092
5093                         free_percent = btrfs_block_group_used(&block_group->item);
5094                         free_percent *= 100;
5095                         free_percent = div64_u64(free_percent,
5096                                                  block_group->key.offset);
5097                         free_percent = 100 - free_percent;
5098                         if (free_percent > ideal_cache_percent &&
5099                             likely(!block_group->ro)) {
5100                                 ideal_cache_offset = block_group->key.objectid;
5101                                 ideal_cache_percent = free_percent;
5102                         }
5103
5104                         /*
5105                          * We only want to start kthread caching if we are at
5106                          * the point where we will wait for caching to make
5107                          * progress, or if our ideal search is over and we've
5108                          * found somebody to start caching.
5109                          */
5110                         if (loop > LOOP_CACHING_NOWAIT ||
5111                             (loop > LOOP_FIND_IDEAL &&
5112                              atomic_read(&space_info->caching_threads) < 2)) {
5113                                 ret = cache_block_group(block_group, trans,
5114                                                         orig_root, 0);
5115                                 BUG_ON(ret);
5116                         }
5117                         found_uncached_bg = true;
5118
5119                         /*
5120                          * If loop is set for cached only, try the next block
5121                          * group.
5122                          */
5123                         if (loop == LOOP_FIND_IDEAL)
5124                                 goto loop;
5125                 }
5126
5127                 cached = block_group_cache_done(block_group);
5128                 if (unlikely(!cached))
5129                         found_uncached_bg = true;
5130
5131                 if (unlikely(block_group->ro))
5132                         goto loop;
5133
5134                 /*
5135                  * Ok we want to try and use the cluster allocator, so lets look
5136                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5137                  * have tried the cluster allocator plenty of times at this
5138                  * point and not have found anything, so we are likely way too
5139                  * fragmented for the clustering stuff to find anything, so lets
5140                  * just skip it and let the allocator find whatever block it can
5141                  * find
5142                  */
5143                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5144                         /*
5145                          * the refill lock keeps out other
5146                          * people trying to start a new cluster
5147                          */
5148                         spin_lock(&last_ptr->refill_lock);
5149                         if (last_ptr->block_group &&
5150                             (last_ptr->block_group->ro ||
5151                             !block_group_bits(last_ptr->block_group, data))) {
5152                                 offset = 0;
5153                                 goto refill_cluster;
5154                         }
5155
5156                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5157                                                  num_bytes, search_start);
5158                         if (offset) {
5159                                 /* we have a block, we're done */
5160                                 spin_unlock(&last_ptr->refill_lock);
5161                                 goto checks;
5162                         }
5163
5164                         spin_lock(&last_ptr->lock);
5165                         /*
5166                          * whoops, this cluster doesn't actually point to
5167                          * this block group.  Get a ref on the block
5168                          * group is does point to and try again
5169                          */
5170                         if (!last_ptr_loop && last_ptr->block_group &&
5171                             last_ptr->block_group != block_group) {
5172
5173                                 btrfs_put_block_group(block_group);
5174                                 block_group = last_ptr->block_group;
5175                                 btrfs_get_block_group(block_group);
5176                                 spin_unlock(&last_ptr->lock);
5177                                 spin_unlock(&last_ptr->refill_lock);
5178
5179                                 last_ptr_loop = 1;
5180                                 search_start = block_group->key.objectid;
5181                                 /*
5182                                  * we know this block group is properly
5183                                  * in the list because
5184                                  * btrfs_remove_block_group, drops the
5185                                  * cluster before it removes the block
5186                                  * group from the list
5187                                  */
5188                                 goto have_block_group;
5189                         }
5190                         spin_unlock(&last_ptr->lock);
5191 refill_cluster:
5192                         /*
5193                          * this cluster didn't work out, free it and
5194                          * start over
5195                          */
5196                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5197
5198                         last_ptr_loop = 0;
5199
5200                         /* allocate a cluster in this block group */
5201                         ret = btrfs_find_space_cluster(trans, root,
5202                                                block_group, last_ptr,
5203                                                offset, num_bytes,
5204                                                empty_cluster + empty_size);
5205                         if (ret == 0) {
5206                                 /*
5207                                  * now pull our allocation out of this
5208                                  * cluster
5209                                  */
5210                                 offset = btrfs_alloc_from_cluster(block_group,
5211                                                   last_ptr, num_bytes,
5212                                                   search_start);
5213                                 if (offset) {
5214                                         /* we found one, proceed */
5215                                         spin_unlock(&last_ptr->refill_lock);
5216                                         goto checks;
5217                                 }
5218                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5219                                    && !failed_cluster_refill) {
5220                                 spin_unlock(&last_ptr->refill_lock);
5221
5222                                 failed_cluster_refill = true;
5223                                 wait_block_group_cache_progress(block_group,
5224                                        num_bytes + empty_cluster + empty_size);
5225                                 goto have_block_group;
5226                         }
5227
5228                         /*
5229                          * at this point we either didn't find a cluster
5230                          * or we weren't able to allocate a block from our
5231                          * cluster.  Free the cluster we've been trying
5232                          * to use, and go to the next block group
5233                          */
5234                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5235                         spin_unlock(&last_ptr->refill_lock);
5236                         goto loop;
5237                 }
5238
5239                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5240                                                     num_bytes, empty_size);
5241                 /*
5242                  * If we didn't find a chunk, and we haven't failed on this
5243                  * block group before, and this block group is in the middle of
5244                  * caching and we are ok with waiting, then go ahead and wait
5245                  * for progress to be made, and set failed_alloc to true.
5246                  *
5247                  * If failed_alloc is true then we've already waited on this
5248                  * block group once and should move on to the next block group.
5249                  */
5250                 if (!offset && !failed_alloc && !cached &&
5251                     loop > LOOP_CACHING_NOWAIT) {
5252                         wait_block_group_cache_progress(block_group,
5253                                                 num_bytes + empty_size);
5254                         failed_alloc = true;
5255                         goto have_block_group;
5256                 } else if (!offset) {
5257                         goto loop;
5258                 }
5259 checks:
5260                 search_start = stripe_align(root, offset);
5261                 /* move on to the next group */
5262                 if (search_start + num_bytes >= search_end) {
5263                         btrfs_add_free_space(block_group, offset, num_bytes);
5264                         goto loop;
5265                 }
5266
5267                 /* move on to the next group */
5268                 if (search_start + num_bytes >
5269                     block_group->key.objectid + block_group->key.offset) {
5270                         btrfs_add_free_space(block_group, offset, num_bytes);
5271                         goto loop;
5272                 }
5273
5274                 ins->objectid = search_start;
5275                 ins->offset = num_bytes;
5276
5277                 if (offset < search_start)
5278                         btrfs_add_free_space(block_group, offset,
5279                                              search_start - offset);
5280                 BUG_ON(offset > search_start);
5281
5282                 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
5283                                             (data & BTRFS_BLOCK_GROUP_DATA));
5284                 if (ret == -EAGAIN) {
5285                         btrfs_add_free_space(block_group, offset, num_bytes);
5286                         goto loop;
5287                 }
5288
5289                 /* we are all good, lets return */
5290                 ins->objectid = search_start;
5291                 ins->offset = num_bytes;
5292
5293                 if (offset < search_start)
5294                         btrfs_add_free_space(block_group, offset,
5295                                              search_start - offset);
5296                 BUG_ON(offset > search_start);
5297                 break;
5298 loop:
5299                 failed_cluster_refill = false;
5300                 failed_alloc = false;
5301                 BUG_ON(index != get_block_group_index(block_group));
5302                 btrfs_put_block_group(block_group);
5303         }
5304         up_read(&space_info->groups_sem);
5305
5306         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5307                 goto search;
5308
5309         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5310          *                      for them to make caching progress.  Also
5311          *                      determine the best possible bg to cache
5312          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5313          *                      caching kthreads as we move along
5314          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5315          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5316          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5317          *                      again
5318          */
5319         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5320             (found_uncached_bg || empty_size || empty_cluster ||
5321              allowed_chunk_alloc)) {
5322                 index = 0;
5323                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5324                         found_uncached_bg = false;
5325                         loop++;
5326                         if (!ideal_cache_percent &&
5327                             atomic_read(&space_info->caching_threads))
5328                                 goto search;
5329
5330                         /*
5331                          * 1 of the following 2 things have happened so far
5332                          *
5333                          * 1) We found an ideal block group for caching that
5334                          * is mostly full and will cache quickly, so we might
5335                          * as well wait for it.
5336                          *
5337                          * 2) We searched for cached only and we didn't find
5338                          * anything, and we didn't start any caching kthreads
5339                          * either, so chances are we will loop through and
5340                          * start a couple caching kthreads, and then come back
5341                          * around and just wait for them.  This will be slower
5342                          * because we will have 2 caching kthreads reading at
5343                          * the same time when we could have just started one
5344                          * and waited for it to get far enough to give us an
5345                          * allocation, so go ahead and go to the wait caching
5346                          * loop.
5347                          */
5348                         loop = LOOP_CACHING_WAIT;
5349                         search_start = ideal_cache_offset;
5350                         ideal_cache_percent = 0;
5351                         goto ideal_cache;
5352                 } else if (loop == LOOP_FIND_IDEAL) {
5353                         /*
5354                          * Didn't find a uncached bg, wait on anything we find
5355                          * next.
5356                          */
5357                         loop = LOOP_CACHING_WAIT;
5358                         goto search;
5359                 }
5360
5361                 if (loop < LOOP_CACHING_WAIT) {
5362                         loop++;
5363                         goto search;
5364                 }
5365
5366                 if (loop == LOOP_ALLOC_CHUNK) {
5367                         empty_size = 0;
5368                         empty_cluster = 0;
5369                 }
5370
5371                 if (allowed_chunk_alloc) {
5372                         ret = do_chunk_alloc(trans, root, num_bytes +
5373                                              2 * 1024 * 1024, data,
5374                                              CHUNK_ALLOC_LIMITED);
5375                         allowed_chunk_alloc = 0;
5376                         done_chunk_alloc = 1;
5377                 } else if (!done_chunk_alloc &&
5378                            space_info->force_alloc == CHUNK_ALLOC_NO_FORCE) {
5379                         space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5380                 }
5381
5382                 if (loop < LOOP_NO_EMPTY_SIZE) {
5383                         loop++;
5384                         goto search;
5385                 }
5386                 ret = -ENOSPC;
5387         } else if (!ins->objectid) {
5388                 ret = -ENOSPC;
5389         }
5390
5391         /* we found what we needed */
5392         if (ins->objectid) {
5393                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5394                         trans->block_group = block_group->key.objectid;
5395
5396                 btrfs_put_block_group(block_group);
5397                 ret = 0;
5398         }
5399
5400         return ret;
5401 }
5402
5403 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5404                             int dump_block_groups)
5405 {
5406         struct btrfs_block_group_cache *cache;
5407         int index = 0;
5408
5409         spin_lock(&info->lock);
5410         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5411                (unsigned long long)(info->total_bytes - info->bytes_used -
5412                                     info->bytes_pinned - info->bytes_reserved -
5413                                     info->bytes_readonly),
5414                (info->full) ? "" : "not ");
5415         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5416                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5417                (unsigned long long)info->total_bytes,
5418                (unsigned long long)info->bytes_used,
5419                (unsigned long long)info->bytes_pinned,
5420                (unsigned long long)info->bytes_reserved,
5421                (unsigned long long)info->bytes_may_use,
5422                (unsigned long long)info->bytes_readonly);
5423         spin_unlock(&info->lock);
5424
5425         if (!dump_block_groups)
5426                 return;
5427
5428         down_read(&info->groups_sem);
5429 again:
5430         list_for_each_entry(cache, &info->block_groups[index], list) {
5431                 spin_lock(&cache->lock);
5432                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5433                        "%llu pinned %llu reserved\n",
5434                        (unsigned long long)cache->key.objectid,
5435                        (unsigned long long)cache->key.offset,
5436                        (unsigned long long)btrfs_block_group_used(&cache->item),
5437                        (unsigned long long)cache->pinned,
5438                        (unsigned long long)cache->reserved);
5439                 btrfs_dump_free_space(cache, bytes);
5440                 spin_unlock(&cache->lock);
5441         }
5442         if (++index < BTRFS_NR_RAID_TYPES)
5443                 goto again;
5444         up_read(&info->groups_sem);
5445 }
5446
5447 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5448                          struct btrfs_root *root,
5449                          u64 num_bytes, u64 min_alloc_size,
5450                          u64 empty_size, u64 hint_byte,
5451                          u64 search_end, struct btrfs_key *ins,
5452                          u64 data)
5453 {
5454         int ret;
5455         u64 search_start = 0;
5456
5457         data = btrfs_get_alloc_profile(root, data);
5458 again:
5459         /*
5460          * the only place that sets empty_size is btrfs_realloc_node, which
5461          * is not called recursively on allocations
5462          */
5463         if (empty_size || root->ref_cows)
5464                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5465                                      num_bytes + 2 * 1024 * 1024, data,
5466                                      CHUNK_ALLOC_NO_FORCE);
5467
5468         WARN_ON(num_bytes < root->sectorsize);
5469         ret = find_free_extent(trans, root, num_bytes, empty_size,
5470                                search_start, search_end, hint_byte,
5471                                ins, data);
5472
5473         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5474                 num_bytes = num_bytes >> 1;
5475                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5476                 num_bytes = max(num_bytes, min_alloc_size);
5477                 do_chunk_alloc(trans, root->fs_info->extent_root,
5478                                num_bytes, data, CHUNK_ALLOC_FORCE);
5479                 goto again;
5480         }
5481         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5482                 struct btrfs_space_info *sinfo;
5483
5484                 sinfo = __find_space_info(root->fs_info, data);
5485                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5486                        "wanted %llu\n", (unsigned long long)data,
5487                        (unsigned long long)num_bytes);
5488                 dump_space_info(sinfo, num_bytes, 1);
5489         }
5490
5491         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5492
5493         return ret;
5494 }
5495
5496 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5497 {
5498         struct btrfs_block_group_cache *cache;
5499         int ret = 0;
5500
5501         cache = btrfs_lookup_block_group(root->fs_info, start);
5502         if (!cache) {
5503                 printk(KERN_ERR "Unable to find block group for %llu\n",
5504                        (unsigned long long)start);
5505                 return -ENOSPC;
5506         }
5507
5508         if (btrfs_test_opt(root, DISCARD))
5509                 ret = btrfs_discard_extent(root, start, len, NULL);
5510
5511         btrfs_add_free_space(cache, start, len);
5512         btrfs_update_reserved_bytes(cache, len, 0, 1);
5513         btrfs_put_block_group(cache);
5514
5515         trace_btrfs_reserved_extent_free(root, start, len);
5516
5517         return ret;
5518 }
5519
5520 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5521                                       struct btrfs_root *root,
5522                                       u64 parent, u64 root_objectid,
5523                                       u64 flags, u64 owner, u64 offset,
5524                                       struct btrfs_key *ins, int ref_mod)
5525 {
5526         int ret;
5527         struct btrfs_fs_info *fs_info = root->fs_info;
5528         struct btrfs_extent_item *extent_item;
5529         struct btrfs_extent_inline_ref *iref;
5530         struct btrfs_path *path;
5531         struct extent_buffer *leaf;
5532         int type;
5533         u32 size;
5534
5535         if (parent > 0)
5536                 type = BTRFS_SHARED_DATA_REF_KEY;
5537         else
5538                 type = BTRFS_EXTENT_DATA_REF_KEY;
5539
5540         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5541
5542         path = btrfs_alloc_path();
5543         if (!path)
5544                 return -ENOMEM;
5545
5546         path->leave_spinning = 1;
5547         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5548                                       ins, size);
5549         BUG_ON(ret);
5550
5551         leaf = path->nodes[0];
5552         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5553                                      struct btrfs_extent_item);
5554         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5555         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5556         btrfs_set_extent_flags(leaf, extent_item,
5557                                flags | BTRFS_EXTENT_FLAG_DATA);
5558
5559         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5560         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5561         if (parent > 0) {
5562                 struct btrfs_shared_data_ref *ref;
5563                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5564                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5565                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5566         } else {
5567                 struct btrfs_extent_data_ref *ref;
5568                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5569                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5570                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5571                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5572                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5573         }
5574
5575         btrfs_mark_buffer_dirty(path->nodes[0]);
5576         btrfs_free_path(path);
5577
5578         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5579         if (ret) {
5580                 printk(KERN_ERR "btrfs update block group failed for %llu "
5581                        "%llu\n", (unsigned long long)ins->objectid,
5582                        (unsigned long long)ins->offset);
5583                 BUG();
5584         }
5585         return ret;
5586 }
5587
5588 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5589                                      struct btrfs_root *root,
5590                                      u64 parent, u64 root_objectid,
5591                                      u64 flags, struct btrfs_disk_key *key,
5592                                      int level, struct btrfs_key *ins)
5593 {
5594         int ret;
5595         struct btrfs_fs_info *fs_info = root->fs_info;
5596         struct btrfs_extent_item *extent_item;
5597         struct btrfs_tree_block_info *block_info;
5598         struct btrfs_extent_inline_ref *iref;
5599         struct btrfs_path *path;
5600         struct extent_buffer *leaf;
5601         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5602
5603         path = btrfs_alloc_path();
5604         BUG_ON(!path);
5605
5606         path->leave_spinning = 1;
5607         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5608                                       ins, size);
5609         BUG_ON(ret);
5610
5611         leaf = path->nodes[0];
5612         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5613                                      struct btrfs_extent_item);
5614         btrfs_set_extent_refs(leaf, extent_item, 1);
5615         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5616         btrfs_set_extent_flags(leaf, extent_item,
5617                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5618         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5619
5620         btrfs_set_tree_block_key(leaf, block_info, key);
5621         btrfs_set_tree_block_level(leaf, block_info, level);
5622
5623         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5624         if (parent > 0) {
5625                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5626                 btrfs_set_extent_inline_ref_type(leaf, iref,
5627                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5628                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5629         } else {
5630                 btrfs_set_extent_inline_ref_type(leaf, iref,
5631                                                  BTRFS_TREE_BLOCK_REF_KEY);
5632                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5633         }
5634
5635         btrfs_mark_buffer_dirty(leaf);
5636         btrfs_free_path(path);
5637
5638         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5639         if (ret) {
5640                 printk(KERN_ERR "btrfs update block group failed for %llu "
5641                        "%llu\n", (unsigned long long)ins->objectid,
5642                        (unsigned long long)ins->offset);
5643                 BUG();
5644         }
5645         return ret;
5646 }
5647
5648 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5649                                      struct btrfs_root *root,
5650                                      u64 root_objectid, u64 owner,
5651                                      u64 offset, struct btrfs_key *ins)
5652 {
5653         int ret;
5654
5655         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5656
5657         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5658                                          0, root_objectid, owner, offset,
5659                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5660         return ret;
5661 }
5662
5663 /*
5664  * this is used by the tree logging recovery code.  It records that
5665  * an extent has been allocated and makes sure to clear the free
5666  * space cache bits as well
5667  */
5668 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5669                                    struct btrfs_root *root,
5670                                    u64 root_objectid, u64 owner, u64 offset,
5671                                    struct btrfs_key *ins)
5672 {
5673         int ret;
5674         struct btrfs_block_group_cache *block_group;
5675         struct btrfs_caching_control *caching_ctl;
5676         u64 start = ins->objectid;
5677         u64 num_bytes = ins->offset;
5678
5679         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5680         cache_block_group(block_group, trans, NULL, 0);
5681         caching_ctl = get_caching_control(block_group);
5682
5683         if (!caching_ctl) {
5684                 BUG_ON(!block_group_cache_done(block_group));
5685                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5686                 BUG_ON(ret);
5687         } else {
5688                 mutex_lock(&caching_ctl->mutex);
5689
5690                 if (start >= caching_ctl->progress) {
5691                         ret = add_excluded_extent(root, start, num_bytes);
5692                         BUG_ON(ret);
5693                 } else if (start + num_bytes <= caching_ctl->progress) {
5694                         ret = btrfs_remove_free_space(block_group,
5695                                                       start, num_bytes);
5696                         BUG_ON(ret);
5697                 } else {
5698                         num_bytes = caching_ctl->progress - start;
5699                         ret = btrfs_remove_free_space(block_group,
5700                                                       start, num_bytes);
5701                         BUG_ON(ret);
5702
5703                         start = caching_ctl->progress;
5704                         num_bytes = ins->objectid + ins->offset -
5705                                     caching_ctl->progress;
5706                         ret = add_excluded_extent(root, start, num_bytes);
5707                         BUG_ON(ret);
5708                 }
5709
5710                 mutex_unlock(&caching_ctl->mutex);
5711                 put_caching_control(caching_ctl);
5712         }
5713
5714         ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
5715         BUG_ON(ret);
5716         btrfs_put_block_group(block_group);
5717         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5718                                          0, owner, offset, ins, 1);
5719         return ret;
5720 }
5721
5722 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5723                                             struct btrfs_root *root,
5724                                             u64 bytenr, u32 blocksize,
5725                                             int level)
5726 {
5727         struct extent_buffer *buf;
5728
5729         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5730         if (!buf)
5731                 return ERR_PTR(-ENOMEM);
5732         btrfs_set_header_generation(buf, trans->transid);
5733         btrfs_set_buffer_lockdep_class(buf, level);
5734         btrfs_tree_lock(buf);
5735         clean_tree_block(trans, root, buf);
5736
5737         btrfs_set_lock_blocking(buf);
5738         btrfs_set_buffer_uptodate(buf);
5739
5740         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5741                 /*
5742                  * we allow two log transactions at a time, use different
5743                  * EXENT bit to differentiate dirty pages.
5744                  */
5745                 if (root->log_transid % 2 == 0)
5746                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5747                                         buf->start + buf->len - 1, GFP_NOFS);
5748                 else
5749                         set_extent_new(&root->dirty_log_pages, buf->start,
5750                                         buf->start + buf->len - 1, GFP_NOFS);
5751         } else {
5752                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5753                          buf->start + buf->len - 1, GFP_NOFS);
5754         }
5755         trans->blocks_used++;
5756         /* this returns a buffer locked for blocking */
5757         return buf;
5758 }
5759
5760 static struct btrfs_block_rsv *
5761 use_block_rsv(struct btrfs_trans_handle *trans,
5762               struct btrfs_root *root, u32 blocksize)
5763 {
5764         struct btrfs_block_rsv *block_rsv;
5765         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5766         int ret;
5767
5768         block_rsv = get_block_rsv(trans, root);
5769
5770         if (block_rsv->size == 0) {
5771                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5772                                              blocksize, 0);
5773                 /*
5774                  * If we couldn't reserve metadata bytes try and use some from
5775                  * the global reserve.
5776                  */
5777                 if (ret && block_rsv != global_rsv) {
5778                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5779                         if (!ret)
5780                                 return global_rsv;
5781                         return ERR_PTR(ret);
5782                 } else if (ret) {
5783                         return ERR_PTR(ret);
5784                 }
5785                 return block_rsv;
5786         }
5787
5788         ret = block_rsv_use_bytes(block_rsv, blocksize);
5789         if (!ret)
5790                 return block_rsv;
5791         if (ret) {
5792                 WARN_ON(1);
5793                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5794                                              0);
5795                 if (!ret) {
5796                         spin_lock(&block_rsv->lock);
5797                         block_rsv->size += blocksize;
5798                         spin_unlock(&block_rsv->lock);
5799                         return block_rsv;
5800                 } else if (ret && block_rsv != global_rsv) {
5801                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5802                         if (!ret)
5803                                 return global_rsv;
5804                 }
5805         }
5806
5807         return ERR_PTR(-ENOSPC);
5808 }
5809
5810 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5811 {
5812         block_rsv_add_bytes(block_rsv, blocksize, 0);
5813         block_rsv_release_bytes(block_rsv, NULL, 0);
5814 }
5815
5816 /*
5817  * finds a free extent and does all the dirty work required for allocation
5818  * returns the key for the extent through ins, and a tree buffer for
5819  * the first block of the extent through buf.
5820  *
5821  * returns the tree buffer or NULL.
5822  */
5823 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5824                                         struct btrfs_root *root, u32 blocksize,
5825                                         u64 parent, u64 root_objectid,
5826                                         struct btrfs_disk_key *key, int level,
5827                                         u64 hint, u64 empty_size)
5828 {
5829         struct btrfs_key ins;
5830         struct btrfs_block_rsv *block_rsv;
5831         struct extent_buffer *buf;
5832         u64 flags = 0;
5833         int ret;
5834
5835
5836         block_rsv = use_block_rsv(trans, root, blocksize);
5837         if (IS_ERR(block_rsv))
5838                 return ERR_CAST(block_rsv);
5839
5840         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5841                                    empty_size, hint, (u64)-1, &ins, 0);
5842         if (ret) {
5843                 unuse_block_rsv(block_rsv, blocksize);
5844                 return ERR_PTR(ret);
5845         }
5846
5847         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5848                                     blocksize, level);
5849         BUG_ON(IS_ERR(buf));
5850
5851         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5852                 if (parent == 0)
5853                         parent = ins.objectid;
5854                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5855         } else
5856                 BUG_ON(parent > 0);
5857
5858         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5859                 struct btrfs_delayed_extent_op *extent_op;
5860                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5861                 BUG_ON(!extent_op);
5862                 if (key)
5863                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5864                 else
5865                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5866                 extent_op->flags_to_set = flags;
5867                 extent_op->update_key = 1;
5868                 extent_op->update_flags = 1;
5869                 extent_op->is_data = 0;
5870
5871                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5872                                         ins.offset, parent, root_objectid,
5873                                         level, BTRFS_ADD_DELAYED_EXTENT,
5874                                         extent_op);
5875                 BUG_ON(ret);
5876         }
5877         return buf;
5878 }
5879
5880 struct walk_control {
5881         u64 refs[BTRFS_MAX_LEVEL];
5882         u64 flags[BTRFS_MAX_LEVEL];
5883         struct btrfs_key update_progress;
5884         int stage;
5885         int level;
5886         int shared_level;
5887         int update_ref;
5888         int keep_locks;
5889         int reada_slot;
5890         int reada_count;
5891 };
5892
5893 #define DROP_REFERENCE  1
5894 #define UPDATE_BACKREF  2
5895
5896 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5897                                      struct btrfs_root *root,
5898                                      struct walk_control *wc,
5899                                      struct btrfs_path *path)
5900 {
5901         u64 bytenr;
5902         u64 generation;
5903         u64 refs;
5904         u64 flags;
5905         u32 nritems;
5906         u32 blocksize;
5907         struct btrfs_key key;
5908         struct extent_buffer *eb;
5909         int ret;
5910         int slot;
5911         int nread = 0;
5912
5913         if (path->slots[wc->level] < wc->reada_slot) {
5914                 wc->reada_count = wc->reada_count * 2 / 3;
5915                 wc->reada_count = max(wc->reada_count, 2);
5916         } else {
5917                 wc->reada_count = wc->reada_count * 3 / 2;
5918                 wc->reada_count = min_t(int, wc->reada_count,
5919                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5920         }
5921
5922         eb = path->nodes[wc->level];
5923         nritems = btrfs_header_nritems(eb);
5924         blocksize = btrfs_level_size(root, wc->level - 1);
5925
5926         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5927                 if (nread >= wc->reada_count)
5928                         break;
5929
5930                 cond_resched();
5931                 bytenr = btrfs_node_blockptr(eb, slot);
5932                 generation = btrfs_node_ptr_generation(eb, slot);
5933
5934                 if (slot == path->slots[wc->level])
5935                         goto reada;
5936
5937                 if (wc->stage == UPDATE_BACKREF &&
5938                     generation <= root->root_key.offset)
5939                         continue;
5940
5941                 /* We don't lock the tree block, it's OK to be racy here */
5942                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5943                                                &refs, &flags);
5944                 BUG_ON(ret);
5945                 BUG_ON(refs == 0);
5946
5947                 if (wc->stage == DROP_REFERENCE) {
5948                         if (refs == 1)
5949                                 goto reada;
5950
5951                         if (wc->level == 1 &&
5952                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5953                                 continue;
5954                         if (!wc->update_ref ||
5955                             generation <= root->root_key.offset)
5956                                 continue;
5957                         btrfs_node_key_to_cpu(eb, &key, slot);
5958                         ret = btrfs_comp_cpu_keys(&key,
5959                                                   &wc->update_progress);
5960                         if (ret < 0)
5961                                 continue;
5962                 } else {
5963                         if (wc->level == 1 &&
5964                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5965                                 continue;
5966                 }
5967 reada:
5968                 ret = readahead_tree_block(root, bytenr, blocksize,
5969                                            generation);
5970                 if (ret)
5971                         break;
5972                 nread++;
5973         }
5974         wc->reada_slot = slot;
5975 }
5976
5977 /*
5978  * hepler to process tree block while walking down the tree.
5979  *
5980  * when wc->stage == UPDATE_BACKREF, this function updates
5981  * back refs for pointers in the block.
5982  *
5983  * NOTE: return value 1 means we should stop walking down.
5984  */
5985 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5986                                    struct btrfs_root *root,
5987                                    struct btrfs_path *path,
5988                                    struct walk_control *wc, int lookup_info)
5989 {
5990         int level = wc->level;
5991         struct extent_buffer *eb = path->nodes[level];
5992         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5993         int ret;
5994
5995         if (wc->stage == UPDATE_BACKREF &&
5996             btrfs_header_owner(eb) != root->root_key.objectid)
5997                 return 1;
5998
5999         /*
6000          * when reference count of tree block is 1, it won't increase
6001          * again. once full backref flag is set, we never clear it.
6002          */
6003         if (lookup_info &&
6004             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6005              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6006                 BUG_ON(!path->locks[level]);
6007                 ret = btrfs_lookup_extent_info(trans, root,
6008                                                eb->start, eb->len,
6009                                                &wc->refs[level],
6010                                                &wc->flags[level]);
6011                 BUG_ON(ret);
6012                 BUG_ON(wc->refs[level] == 0);
6013         }
6014
6015         if (wc->stage == DROP_REFERENCE) {
6016                 if (wc->refs[level] > 1)
6017                         return 1;
6018
6019                 if (path->locks[level] && !wc->keep_locks) {
6020                         btrfs_tree_unlock(eb);
6021                         path->locks[level] = 0;
6022                 }
6023                 return 0;
6024         }
6025
6026         /* wc->stage == UPDATE_BACKREF */
6027         if (!(wc->flags[level] & flag)) {
6028                 BUG_ON(!path->locks[level]);
6029                 ret = btrfs_inc_ref(trans, root, eb, 1);
6030                 BUG_ON(ret);
6031                 ret = btrfs_dec_ref(trans, root, eb, 0);
6032                 BUG_ON(ret);
6033                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6034                                                   eb->len, flag, 0);
6035                 BUG_ON(ret);
6036                 wc->flags[level] |= flag;
6037         }
6038
6039         /*
6040          * the block is shared by multiple trees, so it's not good to
6041          * keep the tree lock
6042          */
6043         if (path->locks[level] && level > 0) {
6044                 btrfs_tree_unlock(eb);
6045                 path->locks[level] = 0;
6046         }
6047         return 0;
6048 }
6049
6050 /*
6051  * hepler to process tree block pointer.
6052  *
6053  * when wc->stage == DROP_REFERENCE, this function checks
6054  * reference count of the block pointed to. if the block
6055  * is shared and we need update back refs for the subtree
6056  * rooted at the block, this function changes wc->stage to
6057  * UPDATE_BACKREF. if the block is shared and there is no
6058  * need to update back, this function drops the reference
6059  * to the block.
6060  *
6061  * NOTE: return value 1 means we should stop walking down.
6062  */
6063 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6064                                  struct btrfs_root *root,
6065                                  struct btrfs_path *path,
6066                                  struct walk_control *wc, int *lookup_info)
6067 {
6068         u64 bytenr;
6069         u64 generation;
6070         u64 parent;
6071         u32 blocksize;
6072         struct btrfs_key key;
6073         struct extent_buffer *next;
6074         int level = wc->level;
6075         int reada = 0;
6076         int ret = 0;
6077
6078         generation = btrfs_node_ptr_generation(path->nodes[level],
6079                                                path->slots[level]);
6080         /*
6081          * if the lower level block was created before the snapshot
6082          * was created, we know there is no need to update back refs
6083          * for the subtree
6084          */
6085         if (wc->stage == UPDATE_BACKREF &&
6086             generation <= root->root_key.offset) {
6087                 *lookup_info = 1;
6088                 return 1;
6089         }
6090
6091         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6092         blocksize = btrfs_level_size(root, level - 1);
6093
6094         next = btrfs_find_tree_block(root, bytenr, blocksize);
6095         if (!next) {
6096                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6097                 if (!next)
6098                         return -ENOMEM;
6099                 reada = 1;
6100         }
6101         btrfs_tree_lock(next);
6102         btrfs_set_lock_blocking(next);
6103
6104         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6105                                        &wc->refs[level - 1],
6106                                        &wc->flags[level - 1]);
6107         BUG_ON(ret);
6108         BUG_ON(wc->refs[level - 1] == 0);
6109         *lookup_info = 0;
6110
6111         if (wc->stage == DROP_REFERENCE) {
6112                 if (wc->refs[level - 1] > 1) {
6113                         if (level == 1 &&
6114                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6115                                 goto skip;
6116
6117                         if (!wc->update_ref ||
6118                             generation <= root->root_key.offset)
6119                                 goto skip;
6120
6121                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6122                                               path->slots[level]);
6123                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6124                         if (ret < 0)
6125                                 goto skip;
6126
6127                         wc->stage = UPDATE_BACKREF;
6128                         wc->shared_level = level - 1;
6129                 }
6130         } else {
6131                 if (level == 1 &&
6132                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6133                         goto skip;
6134         }
6135
6136         if (!btrfs_buffer_uptodate(next, generation)) {
6137                 btrfs_tree_unlock(next);
6138                 free_extent_buffer(next);
6139                 next = NULL;
6140                 *lookup_info = 1;
6141         }
6142
6143         if (!next) {
6144                 if (reada && level == 1)
6145                         reada_walk_down(trans, root, wc, path);
6146                 next = read_tree_block(root, bytenr, blocksize, generation);
6147                 if (!next)
6148                         return -EIO;
6149                 btrfs_tree_lock(next);
6150                 btrfs_set_lock_blocking(next);
6151         }
6152
6153         level--;
6154         BUG_ON(level != btrfs_header_level(next));
6155         path->nodes[level] = next;
6156         path->slots[level] = 0;
6157         path->locks[level] = 1;
6158         wc->level = level;
6159         if (wc->level == 1)
6160                 wc->reada_slot = 0;
6161         return 0;
6162 skip:
6163         wc->refs[level - 1] = 0;
6164         wc->flags[level - 1] = 0;
6165         if (wc->stage == DROP_REFERENCE) {
6166                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6167                         parent = path->nodes[level]->start;
6168                 } else {
6169                         BUG_ON(root->root_key.objectid !=
6170                                btrfs_header_owner(path->nodes[level]));
6171                         parent = 0;
6172                 }
6173
6174                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6175                                         root->root_key.objectid, level - 1, 0);
6176                 BUG_ON(ret);
6177         }
6178         btrfs_tree_unlock(next);
6179         free_extent_buffer(next);
6180         *lookup_info = 1;
6181         return 1;
6182 }
6183
6184 /*
6185  * hepler to process tree block while walking up the tree.
6186  *
6187  * when wc->stage == DROP_REFERENCE, this function drops
6188  * reference count on the block.
6189  *
6190  * when wc->stage == UPDATE_BACKREF, this function changes
6191  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6192  * to UPDATE_BACKREF previously while processing the block.
6193  *
6194  * NOTE: return value 1 means we should stop walking up.
6195  */
6196 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6197                                  struct btrfs_root *root,
6198                                  struct btrfs_path *path,
6199                                  struct walk_control *wc)
6200 {
6201         int ret;
6202         int level = wc->level;
6203         struct extent_buffer *eb = path->nodes[level];
6204         u64 parent = 0;
6205
6206         if (wc->stage == UPDATE_BACKREF) {
6207                 BUG_ON(wc->shared_level < level);
6208                 if (level < wc->shared_level)
6209                         goto out;
6210
6211                 ret = find_next_key(path, level + 1, &wc->update_progress);
6212                 if (ret > 0)
6213                         wc->update_ref = 0;
6214
6215                 wc->stage = DROP_REFERENCE;
6216                 wc->shared_level = -1;
6217                 path->slots[level] = 0;
6218
6219                 /*
6220                  * check reference count again if the block isn't locked.
6221                  * we should start walking down the tree again if reference
6222                  * count is one.
6223                  */
6224                 if (!path->locks[level]) {
6225                         BUG_ON(level == 0);
6226                         btrfs_tree_lock(eb);
6227                         btrfs_set_lock_blocking(eb);
6228                         path->locks[level] = 1;
6229
6230                         ret = btrfs_lookup_extent_info(trans, root,
6231                                                        eb->start, eb->len,
6232                                                        &wc->refs[level],
6233                                                        &wc->flags[level]);
6234                         BUG_ON(ret);
6235                         BUG_ON(wc->refs[level] == 0);
6236                         if (wc->refs[level] == 1) {
6237                                 btrfs_tree_unlock(eb);
6238                                 path->locks[level] = 0;
6239                                 return 1;
6240                         }
6241                 }
6242         }
6243
6244         /* wc->stage == DROP_REFERENCE */
6245         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6246
6247         if (wc->refs[level] == 1) {
6248                 if (level == 0) {
6249                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6250                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6251                         else
6252                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6253                         BUG_ON(ret);
6254                 }
6255                 /* make block locked assertion in clean_tree_block happy */
6256                 if (!path->locks[level] &&
6257                     btrfs_header_generation(eb) == trans->transid) {
6258                         btrfs_tree_lock(eb);
6259                         btrfs_set_lock_blocking(eb);
6260                         path->locks[level] = 1;
6261                 }
6262                 clean_tree_block(trans, root, eb);
6263         }
6264
6265         if (eb == root->node) {
6266                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6267                         parent = eb->start;
6268                 else
6269                         BUG_ON(root->root_key.objectid !=
6270                                btrfs_header_owner(eb));
6271         } else {
6272                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6273                         parent = path->nodes[level + 1]->start;
6274                 else
6275                         BUG_ON(root->root_key.objectid !=
6276                                btrfs_header_owner(path->nodes[level + 1]));
6277         }
6278
6279         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6280 out:
6281         wc->refs[level] = 0;
6282         wc->flags[level] = 0;
6283         return 0;
6284 }
6285
6286 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6287                                    struct btrfs_root *root,
6288                                    struct btrfs_path *path,
6289                                    struct walk_control *wc)
6290 {
6291         int level = wc->level;
6292         int lookup_info = 1;
6293         int ret;
6294
6295         while (level >= 0) {
6296                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6297                 if (ret > 0)
6298                         break;
6299
6300                 if (level == 0)
6301                         break;
6302
6303                 if (path->slots[level] >=
6304                     btrfs_header_nritems(path->nodes[level]))
6305                         break;
6306
6307                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6308                 if (ret > 0) {
6309                         path->slots[level]++;
6310                         continue;
6311                 } else if (ret < 0)
6312                         return ret;
6313                 level = wc->level;
6314         }
6315         return 0;
6316 }
6317
6318 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6319                                  struct btrfs_root *root,
6320                                  struct btrfs_path *path,
6321                                  struct walk_control *wc, int max_level)
6322 {
6323         int level = wc->level;
6324         int ret;
6325
6326         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6327         while (level < max_level && path->nodes[level]) {
6328                 wc->level = level;
6329                 if (path->slots[level] + 1 <
6330                     btrfs_header_nritems(path->nodes[level])) {
6331                         path->slots[level]++;
6332                         return 0;
6333                 } else {
6334                         ret = walk_up_proc(trans, root, path, wc);
6335                         if (ret > 0)
6336                                 return 0;
6337
6338                         if (path->locks[level]) {
6339                                 btrfs_tree_unlock(path->nodes[level]);
6340                                 path->locks[level] = 0;
6341                         }
6342                         free_extent_buffer(path->nodes[level]);
6343                         path->nodes[level] = NULL;
6344                         level++;
6345                 }
6346         }
6347         return 1;
6348 }
6349
6350 /*
6351  * drop a subvolume tree.
6352  *
6353  * this function traverses the tree freeing any blocks that only
6354  * referenced by the tree.
6355  *
6356  * when a shared tree block is found. this function decreases its
6357  * reference count by one. if update_ref is true, this function
6358  * also make sure backrefs for the shared block and all lower level
6359  * blocks are properly updated.
6360  */
6361 int btrfs_drop_snapshot(struct btrfs_root *root,
6362                         struct btrfs_block_rsv *block_rsv, int update_ref)
6363 {
6364         struct btrfs_path *path;
6365         struct btrfs_trans_handle *trans;
6366         struct btrfs_root *tree_root = root->fs_info->tree_root;
6367         struct btrfs_root_item *root_item = &root->root_item;
6368         struct walk_control *wc;
6369         struct btrfs_key key;
6370         int err = 0;
6371         int ret;
6372         int level;
6373
6374         path = btrfs_alloc_path();
6375         BUG_ON(!path);
6376
6377         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6378         BUG_ON(!wc);
6379
6380         trans = btrfs_start_transaction(tree_root, 0);
6381         BUG_ON(IS_ERR(trans));
6382
6383         if (block_rsv)
6384                 trans->block_rsv = block_rsv;
6385
6386         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6387                 level = btrfs_header_level(root->node);
6388                 path->nodes[level] = btrfs_lock_root_node(root);
6389                 btrfs_set_lock_blocking(path->nodes[level]);
6390                 path->slots[level] = 0;
6391                 path->locks[level] = 1;
6392                 memset(&wc->update_progress, 0,
6393                        sizeof(wc->update_progress));
6394         } else {
6395                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6396                 memcpy(&wc->update_progress, &key,
6397                        sizeof(wc->update_progress));
6398
6399                 level = root_item->drop_level;
6400                 BUG_ON(level == 0);
6401                 path->lowest_level = level;
6402                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6403                 path->lowest_level = 0;
6404                 if (ret < 0) {
6405                         err = ret;
6406                         goto out;
6407                 }
6408                 WARN_ON(ret > 0);
6409
6410                 /*
6411                  * unlock our path, this is safe because only this
6412                  * function is allowed to delete this snapshot
6413                  */
6414                 btrfs_unlock_up_safe(path, 0);
6415
6416                 level = btrfs_header_level(root->node);
6417                 while (1) {
6418                         btrfs_tree_lock(path->nodes[level]);
6419                         btrfs_set_lock_blocking(path->nodes[level]);
6420
6421                         ret = btrfs_lookup_extent_info(trans, root,
6422                                                 path->nodes[level]->start,
6423                                                 path->nodes[level]->len,
6424                                                 &wc->refs[level],
6425                                                 &wc->flags[level]);
6426                         BUG_ON(ret);
6427                         BUG_ON(wc->refs[level] == 0);
6428
6429                         if (level == root_item->drop_level)
6430                                 break;
6431
6432                         btrfs_tree_unlock(path->nodes[level]);
6433                         WARN_ON(wc->refs[level] != 1);
6434                         level--;
6435                 }
6436         }
6437
6438         wc->level = level;
6439         wc->shared_level = -1;
6440         wc->stage = DROP_REFERENCE;
6441         wc->update_ref = update_ref;
6442         wc->keep_locks = 0;
6443         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6444
6445         while (1) {
6446                 ret = walk_down_tree(trans, root, path, wc);
6447                 if (ret < 0) {
6448                         err = ret;
6449                         break;
6450                 }
6451
6452                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6453                 if (ret < 0) {
6454                         err = ret;
6455                         break;
6456                 }
6457
6458                 if (ret > 0) {
6459                         BUG_ON(wc->stage != DROP_REFERENCE);
6460                         break;
6461                 }
6462
6463                 if (wc->stage == DROP_REFERENCE) {
6464                         level = wc->level;
6465                         btrfs_node_key(path->nodes[level],
6466                                        &root_item->drop_progress,
6467                                        path->slots[level]);
6468                         root_item->drop_level = level;
6469                 }
6470
6471                 BUG_ON(wc->level == 0);
6472                 if (btrfs_should_end_transaction(trans, tree_root)) {
6473                         ret = btrfs_update_root(trans, tree_root,
6474                                                 &root->root_key,
6475                                                 root_item);
6476                         BUG_ON(ret);
6477
6478                         btrfs_end_transaction_throttle(trans, tree_root);
6479                         trans = btrfs_start_transaction(tree_root, 0);
6480                         BUG_ON(IS_ERR(trans));
6481                         if (block_rsv)
6482                                 trans->block_rsv = block_rsv;
6483                 }
6484         }
6485         btrfs_release_path(root, path);
6486         BUG_ON(err);
6487
6488         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6489         BUG_ON(ret);
6490
6491         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6492                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6493                                            NULL, NULL);
6494                 BUG_ON(ret < 0);
6495                 if (ret > 0) {
6496                         /* if we fail to delete the orphan item this time
6497                          * around, it'll get picked up the next time.
6498                          *
6499                          * The most common failure here is just -ENOENT.
6500                          */
6501                         btrfs_del_orphan_item(trans, tree_root,
6502                                               root->root_key.objectid);
6503                 }
6504         }
6505
6506         if (root->in_radix) {
6507                 btrfs_free_fs_root(tree_root->fs_info, root);
6508         } else {
6509                 free_extent_buffer(root->node);
6510                 free_extent_buffer(root->commit_root);
6511                 kfree(root);
6512         }
6513 out:
6514         btrfs_end_transaction_throttle(trans, tree_root);
6515         kfree(wc);
6516         btrfs_free_path(path);
6517         return err;
6518 }
6519
6520 /*
6521  * drop subtree rooted at tree block 'node'.
6522  *
6523  * NOTE: this function will unlock and release tree block 'node'
6524  */
6525 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6526                         struct btrfs_root *root,
6527                         struct extent_buffer *node,
6528                         struct extent_buffer *parent)
6529 {
6530         struct btrfs_path *path;
6531         struct walk_control *wc;
6532         int level;
6533         int parent_level;
6534         int ret = 0;
6535         int wret;
6536
6537         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6538
6539         path = btrfs_alloc_path();
6540         if (!path)
6541                 return -ENOMEM;
6542
6543         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6544         if (!wc) {
6545                 btrfs_free_path(path);
6546                 return -ENOMEM;
6547         }
6548
6549         btrfs_assert_tree_locked(parent);
6550         parent_level = btrfs_header_level(parent);
6551         extent_buffer_get(parent);
6552         path->nodes[parent_level] = parent;
6553         path->slots[parent_level] = btrfs_header_nritems(parent);
6554
6555         btrfs_assert_tree_locked(node);
6556         level = btrfs_header_level(node);
6557         path->nodes[level] = node;
6558         path->slots[level] = 0;
6559         path->locks[level] = 1;
6560
6561         wc->refs[parent_level] = 1;
6562         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6563         wc->level = level;
6564         wc->shared_level = -1;
6565         wc->stage = DROP_REFERENCE;
6566         wc->update_ref = 0;
6567         wc->keep_locks = 1;
6568         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6569
6570         while (1) {
6571                 wret = walk_down_tree(trans, root, path, wc);
6572                 if (wret < 0) {
6573                         ret = wret;
6574                         break;
6575                 }
6576
6577                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6578                 if (wret < 0)
6579                         ret = wret;
6580                 if (wret != 0)
6581                         break;
6582         }
6583
6584         kfree(wc);
6585         btrfs_free_path(path);
6586         return ret;
6587 }
6588
6589 #if 0
6590 static unsigned long calc_ra(unsigned long start, unsigned long last,
6591                              unsigned long nr)
6592 {
6593         return min(last, start + nr - 1);
6594 }
6595
6596 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6597                                          u64 len)
6598 {
6599         u64 page_start;
6600         u64 page_end;
6601         unsigned long first_index;
6602         unsigned long last_index;
6603         unsigned long i;
6604         struct page *page;
6605         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6606         struct file_ra_state *ra;
6607         struct btrfs_ordered_extent *ordered;
6608         unsigned int total_read = 0;
6609         unsigned int total_dirty = 0;
6610         int ret = 0;
6611
6612         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6613         if (!ra)
6614                 return -ENOMEM;
6615
6616         mutex_lock(&inode->i_mutex);
6617         first_index = start >> PAGE_CACHE_SHIFT;
6618         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6619
6620         /* make sure the dirty trick played by the caller work */
6621         ret = invalidate_inode_pages2_range(inode->i_mapping,
6622                                             first_index, last_index);
6623         if (ret)
6624                 goto out_unlock;
6625
6626         file_ra_state_init(ra, inode->i_mapping);
6627
6628         for (i = first_index ; i <= last_index; i++) {
6629                 if (total_read % ra->ra_pages == 0) {
6630                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6631                                        calc_ra(i, last_index, ra->ra_pages));
6632                 }
6633                 total_read++;
6634 again:
6635                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6636                         BUG_ON(1);
6637                 page = grab_cache_page(inode->i_mapping, i);
6638                 if (!page) {
6639                         ret = -ENOMEM;
6640                         goto out_unlock;
6641                 }
6642                 if (!PageUptodate(page)) {
6643                         btrfs_readpage(NULL, page);
6644                         lock_page(page);
6645                         if (!PageUptodate(page)) {
6646                                 unlock_page(page);
6647                                 page_cache_release(page);
6648                                 ret = -EIO;
6649                                 goto out_unlock;
6650                         }
6651                 }
6652                 wait_on_page_writeback(page);
6653
6654                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6655                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6656                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6657
6658                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6659                 if (ordered) {
6660                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6661                         unlock_page(page);
6662                         page_cache_release(page);
6663                         btrfs_start_ordered_extent(inode, ordered, 1);
6664                         btrfs_put_ordered_extent(ordered);
6665                         goto again;
6666                 }
6667                 set_page_extent_mapped(page);
6668
6669                 if (i == first_index)
6670                         set_extent_bits(io_tree, page_start, page_end,
6671                                         EXTENT_BOUNDARY, GFP_NOFS);
6672                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6673
6674                 set_page_dirty(page);
6675                 total_dirty++;
6676
6677                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6678                 unlock_page(page);
6679                 page_cache_release(page);
6680         }
6681
6682 out_unlock:
6683         kfree(ra);
6684         mutex_unlock(&inode->i_mutex);
6685         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6686         return ret;
6687 }
6688
6689 static noinline int relocate_data_extent(struct inode *reloc_inode,
6690                                          struct btrfs_key *extent_key,
6691                                          u64 offset)
6692 {
6693         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6694         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6695         struct extent_map *em;
6696         u64 start = extent_key->objectid - offset;
6697         u64 end = start + extent_key->offset - 1;
6698
6699         em = alloc_extent_map(GFP_NOFS);
6700         BUG_ON(!em);
6701
6702         em->start = start;
6703         em->len = extent_key->offset;
6704         em->block_len = extent_key->offset;
6705         em->block_start = extent_key->objectid;
6706         em->bdev = root->fs_info->fs_devices->latest_bdev;
6707         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6708
6709         /* setup extent map to cheat btrfs_readpage */
6710         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6711         while (1) {
6712                 int ret;
6713                 write_lock(&em_tree->lock);
6714                 ret = add_extent_mapping(em_tree, em);
6715                 write_unlock(&em_tree->lock);
6716                 if (ret != -EEXIST) {
6717                         free_extent_map(em);
6718                         break;
6719                 }
6720                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6721         }
6722         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6723
6724         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6725 }
6726
6727 struct btrfs_ref_path {
6728         u64 extent_start;
6729         u64 nodes[BTRFS_MAX_LEVEL];
6730         u64 root_objectid;
6731         u64 root_generation;
6732         u64 owner_objectid;
6733         u32 num_refs;
6734         int lowest_level;
6735         int current_level;
6736         int shared_level;
6737
6738         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6739         u64 new_nodes[BTRFS_MAX_LEVEL];
6740 };
6741
6742 struct disk_extent {
6743         u64 ram_bytes;
6744         u64 disk_bytenr;
6745         u64 disk_num_bytes;
6746         u64 offset;
6747         u64 num_bytes;
6748         u8 compression;
6749         u8 encryption;
6750         u16 other_encoding;
6751 };
6752
6753 static int is_cowonly_root(u64 root_objectid)
6754 {
6755         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6756             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6757             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6758             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6759             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6760             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6761                 return 1;
6762         return 0;
6763 }
6764
6765 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6766                                     struct btrfs_root *extent_root,
6767                                     struct btrfs_ref_path *ref_path,
6768                                     int first_time)
6769 {
6770         struct extent_buffer *leaf;
6771         struct btrfs_path *path;
6772         struct btrfs_extent_ref *ref;
6773         struct btrfs_key key;
6774         struct btrfs_key found_key;
6775         u64 bytenr;
6776         u32 nritems;
6777         int level;
6778         int ret = 1;
6779
6780         path = btrfs_alloc_path();
6781         if (!path)
6782                 return -ENOMEM;
6783
6784         if (first_time) {
6785                 ref_path->lowest_level = -1;
6786                 ref_path->current_level = -1;
6787                 ref_path->shared_level = -1;
6788                 goto walk_up;
6789         }
6790 walk_down:
6791         level = ref_path->current_level - 1;
6792         while (level >= -1) {
6793                 u64 parent;
6794                 if (level < ref_path->lowest_level)
6795                         break;
6796
6797                 if (level >= 0)
6798                         bytenr = ref_path->nodes[level];
6799                 else
6800                         bytenr = ref_path->extent_start;
6801                 BUG_ON(bytenr == 0);
6802
6803                 parent = ref_path->nodes[level + 1];
6804                 ref_path->nodes[level + 1] = 0;
6805                 ref_path->current_level = level;
6806                 BUG_ON(parent == 0);
6807
6808                 key.objectid = bytenr;
6809                 key.offset = parent + 1;
6810                 key.type = BTRFS_EXTENT_REF_KEY;
6811
6812                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6813                 if (ret < 0)
6814                         goto out;
6815                 BUG_ON(ret == 0);
6816
6817                 leaf = path->nodes[0];
6818                 nritems = btrfs_header_nritems(leaf);
6819                 if (path->slots[0] >= nritems) {
6820                         ret = btrfs_next_leaf(extent_root, path);
6821                         if (ret < 0)
6822                                 goto out;
6823                         if (ret > 0)
6824                                 goto next;
6825                         leaf = path->nodes[0];
6826                 }
6827
6828                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6829                 if (found_key.objectid == bytenr &&
6830                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6831                         if (level < ref_path->shared_level)
6832                                 ref_path->shared_level = level;
6833                         goto found;
6834                 }
6835 next:
6836                 level--;
6837                 btrfs_release_path(extent_root, path);
6838                 cond_resched();
6839         }
6840         /* reached lowest level */
6841         ret = 1;
6842         goto out;
6843 walk_up:
6844         level = ref_path->current_level;
6845         while (level < BTRFS_MAX_LEVEL - 1) {
6846                 u64 ref_objectid;
6847
6848                 if (level >= 0)
6849                         bytenr = ref_path->nodes[level];
6850                 else
6851                         bytenr = ref_path->extent_start;
6852
6853                 BUG_ON(bytenr == 0);
6854
6855                 key.objectid = bytenr;
6856                 key.offset = 0;
6857                 key.type = BTRFS_EXTENT_REF_KEY;
6858
6859                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6860                 if (ret < 0)
6861                         goto out;
6862
6863                 leaf = path->nodes[0];
6864                 nritems = btrfs_header_nritems(leaf);
6865                 if (path->slots[0] >= nritems) {
6866                         ret = btrfs_next_leaf(extent_root, path);
6867                         if (ret < 0)
6868                                 goto out;
6869                         if (ret > 0) {
6870                                 /* the extent was freed by someone */
6871                                 if (ref_path->lowest_level == level)
6872                                         goto out;
6873                                 btrfs_release_path(extent_root, path);
6874                                 goto walk_down;
6875                         }
6876                         leaf = path->nodes[0];
6877                 }
6878
6879                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6880                 if (found_key.objectid != bytenr ||
6881                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6882                         /* the extent was freed by someone */
6883                         if (ref_path->lowest_level == level) {
6884                                 ret = 1;
6885                                 goto out;
6886                         }
6887                         btrfs_release_path(extent_root, path);
6888                         goto walk_down;
6889                 }
6890 found:
6891                 ref = btrfs_item_ptr(leaf, path->slots[0],
6892                                 struct btrfs_extent_ref);
6893                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6894                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6895                         if (first_time) {
6896                                 level = (int)ref_objectid;
6897                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6898                                 ref_path->lowest_level = level;
6899                                 ref_path->current_level = level;
6900                                 ref_path->nodes[level] = bytenr;
6901                         } else {
6902                                 WARN_ON(ref_objectid != level);
6903                         }
6904                 } else {
6905                         WARN_ON(level != -1);
6906                 }
6907                 first_time = 0;
6908
6909                 if (ref_path->lowest_level == level) {
6910                         ref_path->owner_objectid = ref_objectid;
6911                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6912                 }
6913
6914                 /*
6915                  * the block is tree root or the block isn't in reference
6916                  * counted tree.
6917                  */
6918                 if (found_key.objectid == found_key.offset ||
6919                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6920                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6921                         ref_path->root_generation =
6922                                 btrfs_ref_generation(leaf, ref);
6923                         if (level < 0) {
6924                                 /* special reference from the tree log */
6925                                 ref_path->nodes[0] = found_key.offset;
6926                                 ref_path->current_level = 0;
6927                         }
6928                         ret = 0;
6929                         goto out;
6930                 }
6931
6932                 level++;
6933                 BUG_ON(ref_path->nodes[level] != 0);
6934                 ref_path->nodes[level] = found_key.offset;
6935                 ref_path->current_level = level;
6936
6937                 /*
6938                  * the reference was created in the running transaction,
6939                  * no need to continue walking up.
6940                  */
6941                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6942                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6943                         ref_path->root_generation =
6944                                 btrfs_ref_generation(leaf, ref);
6945                         ret = 0;
6946                         goto out;
6947                 }
6948
6949                 btrfs_release_path(extent_root, path);
6950                 cond_resched();
6951         }
6952         /* reached max tree level, but no tree root found. */
6953         BUG();
6954 out:
6955         btrfs_free_path(path);
6956         return ret;
6957 }
6958
6959 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6960                                 struct btrfs_root *extent_root,
6961                                 struct btrfs_ref_path *ref_path,
6962                                 u64 extent_start)
6963 {
6964         memset(ref_path, 0, sizeof(*ref_path));
6965         ref_path->extent_start = extent_start;
6966
6967         return __next_ref_path(trans, extent_root, ref_path, 1);
6968 }
6969
6970 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6971                                struct btrfs_root *extent_root,
6972                                struct btrfs_ref_path *ref_path)
6973 {
6974         return __next_ref_path(trans, extent_root, ref_path, 0);
6975 }
6976
6977 static noinline int get_new_locations(struct inode *reloc_inode,
6978                                       struct btrfs_key *extent_key,
6979                                       u64 offset, int no_fragment,
6980                                       struct disk_extent **extents,
6981                                       int *nr_extents)
6982 {
6983         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6984         struct btrfs_path *path;
6985         struct btrfs_file_extent_item *fi;
6986         struct extent_buffer *leaf;
6987         struct disk_extent *exts = *extents;
6988         struct btrfs_key found_key;
6989         u64 cur_pos;
6990         u64 last_byte;
6991         u32 nritems;
6992         int nr = 0;
6993         int max = *nr_extents;
6994         int ret;
6995
6996         WARN_ON(!no_fragment && *extents);
6997         if (!exts) {
6998                 max = 1;
6999                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
7000                 if (!exts)
7001                         return -ENOMEM;
7002         }
7003
7004         path = btrfs_alloc_path();
7005         if (!path) {
7006                 if (exts != *extents)
7007                         kfree(exts);
7008                 return -ENOMEM;
7009         }
7010
7011         cur_pos = extent_key->objectid - offset;
7012         last_byte = extent_key->objectid + extent_key->offset;
7013         ret = btrfs_lookup_file_extent(NULL, root, path,
7014                                        btrfs_ino(reloc_inode), cur_pos, 0);
7015         if (ret < 0)
7016                 goto out;
7017         if (ret > 0) {
7018                 ret = -ENOENT;
7019                 goto out;
7020         }
7021
7022         while (1) {
7023                 leaf = path->nodes[0];
7024                 nritems = btrfs_header_nritems(leaf);
7025                 if (path->slots[0] >= nritems) {
7026                         ret = btrfs_next_leaf(root, path);
7027                         if (ret < 0)
7028                                 goto out;
7029                         if (ret > 0)
7030                                 break;
7031                         leaf = path->nodes[0];
7032                 }
7033
7034                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7035                 if (found_key.offset != cur_pos ||
7036                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
7037                     found_key.objectid != btrfs_ino(reloc_inode))
7038                         break;
7039
7040                 fi = btrfs_item_ptr(leaf, path->slots[0],
7041                                     struct btrfs_file_extent_item);
7042                 if (btrfs_file_extent_type(leaf, fi) !=
7043                     BTRFS_FILE_EXTENT_REG ||
7044                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7045                         break;
7046
7047                 if (nr == max) {
7048                         struct disk_extent *old = exts;
7049                         max *= 2;
7050                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
7051                         if (!exts) {
7052                                 ret = -ENOMEM;
7053                                 goto out;
7054                         }
7055                         memcpy(exts, old, sizeof(*exts) * nr);
7056                         if (old != *extents)
7057                                 kfree(old);
7058                 }
7059
7060                 exts[nr].disk_bytenr =
7061                         btrfs_file_extent_disk_bytenr(leaf, fi);
7062                 exts[nr].disk_num_bytes =
7063                         btrfs_file_extent_disk_num_bytes(leaf, fi);
7064                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
7065                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7066                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7067                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
7068                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
7069                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
7070                                                                            fi);
7071                 BUG_ON(exts[nr].offset > 0);
7072                 BUG_ON(exts[nr].compression || exts[nr].encryption);
7073                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
7074
7075                 cur_pos += exts[nr].num_bytes;
7076                 nr++;
7077
7078                 if (cur_pos + offset >= last_byte)
7079                         break;
7080
7081                 if (no_fragment) {
7082                         ret = 1;
7083                         goto out;
7084                 }
7085                 path->slots[0]++;
7086         }
7087
7088         BUG_ON(cur_pos + offset > last_byte);
7089         if (cur_pos + offset < last_byte) {
7090                 ret = -ENOENT;
7091                 goto out;
7092         }
7093         ret = 0;
7094 out:
7095         btrfs_free_path(path);
7096         if (ret) {
7097                 if (exts != *extents)
7098                         kfree(exts);
7099         } else {
7100                 *extents = exts;
7101                 *nr_extents = nr;
7102         }
7103         return ret;
7104 }
7105
7106 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
7107                                         struct btrfs_root *root,
7108                                         struct btrfs_path *path,
7109                                         struct btrfs_key *extent_key,
7110                                         struct btrfs_key *leaf_key,
7111                                         struct btrfs_ref_path *ref_path,
7112                                         struct disk_extent *new_extents,
7113                                         int nr_extents)
7114 {
7115         struct extent_buffer *leaf;
7116         struct btrfs_file_extent_item *fi;
7117         struct inode *inode = NULL;
7118         struct btrfs_key key;
7119         u64 lock_start = 0;
7120         u64 lock_end = 0;
7121         u64 num_bytes;
7122         u64 ext_offset;
7123         u64 search_end = (u64)-1;
7124         u32 nritems;
7125         int nr_scaned = 0;
7126         int extent_locked = 0;
7127         int extent_type;
7128         int ret;
7129
7130         memcpy(&key, leaf_key, sizeof(key));
7131         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7132                 if (key.objectid < ref_path->owner_objectid ||
7133                     (key.objectid == ref_path->owner_objectid &&
7134                      key.type < BTRFS_EXTENT_DATA_KEY)) {
7135                         key.objectid = ref_path->owner_objectid;
7136                         key.type = BTRFS_EXTENT_DATA_KEY;
7137                         key.offset = 0;
7138                 }
7139         }
7140
7141         while (1) {
7142                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7143                 if (ret < 0)
7144                         goto out;
7145
7146                 leaf = path->nodes[0];
7147                 nritems = btrfs_header_nritems(leaf);
7148 next:
7149                 if (extent_locked && ret > 0) {
7150                         /*
7151                          * the file extent item was modified by someone
7152                          * before the extent got locked.
7153                          */
7154                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7155                                       lock_end, GFP_NOFS);
7156                         extent_locked = 0;
7157                 }
7158
7159                 if (path->slots[0] >= nritems) {
7160                         if (++nr_scaned > 2)
7161                                 break;
7162
7163                         BUG_ON(extent_locked);
7164                         ret = btrfs_next_leaf(root, path);
7165                         if (ret < 0)
7166                                 goto out;
7167                         if (ret > 0)
7168                                 break;
7169                         leaf = path->nodes[0];
7170                         nritems = btrfs_header_nritems(leaf);
7171                 }
7172
7173                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7174
7175                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7176                         if ((key.objectid > ref_path->owner_objectid) ||
7177                             (key.objectid == ref_path->owner_objectid &&
7178                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7179                             key.offset >= search_end)
7180                                 break;
7181                 }
7182
7183                 if (inode && key.objectid != btrfs_ino(inode)) {
7184                         BUG_ON(extent_locked);
7185                         btrfs_release_path(root, path);
7186                         mutex_unlock(&inode->i_mutex);
7187                         iput(inode);
7188                         inode = NULL;
7189                         continue;
7190                 }
7191
7192                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7193                         path->slots[0]++;
7194                         ret = 1;
7195                         goto next;
7196                 }
7197                 fi = btrfs_item_ptr(leaf, path->slots[0],
7198                                     struct btrfs_file_extent_item);
7199                 extent_type = btrfs_file_extent_type(leaf, fi);
7200                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7201                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7202                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7203                      extent_key->objectid)) {
7204                         path->slots[0]++;
7205                         ret = 1;
7206                         goto next;
7207                 }
7208
7209                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7210                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7211
7212                 if (search_end == (u64)-1) {
7213                         search_end = key.offset - ext_offset +
7214                                 btrfs_file_extent_ram_bytes(leaf, fi);
7215                 }
7216
7217                 if (!extent_locked) {
7218                         lock_start = key.offset;
7219                         lock_end = lock_start + num_bytes - 1;
7220                 } else {
7221                         if (lock_start > key.offset ||
7222                             lock_end + 1 < key.offset + num_bytes) {
7223                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7224                                               lock_start, lock_end, GFP_NOFS);
7225                                 extent_locked = 0;
7226                         }
7227                 }
7228
7229                 if (!inode) {
7230                         btrfs_release_path(root, path);
7231
7232                         inode = btrfs_iget_locked(root->fs_info->sb,
7233                                                   key.objectid, root);
7234                         if (inode->i_state & I_NEW) {
7235                                 BTRFS_I(inode)->root = root;
7236                                 BTRFS_I(inode)->location.objectid =
7237                                         key.objectid;
7238                                 BTRFS_I(inode)->location.type =
7239                                         BTRFS_INODE_ITEM_KEY;
7240                                 BTRFS_I(inode)->location.offset = 0;
7241                                 btrfs_read_locked_inode(inode);
7242                                 unlock_new_inode(inode);
7243                         }
7244                         /*
7245                          * some code call btrfs_commit_transaction while
7246                          * holding the i_mutex, so we can't use mutex_lock
7247                          * here.
7248                          */
7249                         if (is_bad_inode(inode) ||
7250                             !mutex_trylock(&inode->i_mutex)) {
7251                                 iput(inode);
7252                                 inode = NULL;
7253                                 key.offset = (u64)-1;
7254                                 goto skip;
7255                         }
7256                 }
7257
7258                 if (!extent_locked) {
7259                         struct btrfs_ordered_extent *ordered;
7260
7261                         btrfs_release_path(root, path);
7262
7263                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7264                                     lock_end, GFP_NOFS);
7265                         ordered = btrfs_lookup_first_ordered_extent(inode,
7266                                                                     lock_end);
7267                         if (ordered &&
7268                             ordered->file_offset <= lock_end &&
7269                             ordered->file_offset + ordered->len > lock_start) {
7270                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7271                                               lock_start, lock_end, GFP_NOFS);
7272                                 btrfs_start_ordered_extent(inode, ordered, 1);
7273                                 btrfs_put_ordered_extent(ordered);
7274                                 key.offset += num_bytes;
7275                                 goto skip;
7276                         }
7277                         if (ordered)
7278                                 btrfs_put_ordered_extent(ordered);
7279
7280                         extent_locked = 1;
7281                         continue;
7282                 }
7283
7284                 if (nr_extents == 1) {
7285                         /* update extent pointer in place */
7286                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7287                                                 new_extents[0].disk_bytenr);
7288                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7289                                                 new_extents[0].disk_num_bytes);
7290                         btrfs_mark_buffer_dirty(leaf);
7291
7292                         btrfs_drop_extent_cache(inode, key.offset,
7293                                                 key.offset + num_bytes - 1, 0);
7294
7295                         ret = btrfs_inc_extent_ref(trans, root,
7296                                                 new_extents[0].disk_bytenr,
7297                                                 new_extents[0].disk_num_bytes,
7298                                                 leaf->start,
7299                                                 root->root_key.objectid,
7300                                                 trans->transid,
7301                                                 key.objectid);
7302                         BUG_ON(ret);
7303
7304                         ret = btrfs_free_extent(trans, root,
7305                                                 extent_key->objectid,
7306                                                 extent_key->offset,
7307                                                 leaf->start,
7308                                                 btrfs_header_owner(leaf),
7309                                                 btrfs_header_generation(leaf),
7310                                                 key.objectid, 0);
7311                         BUG_ON(ret);
7312
7313                         btrfs_release_path(root, path);
7314                         key.offset += num_bytes;
7315                 } else {
7316                         BUG_ON(1);
7317 #if 0
7318                         u64 alloc_hint;
7319                         u64 extent_len;
7320                         int i;
7321                         /*
7322                          * drop old extent pointer at first, then insert the
7323                          * new pointers one bye one
7324                          */
7325                         btrfs_release_path(root, path);
7326                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7327                                                  key.offset + num_bytes,
7328                                                  key.offset, &alloc_hint);
7329                         BUG_ON(ret);
7330
7331                         for (i = 0; i < nr_extents; i++) {
7332                                 if (ext_offset >= new_extents[i].num_bytes) {
7333                                         ext_offset -= new_extents[i].num_bytes;
7334                                         continue;
7335                                 }
7336                                 extent_len = min(new_extents[i].num_bytes -
7337                                                  ext_offset, num_bytes);
7338
7339                                 ret = btrfs_insert_empty_item(trans, root,
7340                                                               path, &key,
7341                                                               sizeof(*fi));
7342                                 BUG_ON(ret);
7343
7344                                 leaf = path->nodes[0];
7345                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7346                                                 struct btrfs_file_extent_item);
7347                                 btrfs_set_file_extent_generation(leaf, fi,
7348                                                         trans->transid);
7349                                 btrfs_set_file_extent_type(leaf, fi,
7350                                                         BTRFS_FILE_EXTENT_REG);
7351                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7352                                                 new_extents[i].disk_bytenr);
7353                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7354                                                 new_extents[i].disk_num_bytes);
7355                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7356                                                 new_extents[i].ram_bytes);
7357
7358                                 btrfs_set_file_extent_compression(leaf, fi,
7359                                                 new_extents[i].compression);
7360                                 btrfs_set_file_extent_encryption(leaf, fi,
7361                                                 new_extents[i].encryption);
7362                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7363                                                 new_extents[i].other_encoding);
7364
7365                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7366                                                         extent_len);
7367                                 ext_offset += new_extents[i].offset;
7368                                 btrfs_set_file_extent_offset(leaf, fi,
7369                                                         ext_offset);
7370                                 btrfs_mark_buffer_dirty(leaf);
7371
7372                                 btrfs_drop_extent_cache(inode, key.offset,
7373                                                 key.offset + extent_len - 1, 0);
7374
7375                                 ret = btrfs_inc_extent_ref(trans, root,
7376                                                 new_extents[i].disk_bytenr,
7377                                                 new_extents[i].disk_num_bytes,
7378                                                 leaf->start,
7379                                                 root->root_key.objectid,
7380                                                 trans->transid, key.objectid);
7381                                 BUG_ON(ret);
7382                                 btrfs_release_path(root, path);
7383
7384                                 inode_add_bytes(inode, extent_len);
7385
7386                                 ext_offset = 0;
7387                                 num_bytes -= extent_len;
7388                                 key.offset += extent_len;
7389
7390                                 if (num_bytes == 0)
7391                                         break;
7392                         }
7393                         BUG_ON(i >= nr_extents);
7394 #endif
7395                 }
7396
7397                 if (extent_locked) {
7398                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7399                                       lock_end, GFP_NOFS);
7400                         extent_locked = 0;
7401                 }
7402 skip:
7403                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7404                     key.offset >= search_end)
7405                         break;
7406
7407                 cond_resched();
7408         }
7409         ret = 0;
7410 out:
7411         btrfs_release_path(root, path);
7412         if (inode) {
7413                 mutex_unlock(&inode->i_mutex);
7414                 if (extent_locked) {
7415                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7416                                       lock_end, GFP_NOFS);
7417                 }
7418                 iput(inode);
7419         }
7420         return ret;
7421 }
7422
7423 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7424                                struct btrfs_root *root,
7425                                struct extent_buffer *buf, u64 orig_start)
7426 {
7427         int level;
7428         int ret;
7429
7430         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7431         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7432
7433         level = btrfs_header_level(buf);
7434         if (level == 0) {
7435                 struct btrfs_leaf_ref *ref;
7436                 struct btrfs_leaf_ref *orig_ref;
7437
7438                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7439                 if (!orig_ref)
7440                         return -ENOENT;
7441
7442                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7443                 if (!ref) {
7444                         btrfs_free_leaf_ref(root, orig_ref);
7445                         return -ENOMEM;
7446                 }
7447
7448                 ref->nritems = orig_ref->nritems;
7449                 memcpy(ref->extents, orig_ref->extents,
7450                         sizeof(ref->extents[0]) * ref->nritems);
7451
7452                 btrfs_free_leaf_ref(root, orig_ref);
7453
7454                 ref->root_gen = trans->transid;
7455                 ref->bytenr = buf->start;
7456                 ref->owner = btrfs_header_owner(buf);
7457                 ref->generation = btrfs_header_generation(buf);
7458
7459                 ret = btrfs_add_leaf_ref(root, ref, 0);
7460                 WARN_ON(ret);
7461                 btrfs_free_leaf_ref(root, ref);
7462         }
7463         return 0;
7464 }
7465
7466 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7467                                         struct extent_buffer *leaf,
7468                                         struct btrfs_block_group_cache *group,
7469                                         struct btrfs_root *target_root)
7470 {
7471         struct btrfs_key key;
7472         struct inode *inode = NULL;
7473         struct btrfs_file_extent_item *fi;
7474         struct extent_state *cached_state = NULL;
7475         u64 num_bytes;
7476         u64 skip_objectid = 0;
7477         u32 nritems;
7478         u32 i;
7479
7480         nritems = btrfs_header_nritems(leaf);
7481         for (i = 0; i < nritems; i++) {
7482                 btrfs_item_key_to_cpu(leaf, &key, i);
7483                 if (key.objectid == skip_objectid ||
7484                     key.type != BTRFS_EXTENT_DATA_KEY)
7485                         continue;
7486                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7487                 if (btrfs_file_extent_type(leaf, fi) ==
7488                     BTRFS_FILE_EXTENT_INLINE)
7489                         continue;
7490                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7491                         continue;
7492                 if (!inode || btrfs_ino(inode) != key.objectid) {
7493                         iput(inode);
7494                         inode = btrfs_ilookup(target_root->fs_info->sb,
7495                                               key.objectid, target_root, 1);
7496                 }
7497                 if (!inode) {
7498                         skip_objectid = key.objectid;
7499                         continue;
7500                 }
7501                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7502
7503                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7504                                  key.offset + num_bytes - 1, 0, &cached_state,
7505                                  GFP_NOFS);
7506                 btrfs_drop_extent_cache(inode, key.offset,
7507                                         key.offset + num_bytes - 1, 1);
7508                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7509                                      key.offset + num_bytes - 1, &cached_state,
7510                                      GFP_NOFS);
7511                 cond_resched();
7512         }
7513         iput(inode);
7514         return 0;
7515 }
7516
7517 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7518                                         struct btrfs_root *root,
7519                                         struct extent_buffer *leaf,
7520                                         struct btrfs_block_group_cache *group,
7521                                         struct inode *reloc_inode)
7522 {
7523         struct btrfs_key key;
7524         struct btrfs_key extent_key;
7525         struct btrfs_file_extent_item *fi;
7526         struct btrfs_leaf_ref *ref;
7527         struct disk_extent *new_extent;
7528         u64 bytenr;
7529         u64 num_bytes;
7530         u32 nritems;
7531         u32 i;
7532         int ext_index;
7533         int nr_extent;
7534         int ret;
7535
7536         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7537         if (!new_extent)
7538                 return -ENOMEM;
7539
7540         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7541         BUG_ON(!ref);
7542
7543         ext_index = -1;
7544         nritems = btrfs_header_nritems(leaf);
7545         for (i = 0; i < nritems; i++) {
7546                 btrfs_item_key_to_cpu(leaf, &key, i);
7547                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7548                         continue;
7549                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7550                 if (btrfs_file_extent_type(leaf, fi) ==
7551                     BTRFS_FILE_EXTENT_INLINE)
7552                         continue;
7553                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7554                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7555                 if (bytenr == 0)
7556                         continue;
7557
7558                 ext_index++;
7559                 if (bytenr >= group->key.objectid + group->key.offset ||
7560                     bytenr + num_bytes <= group->key.objectid)
7561                         continue;
7562
7563                 extent_key.objectid = bytenr;
7564                 extent_key.offset = num_bytes;
7565                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7566                 nr_extent = 1;
7567                 ret = get_new_locations(reloc_inode, &extent_key,
7568                                         group->key.objectid, 1,
7569                                         &new_extent, &nr_extent);
7570                 if (ret > 0)
7571                         continue;
7572                 BUG_ON(ret < 0);
7573
7574                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7575                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7576                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7577                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7578
7579                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7580                                                 new_extent->disk_bytenr);
7581                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7582                                                 new_extent->disk_num_bytes);
7583                 btrfs_mark_buffer_dirty(leaf);
7584
7585                 ret = btrfs_inc_extent_ref(trans, root,
7586                                         new_extent->disk_bytenr,
7587                                         new_extent->disk_num_bytes,
7588                                         leaf->start,
7589                                         root->root_key.objectid,
7590                                         trans->transid, key.objectid);
7591                 BUG_ON(ret);
7592
7593                 ret = btrfs_free_extent(trans, root,
7594                                         bytenr, num_bytes, leaf->start,
7595                                         btrfs_header_owner(leaf),
7596                                         btrfs_header_generation(leaf),
7597                                         key.objectid, 0);
7598                 BUG_ON(ret);
7599                 cond_resched();
7600         }
7601         kfree(new_extent);
7602         BUG_ON(ext_index + 1 != ref->nritems);
7603         btrfs_free_leaf_ref(root, ref);
7604         return 0;
7605 }
7606
7607 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7608                           struct btrfs_root *root)
7609 {
7610         struct btrfs_root *reloc_root;
7611         int ret;
7612
7613         if (root->reloc_root) {
7614                 reloc_root = root->reloc_root;
7615                 root->reloc_root = NULL;
7616                 list_add(&reloc_root->dead_list,
7617                          &root->fs_info->dead_reloc_roots);
7618
7619                 btrfs_set_root_bytenr(&reloc_root->root_item,
7620                                       reloc_root->node->start);
7621                 btrfs_set_root_level(&root->root_item,
7622                                      btrfs_header_level(reloc_root->node));
7623                 memset(&reloc_root->root_item.drop_progress, 0,
7624                         sizeof(struct btrfs_disk_key));
7625                 reloc_root->root_item.drop_level = 0;
7626
7627                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7628                                         &reloc_root->root_key,
7629                                         &reloc_root->root_item);
7630                 BUG_ON(ret);
7631         }
7632         return 0;
7633 }
7634
7635 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7636 {
7637         struct btrfs_trans_handle *trans;
7638         struct btrfs_root *reloc_root;
7639         struct btrfs_root *prev_root = NULL;
7640         struct list_head dead_roots;
7641         int ret;
7642         unsigned long nr;
7643
7644         INIT_LIST_HEAD(&dead_roots);
7645         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7646
7647         while (!list_empty(&dead_roots)) {
7648                 reloc_root = list_entry(dead_roots.prev,
7649                                         struct btrfs_root, dead_list);
7650                 list_del_init(&reloc_root->dead_list);
7651
7652                 BUG_ON(reloc_root->commit_root != NULL);
7653                 while (1) {
7654                         trans = btrfs_join_transaction(root, 1);
7655                         BUG_ON(IS_ERR(trans));
7656
7657                         mutex_lock(&root->fs_info->drop_mutex);
7658                         ret = btrfs_drop_snapshot(trans, reloc_root);
7659                         if (ret != -EAGAIN)
7660                                 break;
7661                         mutex_unlock(&root->fs_info->drop_mutex);
7662
7663                         nr = trans->blocks_used;
7664                         ret = btrfs_end_transaction(trans, root);
7665                         BUG_ON(ret);
7666                         btrfs_btree_balance_dirty(root, nr);
7667                 }
7668
7669                 free_extent_buffer(reloc_root->node);
7670
7671                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7672                                      &reloc_root->root_key);
7673                 BUG_ON(ret);
7674                 mutex_unlock(&root->fs_info->drop_mutex);
7675
7676                 nr = trans->blocks_used;
7677                 ret = btrfs_end_transaction(trans, root);
7678                 BUG_ON(ret);
7679                 btrfs_btree_balance_dirty(root, nr);
7680
7681                 kfree(prev_root);
7682                 prev_root = reloc_root;
7683         }
7684         if (prev_root) {
7685                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7686                 kfree(prev_root);
7687         }
7688         return 0;
7689 }
7690
7691 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7692 {
7693         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7694         return 0;
7695 }
7696
7697 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7698 {
7699         struct btrfs_root *reloc_root;
7700         struct btrfs_trans_handle *trans;
7701         struct btrfs_key location;
7702         int found;
7703         int ret;
7704
7705         mutex_lock(&root->fs_info->tree_reloc_mutex);
7706         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7707         BUG_ON(ret);
7708         found = !list_empty(&root->fs_info->dead_reloc_roots);
7709         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7710
7711         if (found) {
7712                 trans = btrfs_start_transaction(root, 1);
7713                 BUG_ON(IS_ERR(trans));
7714                 ret = btrfs_commit_transaction(trans, root);
7715                 BUG_ON(ret);
7716         }
7717
7718         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7719         location.offset = (u64)-1;
7720         location.type = BTRFS_ROOT_ITEM_KEY;
7721
7722         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7723         BUG_ON(!reloc_root);
7724         ret = btrfs_orphan_cleanup(reloc_root);
7725         BUG_ON(ret);
7726         return 0;
7727 }
7728
7729 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7730                                     struct btrfs_root *root)
7731 {
7732         struct btrfs_root *reloc_root;
7733         struct extent_buffer *eb;
7734         struct btrfs_root_item *root_item;
7735         struct btrfs_key root_key;
7736         int ret;
7737
7738         BUG_ON(!root->ref_cows);
7739         if (root->reloc_root)
7740                 return 0;
7741
7742         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7743         if (!root_item)
7744                 return -ENOMEM;
7745
7746         ret = btrfs_copy_root(trans, root, root->commit_root,
7747                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7748         BUG_ON(ret);
7749
7750         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7751         root_key.offset = root->root_key.objectid;
7752         root_key.type = BTRFS_ROOT_ITEM_KEY;
7753
7754         memcpy(root_item, &root->root_item, sizeof(root_item));
7755         btrfs_set_root_refs(root_item, 0);
7756         btrfs_set_root_bytenr(root_item, eb->start);
7757         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7758         btrfs_set_root_generation(root_item, trans->transid);
7759
7760         btrfs_tree_unlock(eb);
7761         free_extent_buffer(eb);
7762
7763         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7764                                 &root_key, root_item);
7765         BUG_ON(ret);
7766         kfree(root_item);
7767
7768         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7769                                                  &root_key);
7770         BUG_ON(IS_ERR(reloc_root));
7771         reloc_root->last_trans = trans->transid;
7772         reloc_root->commit_root = NULL;
7773         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7774
7775         root->reloc_root = reloc_root;
7776         return 0;
7777 }
7778
7779 /*
7780  * Core function of space balance.
7781  *
7782  * The idea is using reloc trees to relocate tree blocks in reference
7783  * counted roots. There is one reloc tree for each subvol, and all
7784  * reloc trees share same root key objectid. Reloc trees are snapshots
7785  * of the latest committed roots of subvols (root->commit_root).
7786  *
7787  * To relocate a tree block referenced by a subvol, there are two steps.
7788  * COW the block through subvol's reloc tree, then update block pointer
7789  * in the subvol to point to the new block. Since all reloc trees share
7790  * same root key objectid, doing special handing for tree blocks owned
7791  * by them is easy. Once a tree block has been COWed in one reloc tree,
7792  * we can use the resulting new block directly when the same block is
7793  * required to COW again through other reloc trees. By this way, relocated
7794  * tree blocks are shared between reloc trees, so they are also shared
7795  * between subvols.
7796  */
7797 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7798                                       struct btrfs_root *root,
7799                                       struct btrfs_path *path,
7800                                       struct btrfs_key *first_key,
7801                                       struct btrfs_ref_path *ref_path,
7802                                       struct btrfs_block_group_cache *group,
7803                                       struct inode *reloc_inode)
7804 {
7805         struct btrfs_root *reloc_root;
7806         struct extent_buffer *eb = NULL;
7807         struct btrfs_key *keys;
7808         u64 *nodes;
7809         int level;
7810         int shared_level;
7811         int lowest_level = 0;
7812         int ret;
7813
7814         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7815                 lowest_level = ref_path->owner_objectid;
7816
7817         if (!root->ref_cows) {
7818                 path->lowest_level = lowest_level;
7819                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7820                 BUG_ON(ret < 0);
7821                 path->lowest_level = 0;
7822                 btrfs_release_path(root, path);
7823                 return 0;
7824         }
7825
7826         mutex_lock(&root->fs_info->tree_reloc_mutex);
7827         ret = init_reloc_tree(trans, root);
7828         BUG_ON(ret);
7829         reloc_root = root->reloc_root;
7830
7831         shared_level = ref_path->shared_level;
7832         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7833
7834         keys = ref_path->node_keys;
7835         nodes = ref_path->new_nodes;
7836         memset(&keys[shared_level + 1], 0,
7837                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7838         memset(&nodes[shared_level + 1], 0,
7839                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7840
7841         if (nodes[lowest_level] == 0) {
7842                 path->lowest_level = lowest_level;
7843                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7844                                         0, 1);
7845                 BUG_ON(ret);
7846                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7847                         eb = path->nodes[level];
7848                         if (!eb || eb == reloc_root->node)
7849                                 break;
7850                         nodes[level] = eb->start;
7851                         if (level == 0)
7852                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7853                         else
7854                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7855                 }
7856                 if (nodes[0] &&
7857                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7858                         eb = path->nodes[0];
7859                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7860                                                       group, reloc_inode);
7861                         BUG_ON(ret);
7862                 }
7863                 btrfs_release_path(reloc_root, path);
7864         } else {
7865                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7866                                        lowest_level);
7867                 BUG_ON(ret);
7868         }
7869
7870         /*
7871          * replace tree blocks in the fs tree with tree blocks in
7872          * the reloc tree.
7873          */
7874         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7875         BUG_ON(ret < 0);
7876
7877         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7878                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7879                                         0, 0);
7880                 BUG_ON(ret);
7881                 extent_buffer_get(path->nodes[0]);
7882                 eb = path->nodes[0];
7883                 btrfs_release_path(reloc_root, path);
7884                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7885                 BUG_ON(ret);
7886                 free_extent_buffer(eb);
7887         }
7888
7889         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7890         path->lowest_level = 0;
7891         return 0;
7892 }
7893
7894 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7895                                         struct btrfs_root *root,
7896                                         struct btrfs_path *path,
7897                                         struct btrfs_key *first_key,
7898                                         struct btrfs_ref_path *ref_path)
7899 {
7900         int ret;
7901
7902         ret = relocate_one_path(trans, root, path, first_key,
7903                                 ref_path, NULL, NULL);
7904         BUG_ON(ret);
7905
7906         return 0;
7907 }
7908
7909 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7910                                     struct btrfs_root *extent_root,
7911                                     struct btrfs_path *path,
7912                                     struct btrfs_key *extent_key)
7913 {
7914         int ret;
7915
7916         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7917         if (ret)
7918                 goto out;
7919         ret = btrfs_del_item(trans, extent_root, path);
7920 out:
7921         btrfs_release_path(extent_root, path);
7922         return ret;
7923 }
7924
7925 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7926                                                 struct btrfs_ref_path *ref_path)
7927 {
7928         struct btrfs_key root_key;
7929
7930         root_key.objectid = ref_path->root_objectid;
7931         root_key.type = BTRFS_ROOT_ITEM_KEY;
7932         if (is_cowonly_root(ref_path->root_objectid))
7933                 root_key.offset = 0;
7934         else
7935                 root_key.offset = (u64)-1;
7936
7937         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7938 }
7939
7940 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7941                                         struct btrfs_path *path,
7942                                         struct btrfs_key *extent_key,
7943                                         struct btrfs_block_group_cache *group,
7944                                         struct inode *reloc_inode, int pass)
7945 {
7946         struct btrfs_trans_handle *trans;
7947         struct btrfs_root *found_root;
7948         struct btrfs_ref_path *ref_path = NULL;
7949         struct disk_extent *new_extents = NULL;
7950         int nr_extents = 0;
7951         int loops;
7952         int ret;
7953         int level;
7954         struct btrfs_key first_key;
7955         u64 prev_block = 0;
7956
7957
7958         trans = btrfs_start_transaction(extent_root, 1);
7959         BUG_ON(IS_ERR(trans));
7960
7961         if (extent_key->objectid == 0) {
7962                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7963                 goto out;
7964         }
7965
7966         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7967         if (!ref_path) {
7968                 ret = -ENOMEM;
7969                 goto out;
7970         }
7971
7972         for (loops = 0; ; loops++) {
7973                 if (loops == 0) {
7974                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7975                                                    extent_key->objectid);
7976                 } else {
7977                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7978                 }
7979                 if (ret < 0)
7980                         goto out;
7981                 if (ret > 0)
7982                         break;
7983
7984                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7985                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7986                         continue;
7987
7988                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7989                 BUG_ON(!found_root);
7990                 /*
7991                  * for reference counted tree, only process reference paths
7992                  * rooted at the latest committed root.
7993                  */
7994                 if (found_root->ref_cows &&
7995                     ref_path->root_generation != found_root->root_key.offset)
7996                         continue;
7997
7998                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7999                         if (pass == 0) {
8000                                 /*
8001                                  * copy data extents to new locations
8002                                  */
8003                                 u64 group_start = group->key.objectid;
8004                                 ret = relocate_data_extent(reloc_inode,
8005                                                            extent_key,
8006                                                            group_start);
8007                                 if (ret < 0)
8008                                         goto out;
8009                                 break;
8010                         }
8011                         level = 0;
8012                 } else {
8013                         level = ref_path->owner_objectid;
8014                 }
8015
8016                 if (prev_block != ref_path->nodes[level]) {
8017                         struct extent_buffer *eb;
8018                         u64 block_start = ref_path->nodes[level];
8019                         u64 block_size = btrfs_level_size(found_root, level);
8020
8021                         eb = read_tree_block(found_root, block_start,
8022                                              block_size, 0);
8023                         if (!eb) {
8024                                 ret = -EIO;
8025                                 goto out;
8026                         }
8027                         btrfs_tree_lock(eb);
8028                         BUG_ON(level != btrfs_header_level(eb));
8029
8030                         if (level == 0)
8031                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
8032                         else
8033                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
8034
8035                         btrfs_tree_unlock(eb);
8036                         free_extent_buffer(eb);
8037                         prev_block = block_start;
8038                 }
8039
8040                 mutex_lock(&extent_root->fs_info->trans_mutex);
8041                 btrfs_record_root_in_trans(found_root);
8042                 mutex_unlock(&extent_root->fs_info->trans_mutex);
8043                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8044                         /*
8045                          * try to update data extent references while
8046                          * keeping metadata shared between snapshots.
8047                          */
8048                         if (pass == 1) {
8049                                 ret = relocate_one_path(trans, found_root,
8050                                                 path, &first_key, ref_path,
8051                                                 group, reloc_inode);
8052                                 if (ret < 0)
8053                                         goto out;
8054                                 continue;
8055                         }
8056                         /*
8057                          * use fallback method to process the remaining
8058                          * references.
8059                          */
8060                         if (!new_extents) {
8061                                 u64 group_start = group->key.objectid;
8062                                 new_extents = kmalloc(sizeof(*new_extents),
8063                                                       GFP_NOFS);
8064                                 if (!new_extents) {
8065                                         ret = -ENOMEM;
8066                                         goto out;
8067                                 }
8068                                 nr_extents = 1;
8069                                 ret = get_new_locations(reloc_inode,
8070                                                         extent_key,
8071                                                         group_start, 1,
8072                                                         &new_extents,
8073                                                         &nr_extents);
8074                                 if (ret)
8075                                         goto out;
8076                         }
8077                         ret = replace_one_extent(trans, found_root,
8078                                                 path, extent_key,
8079                                                 &first_key, ref_path,
8080                                                 new_extents, nr_extents);
8081                 } else {
8082                         ret = relocate_tree_block(trans, found_root, path,
8083                                                   &first_key, ref_path);
8084                 }
8085                 if (ret < 0)
8086                         goto out;
8087         }
8088         ret = 0;
8089 out:
8090         btrfs_end_transaction(trans, extent_root);
8091         kfree(new_extents);
8092         kfree(ref_path);
8093         return ret;
8094 }
8095 #endif
8096
8097 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8098 {
8099         u64 num_devices;
8100         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8101                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8102
8103         /*
8104          * we add in the count of missing devices because we want
8105          * to make sure that any RAID levels on a degraded FS
8106          * continue to be honored.
8107          */
8108         num_devices = root->fs_info->fs_devices->rw_devices +
8109                 root->fs_info->fs_devices->missing_devices;
8110
8111         if (num_devices == 1) {
8112                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8113                 stripped = flags & ~stripped;
8114
8115                 /* turn raid0 into single device chunks */
8116                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8117                         return stripped;
8118
8119                 /* turn mirroring into duplication */
8120                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8121                              BTRFS_BLOCK_GROUP_RAID10))
8122                         return stripped | BTRFS_BLOCK_GROUP_DUP;
8123                 return flags;
8124         } else {
8125                 /* they already had raid on here, just return */
8126                 if (flags & stripped)
8127                         return flags;
8128
8129                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8130                 stripped = flags & ~stripped;
8131
8132                 /* switch duplicated blocks with raid1 */
8133                 if (flags & BTRFS_BLOCK_GROUP_DUP)
8134                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
8135
8136                 /* turn single device chunks into raid0 */
8137                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8138         }
8139         return flags;
8140 }
8141
8142 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8143 {
8144         struct btrfs_space_info *sinfo = cache->space_info;
8145         u64 num_bytes;
8146         int ret = -ENOSPC;
8147
8148         if (cache->ro)
8149                 return 0;
8150
8151         spin_lock(&sinfo->lock);
8152         spin_lock(&cache->lock);
8153         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8154                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8155
8156         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8157             sinfo->bytes_may_use + sinfo->bytes_readonly +
8158             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8159                 sinfo->bytes_readonly += num_bytes;
8160                 sinfo->bytes_reserved += cache->reserved_pinned;
8161                 cache->reserved_pinned = 0;
8162                 cache->ro = 1;
8163                 ret = 0;
8164         }
8165
8166         spin_unlock(&cache->lock);
8167         spin_unlock(&sinfo->lock);
8168         return ret;
8169 }
8170
8171 int btrfs_set_block_group_ro(struct btrfs_root *root,
8172                              struct btrfs_block_group_cache *cache)
8173
8174 {
8175         struct btrfs_trans_handle *trans;
8176         u64 alloc_flags;
8177         int ret;
8178
8179         BUG_ON(cache->ro);
8180
8181         trans = btrfs_join_transaction(root, 1);
8182         BUG_ON(IS_ERR(trans));
8183
8184         alloc_flags = update_block_group_flags(root, cache->flags);
8185         if (alloc_flags != cache->flags)
8186                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8187                                CHUNK_ALLOC_FORCE);
8188
8189         ret = set_block_group_ro(cache);
8190         if (!ret)
8191                 goto out;
8192         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8193         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8194                              CHUNK_ALLOC_FORCE);
8195         if (ret < 0)
8196                 goto out;
8197         ret = set_block_group_ro(cache);
8198 out:
8199         btrfs_end_transaction(trans, root);
8200         return ret;
8201 }
8202
8203 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8204                             struct btrfs_root *root, u64 type)
8205 {
8206         u64 alloc_flags = get_alloc_profile(root, type);
8207         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8208                               CHUNK_ALLOC_FORCE);
8209 }
8210
8211 /*
8212  * helper to account the unused space of all the readonly block group in the
8213  * list. takes mirrors into account.
8214  */
8215 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8216 {
8217         struct btrfs_block_group_cache *block_group;
8218         u64 free_bytes = 0;
8219         int factor;
8220
8221         list_for_each_entry(block_group, groups_list, list) {
8222                 spin_lock(&block_group->lock);
8223
8224                 if (!block_group->ro) {
8225                         spin_unlock(&block_group->lock);
8226                         continue;
8227                 }
8228
8229                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8230                                           BTRFS_BLOCK_GROUP_RAID10 |
8231                                           BTRFS_BLOCK_GROUP_DUP))
8232                         factor = 2;
8233                 else
8234                         factor = 1;
8235
8236                 free_bytes += (block_group->key.offset -
8237                                btrfs_block_group_used(&block_group->item)) *
8238                                factor;
8239
8240                 spin_unlock(&block_group->lock);
8241         }
8242
8243         return free_bytes;
8244 }
8245
8246 /*
8247  * helper to account the unused space of all the readonly block group in the
8248  * space_info. takes mirrors into account.
8249  */
8250 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8251 {
8252         int i;
8253         u64 free_bytes = 0;
8254
8255         spin_lock(&sinfo->lock);
8256
8257         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8258                 if (!list_empty(&sinfo->block_groups[i]))
8259                         free_bytes += __btrfs_get_ro_block_group_free_space(
8260                                                 &sinfo->block_groups[i]);
8261
8262         spin_unlock(&sinfo->lock);
8263
8264         return free_bytes;
8265 }
8266
8267 int btrfs_set_block_group_rw(struct btrfs_root *root,
8268                               struct btrfs_block_group_cache *cache)
8269 {
8270         struct btrfs_space_info *sinfo = cache->space_info;
8271         u64 num_bytes;
8272
8273         BUG_ON(!cache->ro);
8274
8275         spin_lock(&sinfo->lock);
8276         spin_lock(&cache->lock);
8277         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8278                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8279         sinfo->bytes_readonly -= num_bytes;
8280         cache->ro = 0;
8281         spin_unlock(&cache->lock);
8282         spin_unlock(&sinfo->lock);
8283         return 0;
8284 }
8285
8286 /*
8287  * checks to see if its even possible to relocate this block group.
8288  *
8289  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8290  * ok to go ahead and try.
8291  */
8292 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8293 {
8294         struct btrfs_block_group_cache *block_group;
8295         struct btrfs_space_info *space_info;
8296         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8297         struct btrfs_device *device;
8298         int full = 0;
8299         int ret = 0;
8300
8301         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8302
8303         /* odd, couldn't find the block group, leave it alone */
8304         if (!block_group)
8305                 return -1;
8306
8307         /* no bytes used, we're good */
8308         if (!btrfs_block_group_used(&block_group->item))
8309                 goto out;
8310
8311         space_info = block_group->space_info;
8312         spin_lock(&space_info->lock);
8313
8314         full = space_info->full;
8315
8316         /*
8317          * if this is the last block group we have in this space, we can't
8318          * relocate it unless we're able to allocate a new chunk below.
8319          *
8320          * Otherwise, we need to make sure we have room in the space to handle
8321          * all of the extents from this block group.  If we can, we're good
8322          */
8323         if ((space_info->total_bytes != block_group->key.offset) &&
8324            (space_info->bytes_used + space_info->bytes_reserved +
8325             space_info->bytes_pinned + space_info->bytes_readonly +
8326             btrfs_block_group_used(&block_group->item) <
8327             space_info->total_bytes)) {
8328                 spin_unlock(&space_info->lock);
8329                 goto out;
8330         }
8331         spin_unlock(&space_info->lock);
8332
8333         /*
8334          * ok we don't have enough space, but maybe we have free space on our
8335          * devices to allocate new chunks for relocation, so loop through our
8336          * alloc devices and guess if we have enough space.  However, if we
8337          * were marked as full, then we know there aren't enough chunks, and we
8338          * can just return.
8339          */
8340         ret = -1;
8341         if (full)
8342                 goto out;
8343
8344         mutex_lock(&root->fs_info->chunk_mutex);
8345         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8346                 u64 min_free = btrfs_block_group_used(&block_group->item);
8347                 u64 dev_offset;
8348
8349                 /*
8350                  * check to make sure we can actually find a chunk with enough
8351                  * space to fit our block group in.
8352                  */
8353                 if (device->total_bytes > device->bytes_used + min_free) {
8354                         ret = find_free_dev_extent(NULL, device, min_free,
8355                                                    &dev_offset, NULL);
8356                         if (!ret)
8357                                 break;
8358                         ret = -1;
8359                 }
8360         }
8361         mutex_unlock(&root->fs_info->chunk_mutex);
8362 out:
8363         btrfs_put_block_group(block_group);
8364         return ret;
8365 }
8366
8367 static int find_first_block_group(struct btrfs_root *root,
8368                 struct btrfs_path *path, struct btrfs_key *key)
8369 {
8370         int ret = 0;
8371         struct btrfs_key found_key;
8372         struct extent_buffer *leaf;
8373         int slot;
8374
8375         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8376         if (ret < 0)
8377                 goto out;
8378
8379         while (1) {
8380                 slot = path->slots[0];
8381                 leaf = path->nodes[0];
8382                 if (slot >= btrfs_header_nritems(leaf)) {
8383                         ret = btrfs_next_leaf(root, path);
8384                         if (ret == 0)
8385                                 continue;
8386                         if (ret < 0)
8387                                 goto out;
8388                         break;
8389                 }
8390                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8391
8392                 if (found_key.objectid >= key->objectid &&
8393                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8394                         ret = 0;
8395                         goto out;
8396                 }
8397                 path->slots[0]++;
8398         }
8399 out:
8400         return ret;
8401 }
8402
8403 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8404 {
8405         struct btrfs_block_group_cache *block_group;
8406         u64 last = 0;
8407
8408         while (1) {
8409                 struct inode *inode;
8410
8411                 block_group = btrfs_lookup_first_block_group(info, last);
8412                 while (block_group) {
8413                         spin_lock(&block_group->lock);
8414                         if (block_group->iref)
8415                                 break;
8416                         spin_unlock(&block_group->lock);
8417                         block_group = next_block_group(info->tree_root,
8418                                                        block_group);
8419                 }
8420                 if (!block_group) {
8421                         if (last == 0)
8422                                 break;
8423                         last = 0;
8424                         continue;
8425                 }
8426
8427                 inode = block_group->inode;
8428                 block_group->iref = 0;
8429                 block_group->inode = NULL;
8430                 spin_unlock(&block_group->lock);
8431                 iput(inode);
8432                 last = block_group->key.objectid + block_group->key.offset;
8433                 btrfs_put_block_group(block_group);
8434         }
8435 }
8436
8437 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8438 {
8439         struct btrfs_block_group_cache *block_group;
8440         struct btrfs_space_info *space_info;
8441         struct btrfs_caching_control *caching_ctl;
8442         struct rb_node *n;
8443
8444         down_write(&info->extent_commit_sem);
8445         while (!list_empty(&info->caching_block_groups)) {
8446                 caching_ctl = list_entry(info->caching_block_groups.next,
8447                                          struct btrfs_caching_control, list);
8448                 list_del(&caching_ctl->list);
8449                 put_caching_control(caching_ctl);
8450         }
8451         up_write(&info->extent_commit_sem);
8452
8453         spin_lock(&info->block_group_cache_lock);
8454         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8455                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8456                                        cache_node);
8457                 rb_erase(&block_group->cache_node,
8458                          &info->block_group_cache_tree);
8459                 spin_unlock(&info->block_group_cache_lock);
8460
8461                 down_write(&block_group->space_info->groups_sem);
8462                 list_del(&block_group->list);
8463                 up_write(&block_group->space_info->groups_sem);
8464
8465                 if (block_group->cached == BTRFS_CACHE_STARTED)
8466                         wait_block_group_cache_done(block_group);
8467
8468                 /*
8469                  * We haven't cached this block group, which means we could
8470                  * possibly have excluded extents on this block group.
8471                  */
8472                 if (block_group->cached == BTRFS_CACHE_NO)
8473                         free_excluded_extents(info->extent_root, block_group);
8474
8475                 btrfs_remove_free_space_cache(block_group);
8476                 btrfs_put_block_group(block_group);
8477
8478                 spin_lock(&info->block_group_cache_lock);
8479         }
8480         spin_unlock(&info->block_group_cache_lock);
8481
8482         /* now that all the block groups are freed, go through and
8483          * free all the space_info structs.  This is only called during
8484          * the final stages of unmount, and so we know nobody is
8485          * using them.  We call synchronize_rcu() once before we start,
8486          * just to be on the safe side.
8487          */
8488         synchronize_rcu();
8489
8490         release_global_block_rsv(info);
8491
8492         while(!list_empty(&info->space_info)) {
8493                 space_info = list_entry(info->space_info.next,
8494                                         struct btrfs_space_info,
8495                                         list);
8496                 if (space_info->bytes_pinned > 0 ||
8497                     space_info->bytes_reserved > 0) {
8498                         WARN_ON(1);
8499                         dump_space_info(space_info, 0, 0);
8500                 }
8501                 list_del(&space_info->list);
8502                 kfree(space_info);
8503         }
8504         return 0;
8505 }
8506
8507 static void __link_block_group(struct btrfs_space_info *space_info,
8508                                struct btrfs_block_group_cache *cache)
8509 {
8510         int index = get_block_group_index(cache);
8511
8512         down_write(&space_info->groups_sem);
8513         list_add_tail(&cache->list, &space_info->block_groups[index]);
8514         up_write(&space_info->groups_sem);
8515 }
8516
8517 int btrfs_read_block_groups(struct btrfs_root *root)
8518 {
8519         struct btrfs_path *path;
8520         int ret;
8521         struct btrfs_block_group_cache *cache;
8522         struct btrfs_fs_info *info = root->fs_info;
8523         struct btrfs_space_info *space_info;
8524         struct btrfs_key key;
8525         struct btrfs_key found_key;
8526         struct extent_buffer *leaf;
8527         int need_clear = 0;
8528         u64 cache_gen;
8529
8530         root = info->extent_root;
8531         key.objectid = 0;
8532         key.offset = 0;
8533         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8534         path = btrfs_alloc_path();
8535         if (!path)
8536                 return -ENOMEM;
8537
8538         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8539         if (cache_gen != 0 &&
8540             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8541                 need_clear = 1;
8542         if (btrfs_test_opt(root, CLEAR_CACHE))
8543                 need_clear = 1;
8544         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8545                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8546
8547         while (1) {
8548                 ret = find_first_block_group(root, path, &key);
8549                 if (ret > 0)
8550                         break;
8551                 if (ret != 0)
8552                         goto error;
8553                 leaf = path->nodes[0];
8554                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8555                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8556                 if (!cache) {
8557                         ret = -ENOMEM;
8558                         goto error;
8559                 }
8560                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8561                                                 GFP_NOFS);
8562                 if (!cache->free_space_ctl) {
8563                         kfree(cache);
8564                         ret = -ENOMEM;
8565                         goto error;
8566                 }
8567
8568                 atomic_set(&cache->count, 1);
8569                 spin_lock_init(&cache->lock);
8570                 cache->fs_info = info;
8571                 INIT_LIST_HEAD(&cache->list);
8572                 INIT_LIST_HEAD(&cache->cluster_list);
8573
8574                 if (need_clear)
8575                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8576
8577                 read_extent_buffer(leaf, &cache->item,
8578                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8579                                    sizeof(cache->item));
8580                 memcpy(&cache->key, &found_key, sizeof(found_key));
8581
8582                 key.objectid = found_key.objectid + found_key.offset;
8583                 btrfs_release_path(root, path);
8584                 cache->flags = btrfs_block_group_flags(&cache->item);
8585                 cache->sectorsize = root->sectorsize;
8586
8587                 btrfs_init_free_space_ctl(cache);
8588
8589                 /*
8590                  * We need to exclude the super stripes now so that the space
8591                  * info has super bytes accounted for, otherwise we'll think
8592                  * we have more space than we actually do.
8593                  */
8594                 exclude_super_stripes(root, cache);
8595
8596                 /*
8597                  * check for two cases, either we are full, and therefore
8598                  * don't need to bother with the caching work since we won't
8599                  * find any space, or we are empty, and we can just add all
8600                  * the space in and be done with it.  This saves us _alot_ of
8601                  * time, particularly in the full case.
8602                  */
8603                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8604                         cache->last_byte_to_unpin = (u64)-1;
8605                         cache->cached = BTRFS_CACHE_FINISHED;
8606                         free_excluded_extents(root, cache);
8607                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8608                         cache->last_byte_to_unpin = (u64)-1;
8609                         cache->cached = BTRFS_CACHE_FINISHED;
8610                         add_new_free_space(cache, root->fs_info,
8611                                            found_key.objectid,
8612                                            found_key.objectid +
8613                                            found_key.offset);
8614                         free_excluded_extents(root, cache);
8615                 }
8616
8617                 ret = update_space_info(info, cache->flags, found_key.offset,
8618                                         btrfs_block_group_used(&cache->item),
8619                                         &space_info);
8620                 BUG_ON(ret);
8621                 cache->space_info = space_info;
8622                 spin_lock(&cache->space_info->lock);
8623                 cache->space_info->bytes_readonly += cache->bytes_super;
8624                 spin_unlock(&cache->space_info->lock);
8625
8626                 __link_block_group(space_info, cache);
8627
8628                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8629                 BUG_ON(ret);
8630
8631                 set_avail_alloc_bits(root->fs_info, cache->flags);
8632                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8633                         set_block_group_ro(cache);
8634         }
8635
8636         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8637                 if (!(get_alloc_profile(root, space_info->flags) &
8638                       (BTRFS_BLOCK_GROUP_RAID10 |
8639                        BTRFS_BLOCK_GROUP_RAID1 |
8640                        BTRFS_BLOCK_GROUP_DUP)))
8641                         continue;
8642                 /*
8643                  * avoid allocating from un-mirrored block group if there are
8644                  * mirrored block groups.
8645                  */
8646                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8647                         set_block_group_ro(cache);
8648                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8649                         set_block_group_ro(cache);
8650         }
8651
8652         init_global_block_rsv(info);
8653         ret = 0;
8654 error:
8655         btrfs_free_path(path);
8656         return ret;
8657 }
8658
8659 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8660                            struct btrfs_root *root, u64 bytes_used,
8661                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8662                            u64 size)
8663 {
8664         int ret;
8665         struct btrfs_root *extent_root;
8666         struct btrfs_block_group_cache *cache;
8667
8668         extent_root = root->fs_info->extent_root;
8669
8670         root->fs_info->last_trans_log_full_commit = trans->transid;
8671
8672         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8673         if (!cache)
8674                 return -ENOMEM;
8675         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8676                                         GFP_NOFS);
8677         if (!cache->free_space_ctl) {
8678                 kfree(cache);
8679                 return -ENOMEM;
8680         }
8681
8682         cache->key.objectid = chunk_offset;
8683         cache->key.offset = size;
8684         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8685         cache->sectorsize = root->sectorsize;
8686         cache->fs_info = root->fs_info;
8687
8688         atomic_set(&cache->count, 1);
8689         spin_lock_init(&cache->lock);
8690         INIT_LIST_HEAD(&cache->list);
8691         INIT_LIST_HEAD(&cache->cluster_list);
8692
8693         btrfs_init_free_space_ctl(cache);
8694
8695         btrfs_set_block_group_used(&cache->item, bytes_used);
8696         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8697         cache->flags = type;
8698         btrfs_set_block_group_flags(&cache->item, type);
8699
8700         cache->last_byte_to_unpin = (u64)-1;
8701         cache->cached = BTRFS_CACHE_FINISHED;
8702         exclude_super_stripes(root, cache);
8703
8704         add_new_free_space(cache, root->fs_info, chunk_offset,
8705                            chunk_offset + size);
8706
8707         free_excluded_extents(root, cache);
8708
8709         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8710                                 &cache->space_info);
8711         BUG_ON(ret);
8712
8713         spin_lock(&cache->space_info->lock);
8714         cache->space_info->bytes_readonly += cache->bytes_super;
8715         spin_unlock(&cache->space_info->lock);
8716
8717         __link_block_group(cache->space_info, cache);
8718
8719         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8720         BUG_ON(ret);
8721
8722         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8723                                 sizeof(cache->item));
8724         BUG_ON(ret);
8725
8726         set_avail_alloc_bits(extent_root->fs_info, type);
8727
8728         return 0;
8729 }
8730
8731 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8732                              struct btrfs_root *root, u64 group_start)
8733 {
8734         struct btrfs_path *path;
8735         struct btrfs_block_group_cache *block_group;
8736         struct btrfs_free_cluster *cluster;
8737         struct btrfs_root *tree_root = root->fs_info->tree_root;
8738         struct btrfs_key key;
8739         struct inode *inode;
8740         int ret;
8741         int factor;
8742
8743         root = root->fs_info->extent_root;
8744
8745         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8746         BUG_ON(!block_group);
8747         BUG_ON(!block_group->ro);
8748
8749         /*
8750          * Free the reserved super bytes from this block group before
8751          * remove it.
8752          */
8753         free_excluded_extents(root, block_group);
8754
8755         memcpy(&key, &block_group->key, sizeof(key));
8756         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8757                                   BTRFS_BLOCK_GROUP_RAID1 |
8758                                   BTRFS_BLOCK_GROUP_RAID10))
8759                 factor = 2;
8760         else
8761                 factor = 1;
8762
8763         /* make sure this block group isn't part of an allocation cluster */
8764         cluster = &root->fs_info->data_alloc_cluster;
8765         spin_lock(&cluster->refill_lock);
8766         btrfs_return_cluster_to_free_space(block_group, cluster);
8767         spin_unlock(&cluster->refill_lock);
8768
8769         /*
8770          * make sure this block group isn't part of a metadata
8771          * allocation cluster
8772          */
8773         cluster = &root->fs_info->meta_alloc_cluster;
8774         spin_lock(&cluster->refill_lock);
8775         btrfs_return_cluster_to_free_space(block_group, cluster);
8776         spin_unlock(&cluster->refill_lock);
8777
8778         path = btrfs_alloc_path();
8779         BUG_ON(!path);
8780
8781         inode = lookup_free_space_inode(root, block_group, path);
8782         if (!IS_ERR(inode)) {
8783                 btrfs_orphan_add(trans, inode);
8784                 clear_nlink(inode);
8785                 /* One for the block groups ref */
8786                 spin_lock(&block_group->lock);
8787                 if (block_group->iref) {
8788                         block_group->iref = 0;
8789                         block_group->inode = NULL;
8790                         spin_unlock(&block_group->lock);
8791                         iput(inode);
8792                 } else {
8793                         spin_unlock(&block_group->lock);
8794                 }
8795                 /* One for our lookup ref */
8796                 iput(inode);
8797         }
8798
8799         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8800         key.offset = block_group->key.objectid;
8801         key.type = 0;
8802
8803         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8804         if (ret < 0)
8805                 goto out;
8806         if (ret > 0)
8807                 btrfs_release_path(tree_root, path);
8808         if (ret == 0) {
8809                 ret = btrfs_del_item(trans, tree_root, path);
8810                 if (ret)
8811                         goto out;
8812                 btrfs_release_path(tree_root, path);
8813         }
8814
8815         spin_lock(&root->fs_info->block_group_cache_lock);
8816         rb_erase(&block_group->cache_node,
8817                  &root->fs_info->block_group_cache_tree);
8818         spin_unlock(&root->fs_info->block_group_cache_lock);
8819
8820         down_write(&block_group->space_info->groups_sem);
8821         /*
8822          * we must use list_del_init so people can check to see if they
8823          * are still on the list after taking the semaphore
8824          */
8825         list_del_init(&block_group->list);
8826         up_write(&block_group->space_info->groups_sem);
8827
8828         if (block_group->cached == BTRFS_CACHE_STARTED)
8829                 wait_block_group_cache_done(block_group);
8830
8831         btrfs_remove_free_space_cache(block_group);
8832
8833         spin_lock(&block_group->space_info->lock);
8834         block_group->space_info->total_bytes -= block_group->key.offset;
8835         block_group->space_info->bytes_readonly -= block_group->key.offset;
8836         block_group->space_info->disk_total -= block_group->key.offset * factor;
8837         spin_unlock(&block_group->space_info->lock);
8838
8839         memcpy(&key, &block_group->key, sizeof(key));
8840
8841         btrfs_clear_space_info_full(root->fs_info);
8842
8843         btrfs_put_block_group(block_group);
8844         btrfs_put_block_group(block_group);
8845
8846         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8847         if (ret > 0)
8848                 ret = -EIO;
8849         if (ret < 0)
8850                 goto out;
8851
8852         ret = btrfs_del_item(trans, root, path);
8853 out:
8854         btrfs_free_path(path);
8855         return ret;
8856 }
8857
8858 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8859 {
8860         struct btrfs_space_info *space_info;
8861         struct btrfs_super_block *disk_super;
8862         u64 features;
8863         u64 flags;
8864         int mixed = 0;
8865         int ret;
8866
8867         disk_super = &fs_info->super_copy;
8868         if (!btrfs_super_root(disk_super))
8869                 return 1;
8870
8871         features = btrfs_super_incompat_flags(disk_super);
8872         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8873                 mixed = 1;
8874
8875         flags = BTRFS_BLOCK_GROUP_SYSTEM;
8876         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8877         if (ret)
8878                 goto out;
8879
8880         if (mixed) {
8881                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8882                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8883         } else {
8884                 flags = BTRFS_BLOCK_GROUP_METADATA;
8885                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8886                 if (ret)
8887                         goto out;
8888
8889                 flags = BTRFS_BLOCK_GROUP_DATA;
8890                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8891         }
8892 out:
8893         return ret;
8894 }
8895
8896 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8897 {
8898         return unpin_extent_range(root, start, end);
8899 }
8900
8901 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8902                                u64 num_bytes, u64 *actual_bytes)
8903 {
8904         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8905 }
8906
8907 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8908 {
8909         struct btrfs_fs_info *fs_info = root->fs_info;
8910         struct btrfs_block_group_cache *cache = NULL;
8911         u64 group_trimmed;
8912         u64 start;
8913         u64 end;
8914         u64 trimmed = 0;
8915         int ret = 0;
8916
8917         cache = btrfs_lookup_block_group(fs_info, range->start);
8918
8919         while (cache) {
8920                 if (cache->key.objectid >= (range->start + range->len)) {
8921                         btrfs_put_block_group(cache);
8922                         break;
8923                 }
8924
8925                 start = max(range->start, cache->key.objectid);
8926                 end = min(range->start + range->len,
8927                                 cache->key.objectid + cache->key.offset);
8928
8929                 if (end - start >= range->minlen) {
8930                         if (!block_group_cache_done(cache)) {
8931                                 ret = cache_block_group(cache, NULL, root, 0);
8932                                 if (!ret)
8933                                         wait_block_group_cache_done(cache);
8934                         }
8935                         ret = btrfs_trim_block_group(cache,
8936                                                      &group_trimmed,
8937                                                      start,
8938                                                      end,
8939                                                      range->minlen);
8940
8941                         trimmed += group_trimmed;
8942                         if (ret) {
8943                                 btrfs_put_block_group(cache);
8944                                 break;
8945                         }
8946                 }
8947
8948                 cache = next_block_group(fs_info->tree_root, cache);
8949         }
8950
8951         range->len = trimmed;
8952         return ret;
8953 }