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 <linux/ratelimit.h>
  27 #include <linux/percpu_counter.h>
  28 #include "hash.h"
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "print-tree.h"
  32 #include "transaction.h"
  33 #include "volumes.h"
  34 #include "raid56.h"
  35 #include "locking.h"
  36 #include "free-space-cache.h"
  37 #include "math.h"
  38
  39 #undef SCRAMBLE_DELAYED_REFS
  40
  41 /*
  42  * control flags for do_chunk_alloc's force field
  43  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  44  * if we really need one.
  45  *
  46  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  47  * if we have very few chunks already allocated.  This is
  48  * used as part of the clustering code to help make sure
  49  * we have a good pool of storage to cluster in, without
  50  * filling the FS with empty chunks
  51  *
  52  * CHUNK_ALLOC_FORCE means it must try to allocate one
  53  *
  54  */
  55 enum {
  56         CHUNK_ALLOC_NO_FORCE = 0,
  57         CHUNK_ALLOC_LIMITED = 1,
  58         CHUNK_ALLOC_FORCE = 2,
  59 };
  60
  61 /*
  62  * Control how reservations are dealt with.
  63  *
  64  * RESERVE_FREE - freeing a reservation.
  65  * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
  66  *   ENOSPC accounting
  67  * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
  68  *   bytes_may_use as the ENOSPC accounting is done elsewhere
  69  */
  70 enum {
  71         RESERVE_FREE = 0,
  72         RESERVE_ALLOC = 1,
  73         RESERVE_ALLOC_NO_ACCOUNT = 2,
  74 };
  75
  76 static int update_block_group(struct btrfs_root *root,
  77                               u64 bytenr, u64 num_bytes, int alloc);
  78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  79                                 struct btrfs_root *root,
  80                                 u64 bytenr, u64 num_bytes, u64 parent,
  81                                 u64 root_objectid, u64 owner_objectid,
  82                                 u64 owner_offset, int refs_to_drop,
  83                                 struct btrfs_delayed_extent_op *extra_op);
  84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  85                                     struct extent_buffer *leaf,
  86                                     struct btrfs_extent_item *ei);
  87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  88                                       struct btrfs_root *root,
  89                                       u64 parent, u64 root_objectid,
  90                                       u64 flags, u64 owner, u64 offset,
  91                                       struct btrfs_key *ins, int ref_mod);
  92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  93                                      struct btrfs_root *root,
  94                                      u64 parent, u64 root_objectid,
  95                                      u64 flags, struct btrfs_disk_key *key,
  96                                      int level, struct btrfs_key *ins);
  97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  98                           struct btrfs_root *extent_root, u64 flags,
  99                           int force);
 100 static int find_next_key(struct btrfs_path *path, int level,
 101                          struct btrfs_key *key);
 102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
 103                             int dump_block_groups);
 104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
 105                                        u64 num_bytes, int reserve);
 106 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
 107                                u64 num_bytes);
 108 int btrfs_pin_extent(struct btrfs_root *root,
 109                      u64 bytenr, u64 num_bytes, int reserved);
 110
 111 static noinline int
 112 block_group_cache_done(struct btrfs_block_group_cache *cache)
 113 {
 114         smp_mb();
 115         return cache->cached == BTRFS_CACHE_FINISHED ||
 116                 cache->cached == BTRFS_CACHE_ERROR;
 117 }
 118
 119 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 120 {
 121         return (cache->flags & bits) == bits;
 122 }
 123
 124 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 125 {
 126         atomic_inc(&cache->count);
 127 }
 128
 129 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 130 {
 131         if (atomic_dec_and_test(&cache->count)) {
 132                 WARN_ON(cache->pinned > 0);
 133                 WARN_ON(cache->reserved > 0);
 134                 kfree(cache->free_space_ctl);
 135                 kfree(cache);
 136         }
 137 }
 138
 139 /*
 140  * this adds the block group to the fs_info rb tree for the block group
 141  * cache
 142  */
 143 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 144                                 struct btrfs_block_group_cache *block_group)
 145 {
 146         struct rb_node **p;
 147         struct rb_node *parent = NULL;
 148         struct btrfs_block_group_cache *cache;
 149
 150         spin_lock(&info->block_group_cache_lock);
 151         p = &info->block_group_cache_tree.rb_node;
 152
 153         while (*p) {
 154                 parent = *p;
 155                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 156                                  cache_node);
 157                 if (block_group->key.objectid < cache->key.objectid) {
 158                         p = &(*p)->rb_left;
 159                 } else if (block_group->key.objectid > cache->key.objectid) {
 160                         p = &(*p)->rb_right;
 161                 } else {
 162                         spin_unlock(&info->block_group_cache_lock);
 163                         return -EEXIST;
 164                 }
 165         }
 166
 167         rb_link_node(&block_group->cache_node, parent, p);
 168         rb_insert_color(&block_group->cache_node,
 169                         &info->block_group_cache_tree);
 170
 171         if (info->first_logical_byte > block_group->key.objectid)
 172                 info->first_logical_byte = block_group->key.objectid;
 173
 174         spin_unlock(&info->block_group_cache_lock);
 175
 176         return 0;
 177 }
 178
 179 /*
 180  * This will return the block group at or after bytenr if contains is 0, else
 181  * it will return the block group that contains the bytenr
 182  */
 183 static struct btrfs_block_group_cache *
 184 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 185                               int contains)
 186 {
 187         struct btrfs_block_group_cache *cache, *ret = NULL;
 188         struct rb_node *n;
 189         u64 end, start;
 190
 191         spin_lock(&info->block_group_cache_lock);
 192         n = info->block_group_cache_tree.rb_node;
 193
 194         while (n) {
 195                 cache = rb_entry(n, struct btrfs_block_group_cache,
 196                                  cache_node);
 197                 end = cache->key.objectid + cache->key.offset - 1;
 198                 start = cache->key.objectid;
 199
 200                 if (bytenr < start) {
 201                         if (!contains && (!ret || start < ret->key.objectid))
 202                                 ret = cache;
 203                         n = n->rb_left;
 204                 } else if (bytenr > start) {
 205                         if (contains && bytenr <= end) {
 206                                 ret = cache;
 207                                 break;
 208                         }
 209                         n = n->rb_right;
 210                 } else {
 211                         ret = cache;
 212                         break;
 213                 }
 214         }
 215         if (ret) {
 216                 btrfs_get_block_group(ret);
 217                 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
 218                         info->first_logical_byte = ret->key.objectid;
 219         }
 220         spin_unlock(&info->block_group_cache_lock);
 221
 222         return ret;
 223 }
 224
 225 static int add_excluded_extent(struct btrfs_root *root,
 226                                u64 start, u64 num_bytes)
 227 {
 228         u64 end = start + num_bytes - 1;
 229         set_extent_bits(&root->fs_info->freed_extents[0],
 230                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 231         set_extent_bits(&root->fs_info->freed_extents[1],
 232                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 233         return 0;
 234 }
 235
 236 static void free_excluded_extents(struct btrfs_root *root,
 237                                   struct btrfs_block_group_cache *cache)
 238 {
 239         u64 start, end;
 240
 241         start = cache->key.objectid;
 242         end = start + cache->key.offset - 1;
 243
 244         clear_extent_bits(&root->fs_info->freed_extents[0],
 245                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 246         clear_extent_bits(&root->fs_info->freed_extents[1],
 247                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 248 }
 249
 250 static int exclude_super_stripes(struct btrfs_root *root,
 251                                  struct btrfs_block_group_cache *cache)
 252 {
 253         u64 bytenr;
 254         u64 *logical;
 255         int stripe_len;
 256         int i, nr, ret;
 257
 258         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 259                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 260                 cache->bytes_super += stripe_len;
 261                 ret = add_excluded_extent(root, cache->key.objectid,
 262                                           stripe_len);
 263                 if (ret)
 264                         return ret;
 265         }
 266
 267         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 268                 bytenr = btrfs_sb_offset(i);
 269                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 270                                        cache->key.objectid, bytenr,
 271                                        0, &logical, &nr, &stripe_len);
 272                 if (ret)
 273                         return ret;
 274
 275                 while (nr--) {
 276                         u64 start, len;
 277
 278                         if (logical[nr] > cache->key.objectid +
 279                             cache->key.offset)
 280                                 continue;
 281
 282                         if (logical[nr] + stripe_len <= cache->key.objectid)
 283                                 continue;
 284
 285                         start = logical[nr];
 286                         if (start < cache->key.objectid) {
 287                                 start = cache->key.objectid;
 288                                 len = (logical[nr] + stripe_len) - start;
 289                         } else {
 290                                 len = min_t(u64, stripe_len,
 291                                             cache->key.objectid +
 292                                             cache->key.offset - start);
 293                         }
 294
 295                         cache->bytes_super += len;
 296                         ret = add_excluded_extent(root, start, len);
 297                         if (ret) {
 298                                 kfree(logical);
 299                                 return ret;
 300                         }
 301                 }
 302
 303                 kfree(logical);
 304         }
 305         return 0;
 306 }
 307
 308 static struct btrfs_caching_control *
 309 get_caching_control(struct btrfs_block_group_cache *cache)
 310 {
 311         struct btrfs_caching_control *ctl;
 312
 313         spin_lock(&cache->lock);
 314         if (cache->cached != BTRFS_CACHE_STARTED) {
 315                 spin_unlock(&cache->lock);
 316                 return NULL;
 317         }
 318
 319         /* We're loading it the fast way, so we don't have a caching_ctl. */
 320         if (!cache->caching_ctl) {
 321                 spin_unlock(&cache->lock);
 322                 return NULL;
 323         }
 324
 325         ctl = cache->caching_ctl;
 326         atomic_inc(&ctl->count);
 327         spin_unlock(&cache->lock);
 328         return ctl;
 329 }
 330
 331 static void put_caching_control(struct btrfs_caching_control *ctl)
 332 {
 333         if (atomic_dec_and_test(&ctl->count))
 334                 kfree(ctl);
 335 }
 336
 337 /*
 338  * this is only called by cache_block_group, since we could have freed extents
 339  * we need to check the pinned_extents for any extents that can't be used yet
 340  * since their free space will be released as soon as the transaction commits.
 341  */
 342 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 343                               struct btrfs_fs_info *info, u64 start, u64 end)
 344 {
 345         u64 extent_start, extent_end, size, total_added = 0;
 346         int ret;
 347
 348         while (start < end) {
 349                 ret = find_first_extent_bit(info->pinned_extents, start,
 350                                             &extent_start, &extent_end,
 351                                             EXTENT_DIRTY | EXTENT_UPTODATE,
 352                                             NULL);
 353                 if (ret)
 354                         break;
 355
 356                 if (extent_start <= start) {
 357                         start = extent_end + 1;
 358                 } else if (extent_start > start && extent_start < end) {
 359                         size = extent_start - start;
 360                         total_added += size;
 361                         ret = btrfs_add_free_space(block_group, start,
 362                                                    size);
 363                         BUG_ON(ret); /* -ENOMEM or logic error */
 364                         start = extent_end + 1;
 365                 } else {
 366                         break;
 367                 }
 368         }
 369
 370         if (start < end) {
 371                 size = end - start;
 372                 total_added += size;
 373                 ret = btrfs_add_free_space(block_group, start, size);
 374                 BUG_ON(ret); /* -ENOMEM or logic error */
 375         }
 376
 377         return total_added;
 378 }
 379
 380 static noinline void caching_thread(struct btrfs_work *work)
 381 {
 382         struct btrfs_block_group_cache *block_group;
 383         struct btrfs_fs_info *fs_info;
 384         struct btrfs_caching_control *caching_ctl;
 385         struct btrfs_root *extent_root;
 386         struct btrfs_path *path;
 387         struct extent_buffer *leaf;
 388         struct btrfs_key key;
 389         u64 total_found = 0;
 390         u64 last = 0;
 391         u32 nritems;
 392         int ret = -ENOMEM;
 393
 394         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 395         block_group = caching_ctl->block_group;
 396         fs_info = block_group->fs_info;
 397         extent_root = fs_info->extent_root;
 398
 399         path = btrfs_alloc_path();
 400         if (!path)
 401                 goto out;
 402
 403         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 404
 405         /*
 406          * We don't want to deadlock with somebody trying to allocate a new
 407          * extent for the extent root while also trying to search the extent
 408          * root to add free space.  So we skip locking and search the commit
 409          * root, since its read-only
 410          */
 411         path->skip_locking = 1;
 412         path->search_commit_root = 1;
 413         path->reada = 1;
 414
 415         key.objectid = last;
 416         key.offset = 0;
 417         key.type = BTRFS_EXTENT_ITEM_KEY;
 418 again:
 419         mutex_lock(&caching_ctl->mutex);
 420         /* need to make sure the commit_root doesn't disappear */
 421         down_read(&fs_info->extent_commit_sem);
 422
 423 next:
 424         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 425         if (ret < 0)
 426                 goto err;
 427
 428         leaf = path->nodes[0];
 429         nritems = btrfs_header_nritems(leaf);
 430
 431         while (1) {
 432                 if (btrfs_fs_closing(fs_info) > 1) {
 433                         last = (u64)-1;
 434                         break;
 435                 }
 436
 437                 if (path->slots[0] < nritems) {
 438                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 439                 } else {
 440                         ret = find_next_key(path, 0, &key);
 441                         if (ret)
 442                                 break;
 443
 444                         if (need_resched()) {
 445                                 caching_ctl->progress = last;
 446                                 btrfs_release_path(path);
 447                                 up_read(&fs_info->extent_commit_sem);
 448                                 mutex_unlock(&caching_ctl->mutex);
 449                                 cond_resched();
 450                                 goto again;
 451                         }
 452
 453                         ret = btrfs_next_leaf(extent_root, path);
 454                         if (ret < 0)
 455                                 goto err;
 456                         if (ret)
 457                                 break;
 458                         leaf = path->nodes[0];
 459                         nritems = btrfs_header_nritems(leaf);
 460                         continue;
 461                 }
 462
 463                 if (key.objectid < last) {
 464                         key.objectid = last;
 465                         key.offset = 0;
 466                         key.type = BTRFS_EXTENT_ITEM_KEY;
 467
 468                         caching_ctl->progress = last;
 469                         btrfs_release_path(path);
 470                         goto next;
 471                 }
 472
 473                 if (key.objectid < block_group->key.objectid) {
 474                         path->slots[0]++;
 475                         continue;
 476                 }
 477
 478                 if (key.objectid >= block_group->key.objectid +
 479                     block_group->key.offset)
 480                         break;
 481
 482                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
 483                     key.type == BTRFS_METADATA_ITEM_KEY) {
 484                         total_found += add_new_free_space(block_group,
 485                                                           fs_info, last,
 486                                                           key.objectid);
 487                         if (key.type == BTRFS_METADATA_ITEM_KEY)
 488                                 last = key.objectid +
 489                                         fs_info->tree_root->leafsize;
 490                         else
 491                                 last = key.objectid + key.offset;
 492
 493                         if (total_found > (1024 * 1024 * 2)) {
 494                                 total_found = 0;
 495                                 wake_up(&caching_ctl->wait);
 496                         }
 497                 }
 498                 path->slots[0]++;
 499         }
 500         ret = 0;
 501
 502         total_found += add_new_free_space(block_group, fs_info, last,
 503                                           block_group->key.objectid +
 504                                           block_group->key.offset);
 505         caching_ctl->progress = (u64)-1;
 506
 507         spin_lock(&block_group->lock);
 508         block_group->caching_ctl = NULL;
 509         block_group->cached = BTRFS_CACHE_FINISHED;
 510         spin_unlock(&block_group->lock);
 511
 512 err:
 513         btrfs_free_path(path);
 514         up_read(&fs_info->extent_commit_sem);
 515
 516         free_excluded_extents(extent_root, block_group);
 517
 518         mutex_unlock(&caching_ctl->mutex);
 519 out:
 520         if (ret) {
 521                 spin_lock(&block_group->lock);
 522                 block_group->caching_ctl = NULL;
 523                 block_group->cached = BTRFS_CACHE_ERROR;
 524                 spin_unlock(&block_group->lock);
 525         }
 526         wake_up(&caching_ctl->wait);
 527
 528         put_caching_control(caching_ctl);
 529         btrfs_put_block_group(block_group);
 530 }
 531
 532 static int cache_block_group(struct btrfs_block_group_cache *cache,
 533                              int load_cache_only)
 534 {
 535         DEFINE_WAIT(wait);
 536         struct btrfs_fs_info *fs_info = cache->fs_info;
 537         struct btrfs_caching_control *caching_ctl;
 538         int ret = 0;
 539
 540         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 541         if (!caching_ctl)
 542                 return -ENOMEM;
 543
 544         INIT_LIST_HEAD(&caching_ctl->list);
 545         mutex_init(&caching_ctl->mutex);
 546         init_waitqueue_head(&caching_ctl->wait);
 547         caching_ctl->block_group = cache;
 548         caching_ctl->progress = cache->key.objectid;
 549         atomic_set(&caching_ctl->count, 1);
 550         caching_ctl->work.func = caching_thread;
 551
 552         spin_lock(&cache->lock);
 553         /*
 554          * This should be a rare occasion, but this could happen I think in the
 555          * case where one thread starts to load the space cache info, and then
 556          * some other thread starts a transaction commit which tries to do an
 557          * allocation while the other thread is still loading the space cache
 558          * info.  The previous loop should have kept us from choosing this block
 559          * group, but if we've moved to the state where we will wait on caching
 560          * block groups we need to first check if we're doing a fast load here,
 561          * so we can wait for it to finish, otherwise we could end up allocating
 562          * from a block group who's cache gets evicted for one reason or
 563          * another.
 564          */
 565         while (cache->cached == BTRFS_CACHE_FAST) {
 566                 struct btrfs_caching_control *ctl;
 567
 568                 ctl = cache->caching_ctl;
 569                 atomic_inc(&ctl->count);
 570                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
 571                 spin_unlock(&cache->lock);
 572
 573                 schedule();
 574
 575                 finish_wait(&ctl->wait, &wait);
 576                 put_caching_control(ctl);
 577                 spin_lock(&cache->lock);
 578         }
 579
 580         if (cache->cached != BTRFS_CACHE_NO) {
 581                 spin_unlock(&cache->lock);
 582                 kfree(caching_ctl);
 583                 return 0;
 584         }
 585         WARN_ON(cache->caching_ctl);
 586         cache->caching_ctl = caching_ctl;
 587         cache->cached = BTRFS_CACHE_FAST;
 588         spin_unlock(&cache->lock);
 589
 590         if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
 591                 ret = load_free_space_cache(fs_info, cache);
 592
 593                 spin_lock(&cache->lock);
 594                 if (ret == 1) {
 595                         cache->caching_ctl = NULL;
 596                         cache->cached = BTRFS_CACHE_FINISHED;
 597                         cache->last_byte_to_unpin = (u64)-1;
 598                 } else {
 599                         if (load_cache_only) {
 600                                 cache->caching_ctl = NULL;
 601                                 cache->cached = BTRFS_CACHE_NO;
 602                         } else {
 603                                 cache->cached = BTRFS_CACHE_STARTED;
 604                         }
 605                 }
 606                 spin_unlock(&cache->lock);
 607                 wake_up(&caching_ctl->wait);
 608                 if (ret == 1) {
 609                         put_caching_control(caching_ctl);
 610                         free_excluded_extents(fs_info->extent_root, cache);
 611                         return 0;
 612                 }
 613         } else {
 614                 /*
 615                  * We are not going to do the fast caching, set cached to the
 616                  * appropriate value and wakeup any waiters.
 617                  */
 618                 spin_lock(&cache->lock);
 619                 if (load_cache_only) {
 620                         cache->caching_ctl = NULL;
 621                         cache->cached = BTRFS_CACHE_NO;
 622                 } else {
 623                         cache->cached = BTRFS_CACHE_STARTED;
 624                 }
 625                 spin_unlock(&cache->lock);
 626                 wake_up(&caching_ctl->wait);
 627         }
 628
 629         if (load_cache_only) {
 630                 put_caching_control(caching_ctl);
 631                 return 0;
 632         }
 633
 634         down_write(&fs_info->extent_commit_sem);
 635         atomic_inc(&caching_ctl->count);
 636         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 637         up_write(&fs_info->extent_commit_sem);
 638
 639         btrfs_get_block_group(cache);
 640
 641         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 642
 643         return ret;
 644 }
 645
 646 /*
 647  * return the block group that starts at or after bytenr
 648  */
 649 static struct btrfs_block_group_cache *
 650 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 651 {
 652         struct btrfs_block_group_cache *cache;
 653
 654         cache = block_group_cache_tree_search(info, bytenr, 0);
 655
 656         return cache;
 657 }
 658
 659 /*
 660  * return the block group that contains the given bytenr
 661  */
 662 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 663                                                  struct btrfs_fs_info *info,
 664                                                  u64 bytenr)
 665 {
 666         struct btrfs_block_group_cache *cache;
 667
 668         cache = block_group_cache_tree_search(info, bytenr, 1);
 669
 670         return cache;
 671 }
 672
 673 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 674                                                   u64 flags)
 675 {
 676         struct list_head *head = &info->space_info;
 677         struct btrfs_space_info *found;
 678
 679         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 680
 681         rcu_read_lock();
 682         list_for_each_entry_rcu(found, head, list) {
 683                 if (found->flags & flags) {
 684                         rcu_read_unlock();
 685                         return found;
 686                 }
 687         }
 688         rcu_read_unlock();
 689         return NULL;
 690 }
 691
 692 /*
 693  * after adding space to the filesystem, we need to clear the full flags
 694  * on all the space infos.
 695  */
 696 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 697 {
 698         struct list_head *head = &info->space_info;
 699         struct btrfs_space_info *found;
 700
 701         rcu_read_lock();
 702         list_for_each_entry_rcu(found, head, list)
 703                 found->full = 0;
 704         rcu_read_unlock();
 705 }
 706
 707 /* simple helper to search for an existing extent at a given offset */
 708 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 709 {
 710         int ret;
 711         struct btrfs_key key;
 712         struct btrfs_path *path;
 713
 714         path = btrfs_alloc_path();
 715         if (!path)
 716                 return -ENOMEM;
 717
 718         key.objectid = start;
 719         key.offset = len;
 720         key.type = BTRFS_EXTENT_ITEM_KEY;
 721         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 722                                 0, 0);
 723         if (ret > 0) {
 724                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 725                 if (key.objectid == start &&
 726                     key.type == BTRFS_METADATA_ITEM_KEY)
 727                         ret = 0;
 728         }
 729         btrfs_free_path(path);
 730         return ret;
 731 }
 732
 733 /*
 734  * helper function to lookup reference count and flags of a tree block.
 735  *
 736  * the head node for delayed ref is used to store the sum of all the
 737  * reference count modifications queued up in the rbtree. the head
 738  * node may also store the extent flags to set. This way you can check
 739  * to see what the reference count and extent flags would be if all of
 740  * the delayed refs are not processed.
 741  */
 742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 743                              struct btrfs_root *root, u64 bytenr,
 744                              u64 offset, int metadata, u64 *refs, u64 *flags)
 745 {
 746         struct btrfs_delayed_ref_head *head;
 747         struct btrfs_delayed_ref_root *delayed_refs;
 748         struct btrfs_path *path;
 749         struct btrfs_extent_item *ei;
 750         struct extent_buffer *leaf;
 751         struct btrfs_key key;
 752         u32 item_size;
 753         u64 num_refs;
 754         u64 extent_flags;
 755         int ret;
 756
 757         /*
 758          * If we don't have skinny metadata, don't bother doing anything
 759          * different
 760          */
 761         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
 762                 offset = root->leafsize;
 763                 metadata = 0;
 764         }
 765
 766         path = btrfs_alloc_path();
 767         if (!path)
 768                 return -ENOMEM;
 769
 770         if (metadata) {
 771                 key.objectid = bytenr;
 772                 key.type = BTRFS_METADATA_ITEM_KEY;
 773                 key.offset = offset;
 774         } else {
 775                 key.objectid = bytenr;
 776                 key.type = BTRFS_EXTENT_ITEM_KEY;
 777                 key.offset = offset;
 778         }
 779
 780         if (!trans) {
 781                 path->skip_locking = 1;
 782                 path->search_commit_root = 1;
 783         }
 784 again:
 785         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 786                                 &key, path, 0, 0);
 787         if (ret < 0)
 788                 goto out_free;
 789
 790         if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
 791                 metadata = 0;
 792                 if (path->slots[0]) {
 793                         path->slots[0]--;
 794                         btrfs_item_key_to_cpu(path->nodes[0], &key,
 795                                               path->slots[0]);
 796                         if (key.objectid == bytenr &&
 797                             key.type == BTRFS_EXTENT_ITEM_KEY &&
 798                             key.offset == root->leafsize)
 799                                 ret = 0;
 800                 }
 801                 if (ret) {
 802                         key.objectid = bytenr;
 803                         key.type = BTRFS_EXTENT_ITEM_KEY;
 804                         key.offset = root->leafsize;
 805                         btrfs_release_path(path);
 806                         goto again;
 807                 }
 808         }
 809
 810         if (ret == 0) {
 811                 leaf = path->nodes[0];
 812                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 813                 if (item_size >= sizeof(*ei)) {
 814                         ei = btrfs_item_ptr(leaf, path->slots[0],
 815                                             struct btrfs_extent_item);
 816                         num_refs = btrfs_extent_refs(leaf, ei);
 817                         extent_flags = btrfs_extent_flags(leaf, ei);
 818                 } else {
 819 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 820                         struct btrfs_extent_item_v0 *ei0;
 821                         BUG_ON(item_size != sizeof(*ei0));
 822                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 823                                              struct btrfs_extent_item_v0);
 824                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 825                         /* FIXME: this isn't correct for data */
 826                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 827 #else
 828                         BUG();
 829 #endif
 830                 }
 831                 BUG_ON(num_refs == 0);
 832         } else {
 833                 num_refs = 0;
 834                 extent_flags = 0;
 835                 ret = 0;
 836         }
 837
 838         if (!trans)
 839                 goto out;
 840
 841         delayed_refs = &trans->transaction->delayed_refs;
 842         spin_lock(&delayed_refs->lock);
 843         head = btrfs_find_delayed_ref_head(trans, bytenr);
 844         if (head) {
 845                 if (!mutex_trylock(&head->mutex)) {
 846                         atomic_inc(&head->node.refs);
 847                         spin_unlock(&delayed_refs->lock);
 848
 849                         btrfs_release_path(path);
 850
 851                         /*
 852                          * Mutex was contended, block until it's released and try
 853                          * again
 854                          */
 855                         mutex_lock(&head->mutex);
 856                         mutex_unlock(&head->mutex);
 857                         btrfs_put_delayed_ref(&head->node);
 858                         goto again;
 859                 }
 860                 if (head->extent_op && head->extent_op->update_flags)
 861                         extent_flags |= head->extent_op->flags_to_set;
 862                 else
 863                         BUG_ON(num_refs == 0);
 864
 865                 num_refs += head->node.ref_mod;
 866                 mutex_unlock(&head->mutex);
 867         }
 868         spin_unlock(&delayed_refs->lock);
 869 out:
 870         WARN_ON(num_refs == 0);
 871         if (refs)
 872                 *refs = num_refs;
 873         if (flags)
 874                 *flags = extent_flags;
 875 out_free:
 876         btrfs_free_path(path);
 877         return ret;
 878 }
 879
 880 /*
 881  * Back reference rules.  Back refs have three main goals:
 882  *
 883  * 1) differentiate between all holders of references to an extent so that
 884  *    when a reference is dropped we can make sure it was a valid reference
 885  *    before freeing the extent.
 886  *
 887  * 2) Provide enough information to quickly find the holders of an extent
 888  *    if we notice a given block is corrupted or bad.
 889  *
 890  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 891  *    maintenance.  This is actually the same as #2, but with a slightly
 892  *    different use case.
 893  *
 894  * There are two kinds of back refs. The implicit back refs is optimized
 895  * for pointers in non-shared tree blocks. For a given pointer in a block,
 896  * back refs of this kind provide information about the block's owner tree
 897  * and the pointer's key. These information allow us to find the block by
 898  * b-tree searching. The full back refs is for pointers in tree blocks not
 899  * referenced by their owner trees. The location of tree block is recorded
 900  * in the back refs. Actually the full back refs is generic, and can be
 901  * used in all cases the implicit back refs is used. The major shortcoming
 902  * of the full back refs is its overhead. Every time a tree block gets
 903  * COWed, we have to update back refs entry for all pointers in it.
 904  *
 905  * For a newly allocated tree block, we use implicit back refs for
 906  * pointers in it. This means most tree related operations only involve
 907  * implicit back refs. For a tree block created in old transaction, the
 908  * only way to drop a reference to it is COW it. So we can detect the
 909  * event that tree block loses its owner tree's reference and do the
 910  * back refs conversion.
 911  *
 912  * When a tree block is COW'd through a tree, there are four cases:
 913  *
 914  * The reference count of the block is one and the tree is the block's
 915  * owner tree. Nothing to do in this case.
 916  *
 917  * The reference count of the block is one and the tree is not the
 918  * block's owner tree. In this case, full back refs is used for pointers
 919  * in the block. Remove these full back refs, add implicit back refs for
 920  * every pointers in the new block.
 921  *
 922  * The reference count of the block is greater than one and the tree is
 923  * the block's owner tree. In this case, implicit back refs is used for
 924  * pointers in the block. Add full back refs for every pointers in the
 925  * block, increase lower level extents' reference counts. The original
 926  * implicit back refs are entailed to the new block.
 927  *
 928  * The reference count of the block is greater than one and the tree is
 929  * not the block's owner tree. Add implicit back refs for every pointer in
 930  * the new block, increase lower level extents' reference count.
 931  *
 932  * Back Reference Key composing:
 933  *
 934  * The key objectid corresponds to the first byte in the extent,
 935  * The key type is used to differentiate between types of back refs.
 936  * There are different meanings of the key offset for different types
 937  * of back refs.
 938  *
 939  * File extents can be referenced by:
 940  *
 941  * - multiple snapshots, subvolumes, or different generations in one subvol
 942  * - different files inside a single subvolume
 943  * - different offsets inside a file (bookend extents in file.c)
 944  *
 945  * The extent ref structure for the implicit back refs has fields for:
 946  *
 947  * - Objectid of the subvolume root
 948  * - objectid of the file holding the reference
 949  * - original offset in the file
 950  * - how many bookend extents
 951  *
 952  * The key offset for the implicit back refs is hash of the first
 953  * three fields.
 954  *
 955  * The extent ref structure for the full back refs has field for:
 956  *
 957  * - number of pointers in the tree leaf
 958  *
 959  * The key offset for the implicit back refs is the first byte of
 960  * the tree leaf
 961  *
 962  * When a file extent is allocated, The implicit back refs is used.
 963  * the fields are filled in:
 964  *
 965  *     (root_key.objectid, inode objectid, offset in file, 1)
 966  *
 967  * When a file extent is removed file truncation, we find the
 968  * corresponding implicit back refs and check the following fields:
 969  *
 970  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 971  *
 972  * Btree extents can be referenced by:
 973  *
 974  * - Different subvolumes
 975  *
 976  * Both the implicit back refs and the full back refs for tree blocks
 977  * only consist of key. The key offset for the implicit back refs is
 978  * objectid of block's owner tree. The key offset for the full back refs
 979  * is the first byte of parent block.
 980  *
 981  * When implicit back refs is used, information about the lowest key and
 982  * level of the tree block are required. These information are stored in
 983  * tree block info structure.
 984  */
 985
 986 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 987 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 988                                   struct btrfs_root *root,
 989                                   struct btrfs_path *path,
 990                                   u64 owner, u32 extra_size)
 991 {
 992         struct btrfs_extent_item *item;
 993         struct btrfs_extent_item_v0 *ei0;
 994         struct btrfs_extent_ref_v0 *ref0;
 995         struct btrfs_tree_block_info *bi;
 996         struct extent_buffer *leaf;
 997         struct btrfs_key key;
 998         struct btrfs_key found_key;
 999         u32 new_size = sizeof(*item);
1000         u64 refs;
1001         int ret;
1002
1003         leaf = path->nodes[0];
1004         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1005
1006         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1007         ei0 = btrfs_item_ptr(leaf, path->slots[0],
1008                              struct btrfs_extent_item_v0);
1009         refs = btrfs_extent_refs_v0(leaf, ei0);
1010
1011         if (owner == (u64)-1) {
1012                 while (1) {
1013                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1014                                 ret = btrfs_next_leaf(root, path);
1015                                 if (ret < 0)
1016                                         return ret;
1017                                 BUG_ON(ret > 0); /* Corruption */
1018                                 leaf = path->nodes[0];
1019                         }
1020                         btrfs_item_key_to_cpu(leaf, &found_key,
1021                                               path->slots[0]);
1022                         BUG_ON(key.objectid != found_key.objectid);
1023                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1024                                 path->slots[0]++;
1025                                 continue;
1026                         }
1027                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1028                                               struct btrfs_extent_ref_v0);
1029                         owner = btrfs_ref_objectid_v0(leaf, ref0);
1030                         break;
1031                 }
1032         }
1033         btrfs_release_path(path);
1034
1035         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1036                 new_size += sizeof(*bi);
1037
1038         new_size -= sizeof(*ei0);
1039         ret = btrfs_search_slot(trans, root, &key, path,
1040                                 new_size + extra_size, 1);
1041         if (ret < 0)
1042                 return ret;
1043         BUG_ON(ret); /* Corruption */
1044
1045         btrfs_extend_item(root, path, new_size);
1046
1047         leaf = path->nodes[0];
1048         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1049         btrfs_set_extent_refs(leaf, item, refs);
1050         /* FIXME: get real generation */
1051         btrfs_set_extent_generation(leaf, item, 0);
1052         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1053                 btrfs_set_extent_flags(leaf, item,
1054                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1055                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1056                 bi = (struct btrfs_tree_block_info *)(item + 1);
1057                 /* FIXME: get first key of the block */
1058                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1059                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1060         } else {
1061                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1062         }
1063         btrfs_mark_buffer_dirty(leaf);
1064         return 0;
1065 }
1066 #endif
1067
1068 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1069 {
1070         u32 high_crc = ~(u32)0;
1071         u32 low_crc = ~(u32)0;
1072         __le64 lenum;
1073
1074         lenum = cpu_to_le64(root_objectid);
1075         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1076         lenum = cpu_to_le64(owner);
1077         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1078         lenum = cpu_to_le64(offset);
1079         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1080
1081         return ((u64)high_crc << 31) ^ (u64)low_crc;
1082 }
1083
1084 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1085                                      struct btrfs_extent_data_ref *ref)
1086 {
1087         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1088                                     btrfs_extent_data_ref_objectid(leaf, ref),
1089                                     btrfs_extent_data_ref_offset(leaf, ref));
1090 }
1091
1092 static int match_extent_data_ref(struct extent_buffer *leaf,
1093                                  struct btrfs_extent_data_ref *ref,
1094                                  u64 root_objectid, u64 owner, u64 offset)
1095 {
1096         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1097             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1098             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1099                 return 0;
1100         return 1;
1101 }
1102
1103 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1104                                            struct btrfs_root *root,
1105                                            struct btrfs_path *path,
1106                                            u64 bytenr, u64 parent,
1107                                            u64 root_objectid,
1108                                            u64 owner, u64 offset)
1109 {
1110         struct btrfs_key key;
1111         struct btrfs_extent_data_ref *ref;
1112         struct extent_buffer *leaf;
1113         u32 nritems;
1114         int ret;
1115         int recow;
1116         int err = -ENOENT;
1117
1118         key.objectid = bytenr;
1119         if (parent) {
1120                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1121                 key.offset = parent;
1122         } else {
1123                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1124                 key.offset = hash_extent_data_ref(root_objectid,
1125                                                   owner, offset);
1126         }
1127 again:
1128         recow = 0;
1129         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1130         if (ret < 0) {
1131                 err = ret;
1132                 goto fail;
1133         }
1134
1135         if (parent) {
1136                 if (!ret)
1137                         return 0;
1138 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1139                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1140                 btrfs_release_path(path);
1141                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1142                 if (ret < 0) {
1143                         err = ret;
1144                         goto fail;
1145                 }
1146                 if (!ret)
1147                         return 0;
1148 #endif
1149                 goto fail;
1150         }
1151
1152         leaf = path->nodes[0];
1153         nritems = btrfs_header_nritems(leaf);
1154         while (1) {
1155                 if (path->slots[0] >= nritems) {
1156                         ret = btrfs_next_leaf(root, path);
1157                         if (ret < 0)
1158                                 err = ret;
1159                         if (ret)
1160                                 goto fail;
1161
1162                         leaf = path->nodes[0];
1163                         nritems = btrfs_header_nritems(leaf);
1164                         recow = 1;
1165                 }
1166
1167                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1168                 if (key.objectid != bytenr ||
1169                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1170                         goto fail;
1171
1172                 ref = btrfs_item_ptr(leaf, path->slots[0],
1173                                      struct btrfs_extent_data_ref);
1174
1175                 if (match_extent_data_ref(leaf, ref, root_objectid,
1176                                           owner, offset)) {
1177                         if (recow) {
1178                                 btrfs_release_path(path);
1179                                 goto again;
1180                         }
1181                         err = 0;
1182                         break;
1183                 }
1184                 path->slots[0]++;
1185         }
1186 fail:
1187         return err;
1188 }
1189
1190 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1191                                            struct btrfs_root *root,
1192                                            struct btrfs_path *path,
1193                                            u64 bytenr, u64 parent,
1194                                            u64 root_objectid, u64 owner,
1195                                            u64 offset, int refs_to_add)
1196 {
1197         struct btrfs_key key;
1198         struct extent_buffer *leaf;
1199         u32 size;
1200         u32 num_refs;
1201         int ret;
1202
1203         key.objectid = bytenr;
1204         if (parent) {
1205                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1206                 key.offset = parent;
1207                 size = sizeof(struct btrfs_shared_data_ref);
1208         } else {
1209                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1210                 key.offset = hash_extent_data_ref(root_objectid,
1211                                                   owner, offset);
1212                 size = sizeof(struct btrfs_extent_data_ref);
1213         }
1214
1215         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1216         if (ret && ret != -EEXIST)
1217                 goto fail;
1218
1219         leaf = path->nodes[0];
1220         if (parent) {
1221                 struct btrfs_shared_data_ref *ref;
1222                 ref = btrfs_item_ptr(leaf, path->slots[0],
1223                                      struct btrfs_shared_data_ref);
1224                 if (ret == 0) {
1225                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1226                 } else {
1227                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1228                         num_refs += refs_to_add;
1229                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1230                 }
1231         } else {
1232                 struct btrfs_extent_data_ref *ref;
1233                 while (ret == -EEXIST) {
1234                         ref = btrfs_item_ptr(leaf, path->slots[0],
1235                                              struct btrfs_extent_data_ref);
1236                         if (match_extent_data_ref(leaf, ref, root_objectid,
1237                                                   owner, offset))
1238                                 break;
1239                         btrfs_release_path(path);
1240                         key.offset++;
1241                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1242                                                       size);
1243                         if (ret && ret != -EEXIST)
1244                                 goto fail;
1245
1246                         leaf = path->nodes[0];
1247                 }
1248                 ref = btrfs_item_ptr(leaf, path->slots[0],
1249                                      struct btrfs_extent_data_ref);
1250                 if (ret == 0) {
1251                         btrfs_set_extent_data_ref_root(leaf, ref,
1252                                                        root_objectid);
1253                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1254                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1255                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1256                 } else {
1257                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1258                         num_refs += refs_to_add;
1259                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1260                 }
1261         }
1262         btrfs_mark_buffer_dirty(leaf);
1263         ret = 0;
1264 fail:
1265         btrfs_release_path(path);
1266         return ret;
1267 }
1268
1269 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1270                                            struct btrfs_root *root,
1271                                            struct btrfs_path *path,
1272                                            int refs_to_drop)
1273 {
1274         struct btrfs_key key;
1275         struct btrfs_extent_data_ref *ref1 = NULL;
1276         struct btrfs_shared_data_ref *ref2 = NULL;
1277         struct extent_buffer *leaf;
1278         u32 num_refs = 0;
1279         int ret = 0;
1280
1281         leaf = path->nodes[0];
1282         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1283
1284         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1285                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1286                                       struct btrfs_extent_data_ref);
1287                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1288         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1289                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1290                                       struct btrfs_shared_data_ref);
1291                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1292 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1293         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1294                 struct btrfs_extent_ref_v0 *ref0;
1295                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1296                                       struct btrfs_extent_ref_v0);
1297                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1298 #endif
1299         } else {
1300                 BUG();
1301         }
1302
1303         BUG_ON(num_refs < refs_to_drop);
1304         num_refs -= refs_to_drop;
1305
1306         if (num_refs == 0) {
1307                 ret = btrfs_del_item(trans, root, path);
1308         } else {
1309                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1310                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1311                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1312                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1313 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1314                 else {
1315                         struct btrfs_extent_ref_v0 *ref0;
1316                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1317                                         struct btrfs_extent_ref_v0);
1318                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1319                 }
1320 #endif
1321                 btrfs_mark_buffer_dirty(leaf);
1322         }
1323         return ret;
1324 }
1325
1326 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1327                                           struct btrfs_path *path,
1328                                           struct btrfs_extent_inline_ref *iref)
1329 {
1330         struct btrfs_key key;
1331         struct extent_buffer *leaf;
1332         struct btrfs_extent_data_ref *ref1;
1333         struct btrfs_shared_data_ref *ref2;
1334         u32 num_refs = 0;
1335
1336         leaf = path->nodes[0];
1337         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1338         if (iref) {
1339                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1340                     BTRFS_EXTENT_DATA_REF_KEY) {
1341                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1342                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1343                 } else {
1344                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1345                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1346                 }
1347         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1348                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1349                                       struct btrfs_extent_data_ref);
1350                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1351         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1352                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1353                                       struct btrfs_shared_data_ref);
1354                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1355 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1356         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1357                 struct btrfs_extent_ref_v0 *ref0;
1358                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1359                                       struct btrfs_extent_ref_v0);
1360                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1361 #endif
1362         } else {
1363                 WARN_ON(1);
1364         }
1365         return num_refs;
1366 }
1367
1368 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1369                                           struct btrfs_root *root,
1370                                           struct btrfs_path *path,
1371                                           u64 bytenr, u64 parent,
1372                                           u64 root_objectid)
1373 {
1374         struct btrfs_key key;
1375         int ret;
1376
1377         key.objectid = bytenr;
1378         if (parent) {
1379                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1380                 key.offset = parent;
1381         } else {
1382                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1383                 key.offset = root_objectid;
1384         }
1385
1386         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1387         if (ret > 0)
1388                 ret = -ENOENT;
1389 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1390         if (ret == -ENOENT && parent) {
1391                 btrfs_release_path(path);
1392                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1393                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1394                 if (ret > 0)
1395                         ret = -ENOENT;
1396         }
1397 #endif
1398         return ret;
1399 }
1400
1401 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1402                                           struct btrfs_root *root,
1403                                           struct btrfs_path *path,
1404                                           u64 bytenr, u64 parent,
1405                                           u64 root_objectid)
1406 {
1407         struct btrfs_key key;
1408         int ret;
1409
1410         key.objectid = bytenr;
1411         if (parent) {
1412                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1413                 key.offset = parent;
1414         } else {
1415                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1416                 key.offset = root_objectid;
1417         }
1418
1419         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1420         btrfs_release_path(path);
1421         return ret;
1422 }
1423
1424 static inline int extent_ref_type(u64 parent, u64 owner)
1425 {
1426         int type;
1427         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1428                 if (parent > 0)
1429                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1430                 else
1431                         type = BTRFS_TREE_BLOCK_REF_KEY;
1432         } else {
1433                 if (parent > 0)
1434                         type = BTRFS_SHARED_DATA_REF_KEY;
1435                 else
1436                         type = BTRFS_EXTENT_DATA_REF_KEY;
1437         }
1438         return type;
1439 }
1440
1441 static int find_next_key(struct btrfs_path *path, int level,
1442                          struct btrfs_key *key)
1443
1444 {
1445         for (; level < BTRFS_MAX_LEVEL; level++) {
1446                 if (!path->nodes[level])
1447                         break;
1448                 if (path->slots[level] + 1 >=
1449                     btrfs_header_nritems(path->nodes[level]))
1450                         continue;
1451                 if (level == 0)
1452                         btrfs_item_key_to_cpu(path->nodes[level], key,
1453                                               path->slots[level] + 1);
1454                 else
1455                         btrfs_node_key_to_cpu(path->nodes[level], key,
1456                                               path->slots[level] + 1);
1457                 return 0;
1458         }
1459         return 1;
1460 }
1461
1462 /*
1463  * look for inline back ref. if back ref is found, *ref_ret is set
1464  * to the address of inline back ref, and 0 is returned.
1465  *
1466  * if back ref isn't found, *ref_ret is set to the address where it
1467  * should be inserted, and -ENOENT is returned.
1468  *
1469  * if insert is true and there are too many inline back refs, the path
1470  * points to the extent item, and -EAGAIN is returned.
1471  *
1472  * NOTE: inline back refs are ordered in the same way that back ref
1473  *       items in the tree are ordered.
1474  */
1475 static noinline_for_stack
1476 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1477                                  struct btrfs_root *root,
1478                                  struct btrfs_path *path,
1479                                  struct btrfs_extent_inline_ref **ref_ret,
1480                                  u64 bytenr, u64 num_bytes,
1481                                  u64 parent, u64 root_objectid,
1482                                  u64 owner, u64 offset, int insert)
1483 {
1484         struct btrfs_key key;
1485         struct extent_buffer *leaf;
1486         struct btrfs_extent_item *ei;
1487         struct btrfs_extent_inline_ref *iref;
1488         u64 flags;
1489         u64 item_size;
1490         unsigned long ptr;
1491         unsigned long end;
1492         int extra_size;
1493         int type;
1494         int want;
1495         int ret;
1496         int err = 0;
1497         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1498                                                  SKINNY_METADATA);
1499
1500         key.objectid = bytenr;
1501         key.type = BTRFS_EXTENT_ITEM_KEY;
1502         key.offset = num_bytes;
1503
1504         want = extent_ref_type(parent, owner);
1505         if (insert) {
1506                 extra_size = btrfs_extent_inline_ref_size(want);
1507                 path->keep_locks = 1;
1508         } else
1509                 extra_size = -1;
1510
1511         /*
1512          * Owner is our parent level, so we can just add one to get the level
1513          * for the block we are interested in.
1514          */
1515         if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1516                 key.type = BTRFS_METADATA_ITEM_KEY;
1517                 key.offset = owner;
1518         }
1519
1520 again:
1521         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1522         if (ret < 0) {
1523                 err = ret;
1524                 goto out;
1525         }
1526
1527         /*
1528          * We may be a newly converted file system which still has the old fat
1529          * extent entries for metadata, so try and see if we have one of those.
1530          */
1531         if (ret > 0 && skinny_metadata) {
1532                 skinny_metadata = false;
1533                 if (path->slots[0]) {
1534                         path->slots[0]--;
1535                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1536                                               path->slots[0]);
1537                         if (key.objectid == bytenr &&
1538                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1539                             key.offset == num_bytes)
1540                                 ret = 0;
1541                 }
1542                 if (ret) {
1543                         key.type = BTRFS_EXTENT_ITEM_KEY;
1544                         key.offset = num_bytes;
1545                         btrfs_release_path(path);
1546                         goto again;
1547                 }
1548         }
1549
1550         if (ret && !insert) {
1551                 err = -ENOENT;
1552                 goto out;
1553         } else if (ret) {
1554                 err = -EIO;
1555                 WARN_ON(1);
1556                 goto out;
1557         }
1558
1559         leaf = path->nodes[0];
1560         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1561 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1562         if (item_size < sizeof(*ei)) {
1563                 if (!insert) {
1564                         err = -ENOENT;
1565                         goto out;
1566                 }
1567                 ret = convert_extent_item_v0(trans, root, path, owner,
1568                                              extra_size);
1569                 if (ret < 0) {
1570                         err = ret;
1571                         goto out;
1572                 }
1573                 leaf = path->nodes[0];
1574                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1575         }
1576 #endif
1577         BUG_ON(item_size < sizeof(*ei));
1578
1579         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1580         flags = btrfs_extent_flags(leaf, ei);
1581
1582         ptr = (unsigned long)(ei + 1);
1583         end = (unsigned long)ei + item_size;
1584
1585         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1586                 ptr += sizeof(struct btrfs_tree_block_info);
1587                 BUG_ON(ptr > end);
1588         }
1589
1590         err = -ENOENT;
1591         while (1) {
1592                 if (ptr >= end) {
1593                         WARN_ON(ptr > end);
1594                         break;
1595                 }
1596                 iref = (struct btrfs_extent_inline_ref *)ptr;
1597                 type = btrfs_extent_inline_ref_type(leaf, iref);
1598                 if (want < type)
1599                         break;
1600                 if (want > type) {
1601                         ptr += btrfs_extent_inline_ref_size(type);
1602                         continue;
1603                 }
1604
1605                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1606                         struct btrfs_extent_data_ref *dref;
1607                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1608                         if (match_extent_data_ref(leaf, dref, root_objectid,
1609                                                   owner, offset)) {
1610                                 err = 0;
1611                                 break;
1612                         }
1613                         if (hash_extent_data_ref_item(leaf, dref) <
1614                             hash_extent_data_ref(root_objectid, owner, offset))
1615                                 break;
1616                 } else {
1617                         u64 ref_offset;
1618                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1619                         if (parent > 0) {
1620                                 if (parent == ref_offset) {
1621                                         err = 0;
1622                                         break;
1623                                 }
1624                                 if (ref_offset < parent)
1625                                         break;
1626                         } else {
1627                                 if (root_objectid == ref_offset) {
1628                                         err = 0;
1629                                         break;
1630                                 }
1631                                 if (ref_offset < root_objectid)
1632                                         break;
1633                         }
1634                 }
1635                 ptr += btrfs_extent_inline_ref_size(type);
1636         }
1637         if (err == -ENOENT && insert) {
1638                 if (item_size + extra_size >=
1639                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1640                         err = -EAGAIN;
1641                         goto out;
1642                 }
1643                 /*
1644                  * To add new inline back ref, we have to make sure
1645                  * there is no corresponding back ref item.
1646                  * For simplicity, we just do not add new inline back
1647                  * ref if there is any kind of item for this block
1648                  */
1649                 if (find_next_key(path, 0, &key) == 0 &&
1650                     key.objectid == bytenr &&
1651                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1652                         err = -EAGAIN;
1653                         goto out;
1654                 }
1655         }
1656         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1657 out:
1658         if (insert) {
1659                 path->keep_locks = 0;
1660                 btrfs_unlock_up_safe(path, 1);
1661         }
1662         return err;
1663 }
1664
1665 /*
1666  * helper to add new inline back ref
1667  */
1668 static noinline_for_stack
1669 void setup_inline_extent_backref(struct btrfs_root *root,
1670                                  struct btrfs_path *path,
1671                                  struct btrfs_extent_inline_ref *iref,
1672                                  u64 parent, u64 root_objectid,
1673                                  u64 owner, u64 offset, int refs_to_add,
1674                                  struct btrfs_delayed_extent_op *extent_op)
1675 {
1676         struct extent_buffer *leaf;
1677         struct btrfs_extent_item *ei;
1678         unsigned long ptr;
1679         unsigned long end;
1680         unsigned long item_offset;
1681         u64 refs;
1682         int size;
1683         int type;
1684
1685         leaf = path->nodes[0];
1686         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1687         item_offset = (unsigned long)iref - (unsigned long)ei;
1688
1689         type = extent_ref_type(parent, owner);
1690         size = btrfs_extent_inline_ref_size(type);
1691
1692         btrfs_extend_item(root, path, size);
1693
1694         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1695         refs = btrfs_extent_refs(leaf, ei);
1696         refs += refs_to_add;
1697         btrfs_set_extent_refs(leaf, ei, refs);
1698         if (extent_op)
1699                 __run_delayed_extent_op(extent_op, leaf, ei);
1700
1701         ptr = (unsigned long)ei + item_offset;
1702         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1703         if (ptr < end - size)
1704                 memmove_extent_buffer(leaf, ptr + size, ptr,
1705                                       end - size - ptr);
1706
1707         iref = (struct btrfs_extent_inline_ref *)ptr;
1708         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1709         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1710                 struct btrfs_extent_data_ref *dref;
1711                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1712                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1713                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1714                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1715                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1716         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1717                 struct btrfs_shared_data_ref *sref;
1718                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1719                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1720                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1721         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1722                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1723         } else {
1724                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1725         }
1726         btrfs_mark_buffer_dirty(leaf);
1727 }
1728
1729 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1730                                  struct btrfs_root *root,
1731                                  struct btrfs_path *path,
1732                                  struct btrfs_extent_inline_ref **ref_ret,
1733                                  u64 bytenr, u64 num_bytes, u64 parent,
1734                                  u64 root_objectid, u64 owner, u64 offset)
1735 {
1736         int ret;
1737
1738         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1739                                            bytenr, num_bytes, parent,
1740                                            root_objectid, owner, offset, 0);
1741         if (ret != -ENOENT)
1742                 return ret;
1743
1744         btrfs_release_path(path);
1745         *ref_ret = NULL;
1746
1747         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1748                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1749                                             root_objectid);
1750         } else {
1751                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1752                                              root_objectid, owner, offset);
1753         }
1754         return ret;
1755 }
1756
1757 /*
1758  * helper to update/remove inline back ref
1759  */
1760 static noinline_for_stack
1761 void update_inline_extent_backref(struct btrfs_root *root,
1762                                   struct btrfs_path *path,
1763                                   struct btrfs_extent_inline_ref *iref,
1764                                   int refs_to_mod,
1765                                   struct btrfs_delayed_extent_op *extent_op)
1766 {
1767         struct extent_buffer *leaf;
1768         struct btrfs_extent_item *ei;
1769         struct btrfs_extent_data_ref *dref = NULL;
1770         struct btrfs_shared_data_ref *sref = NULL;
1771         unsigned long ptr;
1772         unsigned long end;
1773         u32 item_size;
1774         int size;
1775         int type;
1776         u64 refs;
1777
1778         leaf = path->nodes[0];
1779         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1780         refs = btrfs_extent_refs(leaf, ei);
1781         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1782         refs += refs_to_mod;
1783         btrfs_set_extent_refs(leaf, ei, refs);
1784         if (extent_op)
1785                 __run_delayed_extent_op(extent_op, leaf, ei);
1786
1787         type = btrfs_extent_inline_ref_type(leaf, iref);
1788
1789         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1790                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1791                 refs = btrfs_extent_data_ref_count(leaf, dref);
1792         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1793                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1794                 refs = btrfs_shared_data_ref_count(leaf, sref);
1795         } else {
1796                 refs = 1;
1797                 BUG_ON(refs_to_mod != -1);
1798         }
1799
1800         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1801         refs += refs_to_mod;
1802
1803         if (refs > 0) {
1804                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1805                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1806                 else
1807                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1808         } else {
1809                 size =  btrfs_extent_inline_ref_size(type);
1810                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1811                 ptr = (unsigned long)iref;
1812                 end = (unsigned long)ei + item_size;
1813                 if (ptr + size < end)
1814                         memmove_extent_buffer(leaf, ptr, ptr + size,
1815                                               end - ptr - size);
1816                 item_size -= size;
1817                 btrfs_truncate_item(root, path, item_size, 1);
1818         }
1819         btrfs_mark_buffer_dirty(leaf);
1820 }
1821
1822 static noinline_for_stack
1823 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1824                                  struct btrfs_root *root,
1825                                  struct btrfs_path *path,
1826                                  u64 bytenr, u64 num_bytes, u64 parent,
1827                                  u64 root_objectid, u64 owner,
1828                                  u64 offset, int refs_to_add,
1829                                  struct btrfs_delayed_extent_op *extent_op)
1830 {
1831         struct btrfs_extent_inline_ref *iref;
1832         int ret;
1833
1834         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1835                                            bytenr, num_bytes, parent,
1836                                            root_objectid, owner, offset, 1);
1837         if (ret == 0) {
1838                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1839                 update_inline_extent_backref(root, path, iref,
1840                                              refs_to_add, extent_op);
1841         } else if (ret == -ENOENT) {
1842                 setup_inline_extent_backref(root, path, iref, parent,
1843                                             root_objectid, owner, offset,
1844                                             refs_to_add, extent_op);
1845                 ret = 0;
1846         }
1847         return ret;
1848 }
1849
1850 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1851                                  struct btrfs_root *root,
1852                                  struct btrfs_path *path,
1853                                  u64 bytenr, u64 parent, u64 root_objectid,
1854                                  u64 owner, u64 offset, int refs_to_add)
1855 {
1856         int ret;
1857         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1858                 BUG_ON(refs_to_add != 1);
1859                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1860                                             parent, root_objectid);
1861         } else {
1862                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1863                                              parent, root_objectid,
1864                                              owner, offset, refs_to_add);
1865         }
1866         return ret;
1867 }
1868
1869 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1870                                  struct btrfs_root *root,
1871                                  struct btrfs_path *path,
1872                                  struct btrfs_extent_inline_ref *iref,
1873                                  int refs_to_drop, int is_data)
1874 {
1875         int ret = 0;
1876
1877         BUG_ON(!is_data && refs_to_drop != 1);
1878         if (iref) {
1879                 update_inline_extent_backref(root, path, iref,
1880                                              -refs_to_drop, NULL);
1881         } else if (is_data) {
1882                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1883         } else {
1884                 ret = btrfs_del_item(trans, root, path);
1885         }
1886         return ret;
1887 }
1888
1889 static int btrfs_issue_discard(struct block_device *bdev,
1890                                 u64 start, u64 len)
1891 {
1892         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1893 }
1894
1895 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1896                                 u64 num_bytes, u64 *actual_bytes)
1897 {
1898         int ret;
1899         u64 discarded_bytes = 0;
1900         struct btrfs_bio *bbio = NULL;
1901
1902
1903         /* Tell the block device(s) that the sectors can be discarded */
1904         ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1905                               bytenr, &num_bytes, &bbio, 0);
1906         /* Error condition is -ENOMEM */
1907         if (!ret) {
1908                 struct btrfs_bio_stripe *stripe = bbio->stripes;
1909                 int i;
1910
1911
1912                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1913                         if (!stripe->dev->can_discard)
1914                                 continue;
1915
1916                         ret = btrfs_issue_discard(stripe->dev->bdev,
1917                                                   stripe->physical,
1918                                                   stripe->length);
1919                         if (!ret)
1920                                 discarded_bytes += stripe->length;
1921                         else if (ret != -EOPNOTSUPP)
1922                                 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1923
1924                         /*
1925                          * Just in case we get back EOPNOTSUPP for some reason,
1926                          * just ignore the return value so we don't screw up
1927                          * people calling discard_extent.
1928                          */
1929                         ret = 0;
1930                 }
1931                 kfree(bbio);
1932         }
1933
1934         if (actual_bytes)
1935                 *actual_bytes = discarded_bytes;
1936
1937
1938         if (ret == -EOPNOTSUPP)
1939                 ret = 0;
1940         return ret;
1941 }
1942
1943 /* Can return -ENOMEM */
1944 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1945                          struct btrfs_root *root,
1946                          u64 bytenr, u64 num_bytes, u64 parent,
1947                          u64 root_objectid, u64 owner, u64 offset, int for_cow)
1948 {
1949         int ret;
1950         struct btrfs_fs_info *fs_info = root->fs_info;
1951
1952         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1953                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1954
1955         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1956                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1957                                         num_bytes,
1958                                         parent, root_objectid, (int)owner,
1959                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1960         } else {
1961                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1962                                         num_bytes,
1963                                         parent, root_objectid, owner, offset,
1964                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1965         }
1966         return ret;
1967 }
1968
1969 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1970                                   struct btrfs_root *root,
1971                                   u64 bytenr, u64 num_bytes,
1972                                   u64 parent, u64 root_objectid,
1973                                   u64 owner, u64 offset, int refs_to_add,
1974                                   struct btrfs_delayed_extent_op *extent_op)
1975 {
1976         struct btrfs_path *path;
1977         struct extent_buffer *leaf;
1978         struct btrfs_extent_item *item;
1979         u64 refs;
1980         int ret;
1981
1982         path = btrfs_alloc_path();
1983         if (!path)
1984                 return -ENOMEM;
1985
1986         path->reada = 1;
1987         path->leave_spinning = 1;
1988         /* this will setup the path even if it fails to insert the back ref */
1989         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1990                                            path, bytenr, num_bytes, parent,
1991                                            root_objectid, owner, offset,
1992                                            refs_to_add, extent_op);
1993         if (ret != -EAGAIN)
1994                 goto out;
1995
1996         leaf = path->nodes[0];
1997         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1998         refs = btrfs_extent_refs(leaf, item);
1999         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2000         if (extent_op)
2001                 __run_delayed_extent_op(extent_op, leaf, item);
2002
2003         btrfs_mark_buffer_dirty(leaf);
2004         btrfs_release_path(path);
2005
2006         path->reada = 1;
2007         path->leave_spinning = 1;
2008
2009         /* now insert the actual backref */
2010         ret = insert_extent_backref(trans, root->fs_info->extent_root,
2011                                     path, bytenr, parent, root_objectid,
2012                                     owner, offset, refs_to_add);
2013         if (ret)
2014                 btrfs_abort_transaction(trans, root, ret);
2015 out:
2016         btrfs_free_path(path);
2017         return ret;
2018 }
2019
2020 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2021                                 struct btrfs_root *root,
2022                                 struct btrfs_delayed_ref_node *node,
2023                                 struct btrfs_delayed_extent_op *extent_op,
2024                                 int insert_reserved)
2025 {
2026         int ret = 0;
2027         struct btrfs_delayed_data_ref *ref;
2028         struct btrfs_key ins;
2029         u64 parent = 0;
2030         u64 ref_root = 0;
2031         u64 flags = 0;
2032
2033         ins.objectid = node->bytenr;
2034         ins.offset = node->num_bytes;
2035         ins.type = BTRFS_EXTENT_ITEM_KEY;
2036
2037         ref = btrfs_delayed_node_to_data_ref(node);
2038         trace_run_delayed_data_ref(node, ref, node->action);
2039
2040         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2041                 parent = ref->parent;
2042         else
2043                 ref_root = ref->root;
2044
2045         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2046                 if (extent_op)
2047                         flags |= extent_op->flags_to_set;
2048                 ret = alloc_reserved_file_extent(trans, root,
2049                                                  parent, ref_root, flags,
2050                                                  ref->objectid, ref->offset,
2051                                                  &ins, node->ref_mod);
2052         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2053                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2054                                              node->num_bytes, parent,
2055                                              ref_root, ref->objectid,
2056                                              ref->offset, node->ref_mod,
2057                                              extent_op);
2058         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2059                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2060                                           node->num_bytes, parent,
2061                                           ref_root, ref->objectid,
2062                                           ref->offset, node->ref_mod,
2063                                           extent_op);
2064         } else {
2065                 BUG();
2066         }
2067         return ret;
2068 }
2069
2070 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2071                                     struct extent_buffer *leaf,
2072                                     struct btrfs_extent_item *ei)
2073 {
2074         u64 flags = btrfs_extent_flags(leaf, ei);
2075         if (extent_op->update_flags) {
2076                 flags |= extent_op->flags_to_set;
2077                 btrfs_set_extent_flags(leaf, ei, flags);
2078         }
2079
2080         if (extent_op->update_key) {
2081                 struct btrfs_tree_block_info *bi;
2082                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2083                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2084                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2085         }
2086 }
2087
2088 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2089                                  struct btrfs_root *root,
2090                                  struct btrfs_delayed_ref_node *node,
2091                                  struct btrfs_delayed_extent_op *extent_op)
2092 {
2093         struct btrfs_key key;
2094         struct btrfs_path *path;
2095         struct btrfs_extent_item *ei;
2096         struct extent_buffer *leaf;
2097         u32 item_size;
2098         int ret;
2099         int err = 0;
2100         int metadata = !extent_op->is_data;
2101
2102         if (trans->aborted)
2103                 return 0;
2104
2105         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2106                 metadata = 0;
2107
2108         path = btrfs_alloc_path();
2109         if (!path)
2110                 return -ENOMEM;
2111
2112         key.objectid = node->bytenr;
2113
2114         if (metadata) {
2115                 key.type = BTRFS_METADATA_ITEM_KEY;
2116                 key.offset = extent_op->level;
2117         } else {
2118                 key.type = BTRFS_EXTENT_ITEM_KEY;
2119                 key.offset = node->num_bytes;
2120         }
2121
2122 again:
2123         path->reada = 1;
2124         path->leave_spinning = 1;
2125         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2126                                 path, 0, 1);
2127         if (ret < 0) {
2128                 err = ret;
2129                 goto out;
2130         }
2131         if (ret > 0) {
2132                 if (metadata) {
2133                         btrfs_release_path(path);
2134                         metadata = 0;
2135
2136                         key.offset = node->num_bytes;
2137                         key.type = BTRFS_EXTENT_ITEM_KEY;
2138                         goto again;
2139                 }
2140                 err = -EIO;
2141                 goto out;
2142         }
2143
2144         leaf = path->nodes[0];
2145         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2146 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2147         if (item_size < sizeof(*ei)) {
2148                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2149                                              path, (u64)-1, 0);
2150                 if (ret < 0) {
2151                         err = ret;
2152                         goto out;
2153                 }
2154                 leaf = path->nodes[0];
2155                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2156         }
2157 #endif
2158         BUG_ON(item_size < sizeof(*ei));
2159         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2160         __run_delayed_extent_op(extent_op, leaf, ei);
2161
2162         btrfs_mark_buffer_dirty(leaf);
2163 out:
2164         btrfs_free_path(path);
2165         return err;
2166 }
2167
2168 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2169                                 struct btrfs_root *root,
2170                                 struct btrfs_delayed_ref_node *node,
2171                                 struct btrfs_delayed_extent_op *extent_op,
2172                                 int insert_reserved)
2173 {
2174         int ret = 0;
2175         struct btrfs_delayed_tree_ref *ref;
2176         struct btrfs_key ins;
2177         u64 parent = 0;
2178         u64 ref_root = 0;
2179         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2180                                                  SKINNY_METADATA);
2181
2182         ref = btrfs_delayed_node_to_tree_ref(node);
2183         trace_run_delayed_tree_ref(node, ref, node->action);
2184
2185         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2186                 parent = ref->parent;
2187         else
2188                 ref_root = ref->root;
2189
2190         ins.objectid = node->bytenr;
2191         if (skinny_metadata) {
2192                 ins.offset = ref->level;
2193                 ins.type = BTRFS_METADATA_ITEM_KEY;
2194         } else {
2195                 ins.offset = node->num_bytes;
2196                 ins.type = BTRFS_EXTENT_ITEM_KEY;
2197         }
2198
2199         BUG_ON(node->ref_mod != 1);
2200         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2201                 BUG_ON(!extent_op || !extent_op->update_flags);
2202                 ret = alloc_reserved_tree_block(trans, root,
2203                                                 parent, ref_root,
2204                                                 extent_op->flags_to_set,
2205                                                 &extent_op->key,
2206                                                 ref->level, &ins);
2207         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2208                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2209                                              node->num_bytes, parent, ref_root,
2210                                              ref->level, 0, 1, extent_op);
2211         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2212                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2213                                           node->num_bytes, parent, ref_root,
2214                                           ref->level, 0, 1, extent_op);
2215         } else {
2216                 BUG();
2217         }
2218         return ret;
2219 }
2220
2221 /* helper function to actually process a single delayed ref entry */
2222 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2223                                struct btrfs_root *root,
2224                                struct btrfs_delayed_ref_node *node,
2225                                struct btrfs_delayed_extent_op *extent_op,
2226                                int insert_reserved)
2227 {
2228         int ret = 0;
2229
2230         if (trans->aborted) {
2231                 if (insert_reserved)
2232                         btrfs_pin_extent(root, node->bytenr,
2233                                          node->num_bytes, 1);
2234                 return 0;
2235         }
2236
2237         if (btrfs_delayed_ref_is_head(node)) {
2238                 struct btrfs_delayed_ref_head *head;
2239                 /*
2240                  * we've hit the end of the chain and we were supposed
2241                  * to insert this extent into the tree.  But, it got
2242                  * deleted before we ever needed to insert it, so all
2243                  * we have to do is clean up the accounting
2244                  */
2245                 BUG_ON(extent_op);
2246                 head = btrfs_delayed_node_to_head(node);
2247                 trace_run_delayed_ref_head(node, head, node->action);
2248
2249                 if (insert_reserved) {
2250                         btrfs_pin_extent(root, node->bytenr,
2251                                          node->num_bytes, 1);
2252                         if (head->is_data) {
2253                                 ret = btrfs_del_csums(trans, root,
2254                                                       node->bytenr,
2255                                                       node->num_bytes);
2256                         }
2257                 }
2258                 return ret;
2259         }
2260
2261         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2262             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2263                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2264                                            insert_reserved);
2265         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2266                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2267                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2268                                            insert_reserved);
2269         else
2270                 BUG();
2271         return ret;
2272 }
2273
2274 static noinline struct btrfs_delayed_ref_node *
2275 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2276 {
2277         struct rb_node *node;
2278         struct btrfs_delayed_ref_node *ref;
2279         int action = BTRFS_ADD_DELAYED_REF;
2280 again:
2281         /*
2282          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2283          * this prevents ref count from going down to zero when
2284          * there still are pending delayed ref.
2285          */
2286         node = rb_prev(&head->node.rb_node);
2287         while (1) {
2288                 if (!node)
2289                         break;
2290                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2291                                 rb_node);
2292                 if (ref->bytenr != head->node.bytenr)
2293                         break;
2294                 if (ref->action == action)
2295                         return ref;
2296                 node = rb_prev(node);
2297         }
2298         if (action == BTRFS_ADD_DELAYED_REF) {
2299                 action = BTRFS_DROP_DELAYED_REF;
2300                 goto again;
2301         }
2302         return NULL;
2303 }
2304
2305 /*
2306  * Returns 0 on success or if called with an already aborted transaction.
2307  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2308  */
2309 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2310                                        struct btrfs_root *root,
2311                                        struct list_head *cluster)
2312 {
2313         struct btrfs_delayed_ref_root *delayed_refs;
2314         struct btrfs_delayed_ref_node *ref;
2315         struct btrfs_delayed_ref_head *locked_ref = NULL;
2316         struct btrfs_delayed_extent_op *extent_op;
2317         struct btrfs_fs_info *fs_info = root->fs_info;
2318         int ret;
2319         int count = 0;
2320         int must_insert_reserved = 0;
2321
2322         delayed_refs = &trans->transaction->delayed_refs;
2323         while (1) {
2324                 if (!locked_ref) {
2325                         /* pick a new head ref from the cluster list */
2326                         if (list_empty(cluster))
2327                                 break;
2328
2329                         locked_ref = list_entry(cluster->next,
2330                                      struct btrfs_delayed_ref_head, cluster);
2331
2332                         /* grab the lock that says we are going to process
2333                          * all the refs for this head */
2334                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2335
2336                         /*
2337                          * we may have dropped the spin lock to get the head
2338                          * mutex lock, and that might have given someone else
2339                          * time to free the head.  If that's true, it has been
2340                          * removed from our list and we can move on.
2341                          */
2342                         if (ret == -EAGAIN) {
2343                                 locked_ref = NULL;
2344                                 count++;
2345                                 continue;
2346                         }
2347                 }
2348
2349                 /*
2350                  * We need to try and merge add/drops of the same ref since we
2351                  * can run into issues with relocate dropping the implicit ref
2352                  * and then it being added back again before the drop can
2353                  * finish.  If we merged anything we need to re-loop so we can
2354                  * get a good ref.
2355                  */
2356                 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2357                                          locked_ref);
2358
2359                 /*
2360                  * locked_ref is the head node, so we have to go one
2361                  * node back for any delayed ref updates
2362                  */
2363                 ref = select_delayed_ref(locked_ref);
2364
2365                 if (ref && ref->seq &&
2366                     btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2367                         /*
2368                          * there are still refs with lower seq numbers in the
2369                          * process of being added. Don't run this ref yet.
2370                          */
2371                         list_del_init(&locked_ref->cluster);
2372                         btrfs_delayed_ref_unlock(locked_ref);
2373                         locked_ref = NULL;
2374                         delayed_refs->num_heads_ready++;
2375                         spin_unlock(&delayed_refs->lock);
2376                         cond_resched();
2377                         spin_lock(&delayed_refs->lock);
2378                         continue;
2379                 }
2380
2381                 /*
2382                  * record the must insert reserved flag before we
2383                  * drop the spin lock.
2384                  */
2385                 must_insert_reserved = locked_ref->must_insert_reserved;
2386                 locked_ref->must_insert_reserved = 0;
2387
2388                 extent_op = locked_ref->extent_op;
2389                 locked_ref->extent_op = NULL;
2390
2391                 if (!ref) {
2392                         /* All delayed refs have been processed, Go ahead
2393                          * and send the head node to run_one_delayed_ref,
2394                          * so that any accounting fixes can happen
2395                          */
2396                         ref = &locked_ref->node;
2397
2398                         if (extent_op && must_insert_reserved) {
2399                                 btrfs_free_delayed_extent_op(extent_op);
2400                                 extent_op = NULL;
2401                         }
2402
2403                         if (extent_op) {
2404                                 spin_unlock(&delayed_refs->lock);
2405
2406                                 ret = run_delayed_extent_op(trans, root,
2407                                                             ref, extent_op);
2408                                 btrfs_free_delayed_extent_op(extent_op);
2409
2410                                 if (ret) {
2411                                         /*
2412                                          * Need to reset must_insert_reserved if
2413                                          * there was an error so the abort stuff
2414                                          * can cleanup the reserved space
2415                                          * properly.
2416                                          */
2417                                         if (must_insert_reserved)
2418                                                 locked_ref->must_insert_reserved = 1;
2419                                         btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2420                                         spin_lock(&delayed_refs->lock);
2421                                         btrfs_delayed_ref_unlock(locked_ref);
2422                                         return ret;
2423                                 }
2424
2425                                 goto next;
2426                         }
2427                 }
2428
2429                 ref->in_tree = 0;
2430                 rb_erase(&ref->rb_node, &delayed_refs->root);
2431                 delayed_refs->num_entries--;
2432                 if (!btrfs_delayed_ref_is_head(ref)) {
2433                         /*
2434                          * when we play the delayed ref, also correct the
2435                          * ref_mod on head
2436                          */
2437                         switch (ref->action) {
2438                         case BTRFS_ADD_DELAYED_REF:
2439                         case BTRFS_ADD_DELAYED_EXTENT:
2440                                 locked_ref->node.ref_mod -= ref->ref_mod;
2441                                 break;
2442                         case BTRFS_DROP_DELAYED_REF:
2443                                 locked_ref->node.ref_mod += ref->ref_mod;
2444                                 break;
2445                         default:
2446                                 WARN_ON(1);
2447                         }
2448                 } else {
2449                         list_del_init(&locked_ref->cluster);
2450                 }
2451                 spin_unlock(&delayed_refs->lock);
2452
2453                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2454                                           must_insert_reserved);
2455
2456                 btrfs_free_delayed_extent_op(extent_op);
2457                 if (ret) {
2458                         btrfs_delayed_ref_unlock(locked_ref);
2459                         btrfs_put_delayed_ref(ref);
2460                         btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2461                         spin_lock(&delayed_refs->lock);
2462                         return ret;
2463                 }
2464
2465                 /*
2466                  * If this node is a head, that means all the refs in this head
2467                  * have been dealt with, and we will pick the next head to deal
2468                  * with, so we must unlock the head and drop it from the cluster
2469                  * list before we release it.
2470                  */
2471                 if (btrfs_delayed_ref_is_head(ref)) {
2472                         btrfs_delayed_ref_unlock(locked_ref);
2473                         locked_ref = NULL;
2474                 }
2475                 btrfs_put_delayed_ref(ref);
2476                 count++;
2477 next:
2478                 cond_resched();
2479                 spin_lock(&delayed_refs->lock);
2480         }
2481         return count;
2482 }
2483
2484 #ifdef SCRAMBLE_DELAYED_REFS
2485 /*
2486  * Normally delayed refs get processed in ascending bytenr order. This
2487  * correlates in most cases to the order added. To expose dependencies on this
2488  * order, we start to process the tree in the middle instead of the beginning
2489  */
2490 static u64 find_middle(struct rb_root *root)
2491 {
2492         struct rb_node *n = root->rb_node;
2493         struct btrfs_delayed_ref_node *entry;
2494         int alt = 1;
2495         u64 middle;
2496         u64 first = 0, last = 0;
2497
2498         n = rb_first(root);
2499         if (n) {
2500                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2501                 first = entry->bytenr;
2502         }
2503         n = rb_last(root);
2504         if (n) {
2505                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2506                 last = entry->bytenr;
2507         }
2508         n = root->rb_node;
2509
2510         while (n) {
2511                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2512                 WARN_ON(!entry->in_tree);
2513
2514                 middle = entry->bytenr;
2515
2516                 if (alt)
2517                         n = n->rb_left;
2518                 else
2519                         n = n->rb_right;
2520
2521                 alt = 1 - alt;
2522         }
2523         return middle;
2524 }
2525 #endif
2526
2527 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2528                                          struct btrfs_fs_info *fs_info)
2529 {
2530         struct qgroup_update *qgroup_update;
2531         int ret = 0;
2532
2533         if (list_empty(&trans->qgroup_ref_list) !=
2534             !trans->delayed_ref_elem.seq) {
2535                 /* list without seq or seq without list */
2536                 btrfs_err(fs_info,
2537                         "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2538                         list_empty(&trans->qgroup_ref_list) ? "" : " not",
2539                         (u32)(trans->delayed_ref_elem.seq >> 32),
2540                         (u32)trans->delayed_ref_elem.seq);
2541                 BUG();
2542         }
2543
2544         if (!trans->delayed_ref_elem.seq)
2545                 return 0;
2546
2547         while (!list_empty(&trans->qgroup_ref_list)) {
2548                 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2549                                                  struct qgroup_update, list);
2550                 list_del(&qgroup_update->list);
2551                 if (!ret)
2552                         ret = btrfs_qgroup_account_ref(
2553                                         trans, fs_info, qgroup_update->node,
2554                                         qgroup_update->extent_op);
2555                 kfree(qgroup_update);
2556         }
2557
2558         btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2559
2560         return ret;
2561 }
2562
2563 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2564                       int count)
2565 {
2566         int val = atomic_read(&delayed_refs->ref_seq);
2567
2568         if (val < seq || val >= seq + count)
2569                 return 1;
2570         return 0;
2571 }
2572
2573 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2574 {
2575         u64 num_bytes;
2576
2577         num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2578                              sizeof(struct btrfs_extent_inline_ref));
2579         if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2580                 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2581
2582         /*
2583          * We don't ever fill up leaves all the way so multiply by 2 just to be
2584          * closer to what we're really going to want to ouse.
2585          */
2586         return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2587 }
2588
2589 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2590                                        struct btrfs_root *root)
2591 {
2592         struct btrfs_block_rsv *global_rsv;
2593         u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2594         u64 num_bytes;
2595         int ret = 0;
2596
2597         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2598         num_heads = heads_to_leaves(root, num_heads);
2599         if (num_heads > 1)
2600                 num_bytes += (num_heads - 1) * root->leafsize;
2601         num_bytes <<= 1;
2602         global_rsv = &root->fs_info->global_block_rsv;
2603
2604         /*
2605          * If we can't allocate any more chunks lets make sure we have _lots_ of
2606          * wiggle room since running delayed refs can create more delayed refs.
2607          */
2608         if (global_rsv->space_info->full)
2609                 num_bytes <<= 1;
2610
2611         spin_lock(&global_rsv->lock);
2612         if (global_rsv->reserved <= num_bytes)
2613                 ret = 1;
2614         spin_unlock(&global_rsv->lock);
2615         return ret;
2616 }
2617
2618 /*
2619  * this starts processing the delayed reference count updates and
2620  * extent insertions we have queued up so far.  count can be
2621  * 0, which means to process everything in the tree at the start
2622  * of the run (but not newly added entries), or it can be some target
2623  * number you'd like to process.
2624  *
2625  * Returns 0 on success or if called with an aborted transaction
2626  * Returns <0 on error and aborts the transaction
2627  */
2628 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2629                            struct btrfs_root *root, unsigned long count)
2630 {
2631         struct rb_node *node;
2632         struct btrfs_delayed_ref_root *delayed_refs;
2633         struct btrfs_delayed_ref_node *ref;
2634         struct list_head cluster;
2635         int ret;
2636         u64 delayed_start;
2637         int run_all = count == (unsigned long)-1;
2638         int run_most = 0;
2639         int loops;
2640
2641         /* We'll clean this up in btrfs_cleanup_transaction */
2642         if (trans->aborted)
2643                 return 0;
2644
2645         if (root == root->fs_info->extent_root)
2646                 root = root->fs_info->tree_root;
2647
2648         btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2649
2650         delayed_refs = &trans->transaction->delayed_refs;
2651         INIT_LIST_HEAD(&cluster);
2652         if (count == 0) {
2653                 count = delayed_refs->num_entries * 2;
2654                 run_most = 1;
2655         }
2656
2657         if (!run_all && !run_most) {
2658                 int old;
2659                 int seq = atomic_read(&delayed_refs->ref_seq);
2660
2661 progress:
2662                 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2663                 if (old) {
2664                         DEFINE_WAIT(__wait);
2665                         if (delayed_refs->flushing ||
2666                             !btrfs_should_throttle_delayed_refs(trans, root))
2667                                 return 0;
2668
2669                         prepare_to_wait(&delayed_refs->wait, &__wait,
2670                                         TASK_UNINTERRUPTIBLE);
2671
2672                         old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2673                         if (old) {
2674                                 schedule();
2675                                 finish_wait(&delayed_refs->wait, &__wait);
2676
2677                                 if (!refs_newer(delayed_refs, seq, 256))
2678                                         goto progress;
2679                                 else
2680                                         return 0;
2681                         } else {
2682                                 finish_wait(&delayed_refs->wait, &__wait);
2683                                 goto again;
2684                         }
2685                 }
2686
2687         } else {
2688                 atomic_inc(&delayed_refs->procs_running_refs);
2689         }
2690
2691 again:
2692         loops = 0;
2693         spin_lock(&delayed_refs->lock);
2694
2695 #ifdef SCRAMBLE_DELAYED_REFS
2696         delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2697 #endif
2698
2699         while (1) {
2700                 if (!(run_all || run_most) &&
2701                     !btrfs_should_throttle_delayed_refs(trans, root))
2702                         break;
2703
2704                 /*
2705                  * go find something we can process in the rbtree.  We start at
2706                  * the beginning of the tree, and then build a cluster
2707                  * of refs to process starting at the first one we are able to
2708                  * lock
2709                  */
2710                 delayed_start = delayed_refs->run_delayed_start;
2711                 ret = btrfs_find_ref_cluster(trans, &cluster,
2712                                              delayed_refs->run_delayed_start);
2713                 if (ret)
2714                         break;
2715
2716                 ret = run_clustered_refs(trans, root, &cluster);
2717                 if (ret < 0) {
2718                         btrfs_release_ref_cluster(&cluster);
2719                         spin_unlock(&delayed_refs->lock);
2720                         btrfs_abort_transaction(trans, root, ret);
2721                         atomic_dec(&delayed_refs->procs_running_refs);
2722                         wake_up(&delayed_refs->wait);
2723                         return ret;
2724                 }
2725
2726                 atomic_add(ret, &delayed_refs->ref_seq);
2727
2728                 count -= min_t(unsigned long, ret, count);
2729
2730                 if (count == 0)
2731                         break;
2732
2733                 if (delayed_start >= delayed_refs->run_delayed_start) {
2734                         if (loops == 0) {
2735                                 /*
2736                                  * btrfs_find_ref_cluster looped. let's do one
2737                                  * more cycle. if we don't run any delayed ref
2738                                  * during that cycle (because we can't because
2739                                  * all of them are blocked), bail out.
2740                                  */
2741                                 loops = 1;
2742                         } else {
2743                                 /*
2744                                  * no runnable refs left, stop trying
2745                                  */
2746                                 BUG_ON(run_all);
2747                                 break;
2748                         }
2749                 }
2750                 if (ret) {
2751                         /* refs were run, let's reset staleness detection */
2752                         loops = 0;
2753                 }
2754         }
2755
2756         if (run_all) {
2757                 if (!list_empty(&trans->new_bgs)) {
2758                         spin_unlock(&delayed_refs->lock);
2759                         btrfs_create_pending_block_groups(trans, root);
2760                         spin_lock(&delayed_refs->lock);
2761                 }
2762
2763                 node = rb_first(&delayed_refs->root);
2764                 if (!node)
2765                         goto out;
2766                 count = (unsigned long)-1;
2767
2768                 while (node) {
2769                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2770                                        rb_node);
2771                         if (btrfs_delayed_ref_is_head(ref)) {
2772                                 struct btrfs_delayed_ref_head *head;
2773
2774                                 head = btrfs_delayed_node_to_head(ref);
2775                                 atomic_inc(&ref->refs);
2776
2777                                 spin_unlock(&delayed_refs->lock);
2778                                 /*
2779                                  * Mutex was contended, block until it's
2780                                  * released and try again
2781                                  */
2782                                 mutex_lock(&head->mutex);
2783                                 mutex_unlock(&head->mutex);
2784
2785                                 btrfs_put_delayed_ref(ref);
2786                                 cond_resched();
2787                                 goto again;
2788                         }
2789                         node = rb_next(node);
2790                 }
2791                 spin_unlock(&delayed_refs->lock);
2792                 schedule_timeout(1);
2793                 goto again;
2794         }
2795 out:
2796         atomic_dec(&delayed_refs->procs_running_refs);
2797         smp_mb();
2798         if (waitqueue_active(&delayed_refs->wait))
2799                 wake_up(&delayed_refs->wait);
2800
2801         spin_unlock(&delayed_refs->lock);
2802         assert_qgroups_uptodate(trans);
2803         return 0;
2804 }
2805
2806 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2807                                 struct btrfs_root *root,
2808                                 u64 bytenr, u64 num_bytes, u64 flags,
2809                                 int level, int is_data)
2810 {
2811         struct btrfs_delayed_extent_op *extent_op;
2812         int ret;
2813
2814         extent_op = btrfs_alloc_delayed_extent_op();
2815         if (!extent_op)
2816                 return -ENOMEM;
2817
2818         extent_op->flags_to_set = flags;
2819         extent_op->update_flags = 1;
2820         extent_op->update_key = 0;
2821         extent_op->is_data = is_data ? 1 : 0;
2822         extent_op->level = level;
2823
2824         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2825                                           num_bytes, extent_op);
2826         if (ret)
2827                 btrfs_free_delayed_extent_op(extent_op);
2828         return ret;
2829 }
2830
2831 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2832                                       struct btrfs_root *root,
2833                                       struct btrfs_path *path,
2834                                       u64 objectid, u64 offset, u64 bytenr)
2835 {
2836         struct btrfs_delayed_ref_head *head;
2837         struct btrfs_delayed_ref_node *ref;
2838         struct btrfs_delayed_data_ref *data_ref;
2839         struct btrfs_delayed_ref_root *delayed_refs;
2840         struct rb_node *node;
2841         int ret = 0;
2842
2843         ret = -ENOENT;
2844         delayed_refs = &trans->transaction->delayed_refs;
2845         spin_lock(&delayed_refs->lock);
2846         head = btrfs_find_delayed_ref_head(trans, bytenr);
2847         if (!head)
2848                 goto out;
2849
2850         if (!mutex_trylock(&head->mutex)) {
2851                 atomic_inc(&head->node.refs);
2852                 spin_unlock(&delayed_refs->lock);
2853
2854                 btrfs_release_path(path);
2855
2856                 /*
2857                  * Mutex was contended, block until it's released and let
2858                  * caller try again
2859                  */
2860                 mutex_lock(&head->mutex);
2861                 mutex_unlock(&head->mutex);
2862                 btrfs_put_delayed_ref(&head->node);
2863                 return -EAGAIN;
2864         }
2865
2866         node = rb_prev(&head->node.rb_node);
2867         if (!node)
2868                 goto out_unlock;
2869
2870         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2871
2872         if (ref->bytenr != bytenr)
2873                 goto out_unlock;
2874
2875         ret = 1;
2876         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2877                 goto out_unlock;
2878
2879         data_ref = btrfs_delayed_node_to_data_ref(ref);
2880
2881         node = rb_prev(node);
2882         if (node) {
2883                 int seq = ref->seq;
2884
2885                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2886                 if (ref->bytenr == bytenr && ref->seq == seq)
2887                         goto out_unlock;
2888         }
2889
2890         if (data_ref->root != root->root_key.objectid ||
2891             data_ref->objectid != objectid || data_ref->offset != offset)
2892                 goto out_unlock;
2893
2894         ret = 0;
2895 out_unlock:
2896         mutex_unlock(&head->mutex);
2897 out:
2898         spin_unlock(&delayed_refs->lock);
2899         return ret;
2900 }
2901
2902 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2903                                         struct btrfs_root *root,
2904                                         struct btrfs_path *path,
2905                                         u64 objectid, u64 offset, u64 bytenr)
2906 {
2907         struct btrfs_root *extent_root = root->fs_info->extent_root;
2908         struct extent_buffer *leaf;
2909         struct btrfs_extent_data_ref *ref;
2910         struct btrfs_extent_inline_ref *iref;
2911         struct btrfs_extent_item *ei;
2912         struct btrfs_key key;
2913         u32 item_size;
2914         int ret;
2915
2916         key.objectid = bytenr;
2917         key.offset = (u64)-1;
2918         key.type = BTRFS_EXTENT_ITEM_KEY;
2919
2920         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2921         if (ret < 0)
2922                 goto out;
2923         BUG_ON(ret == 0); /* Corruption */
2924
2925         ret = -ENOENT;
2926         if (path->slots[0] == 0)
2927                 goto out;
2928
2929         path->slots[0]--;
2930         leaf = path->nodes[0];
2931         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2932
2933         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2934                 goto out;
2935
2936         ret = 1;
2937         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2938 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2939         if (item_size < sizeof(*ei)) {
2940                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2941                 goto out;
2942         }
2943 #endif
2944         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2945
2946         if (item_size != sizeof(*ei) +
2947             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2948                 goto out;
2949
2950         if (btrfs_extent_generation(leaf, ei) <=
2951             btrfs_root_last_snapshot(&root->root_item))
2952                 goto out;
2953
2954         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2955         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2956             BTRFS_EXTENT_DATA_REF_KEY)
2957                 goto out;
2958
2959         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2960         if (btrfs_extent_refs(leaf, ei) !=
2961             btrfs_extent_data_ref_count(leaf, ref) ||
2962             btrfs_extent_data_ref_root(leaf, ref) !=
2963             root->root_key.objectid ||
2964             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2965             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2966                 goto out;
2967
2968         ret = 0;
2969 out:
2970         return ret;
2971 }
2972
2973 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2974                           struct btrfs_root *root,
2975                           u64 objectid, u64 offset, u64 bytenr)
2976 {
2977         struct btrfs_path *path;
2978         int ret;
2979         int ret2;
2980
2981         path = btrfs_alloc_path();
2982         if (!path)
2983                 return -ENOENT;
2984
2985         do {
2986                 ret = check_committed_ref(trans, root, path, objectid,
2987                                           offset, bytenr);
2988                 if (ret && ret != -ENOENT)
2989                         goto out;
2990
2991                 ret2 = check_delayed_ref(trans, root, path, objectid,
2992                                          offset, bytenr);
2993         } while (ret2 == -EAGAIN);
2994
2995         if (ret2 && ret2 != -ENOENT) {
2996                 ret = ret2;
2997                 goto out;
2998         }
2999
3000         if (ret != -ENOENT || ret2 != -ENOENT)
3001                 ret = 0;
3002 out:
3003         btrfs_free_path(path);
3004         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3005                 WARN_ON(ret > 0);
3006         return ret;
3007 }
3008
3009 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3010                            struct btrfs_root *root,
3011                            struct extent_buffer *buf,
3012                            int full_backref, int inc, int for_cow)
3013 {
3014         u64 bytenr;
3015         u64 num_bytes;
3016         u64 parent;
3017         u64 ref_root;
3018         u32 nritems;
3019         struct btrfs_key key;
3020         struct btrfs_file_extent_item *fi;
3021         int i;
3022         int level;
3023         int ret = 0;
3024         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3025                             u64, u64, u64, u64, u64, u64, int);
3026
3027         ref_root = btrfs_header_owner(buf);
3028         nritems = btrfs_header_nritems(buf);
3029         level = btrfs_header_level(buf);
3030
3031         if (!root->ref_cows && level == 0)
3032                 return 0;
3033
3034         if (inc)
3035                 process_func = btrfs_inc_extent_ref;
3036         else
3037                 process_func = btrfs_free_extent;
3038
3039         if (full_backref)
3040                 parent = buf->start;
3041         else
3042                 parent = 0;
3043
3044         for (i = 0; i < nritems; i++) {
3045                 if (level == 0) {
3046                         btrfs_item_key_to_cpu(buf, &key, i);
3047                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3048                                 continue;
3049                         fi = btrfs_item_ptr(buf, i,
3050                                             struct btrfs_file_extent_item);
3051                         if (btrfs_file_extent_type(buf, fi) ==
3052                             BTRFS_FILE_EXTENT_INLINE)
3053                                 continue;
3054                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3055                         if (bytenr == 0)
3056                                 continue;
3057
3058                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3059                         key.offset -= btrfs_file_extent_offset(buf, fi);
3060                         ret = process_func(trans, root, bytenr, num_bytes,
3061                                            parent, ref_root, key.objectid,
3062                                            key.offset, for_cow);
3063                         if (ret)
3064                                 goto fail;
3065                 } else {
3066                         bytenr = btrfs_node_blockptr(buf, i);
3067                         num_bytes = btrfs_level_size(root, level - 1);
3068                         ret = process_func(trans, root, bytenr, num_bytes,
3069                                            parent, ref_root, level - 1, 0,
3070                                            for_cow);
3071                         if (ret)
3072                                 goto fail;
3073                 }
3074         }
3075         return 0;
3076 fail:
3077         return ret;
3078 }
3079
3080 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3081                   struct extent_buffer *buf, int full_backref, int for_cow)
3082 {
3083         return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3084 }
3085
3086 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3087                   struct extent_buffer *buf, int full_backref, int for_cow)
3088 {
3089         return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3090 }
3091
3092 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3093                                  struct btrfs_root *root,
3094                                  struct btrfs_path *path,
3095                                  struct btrfs_block_group_cache *cache)
3096 {
3097         int ret;
3098         struct btrfs_root *extent_root = root->fs_info->extent_root;
3099         unsigned long bi;
3100         struct extent_buffer *leaf;
3101
3102         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3103         if (ret < 0)
3104                 goto fail;
3105         BUG_ON(ret); /* Corruption */
3106
3107         leaf = path->nodes[0];
3108         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3109         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3110         btrfs_mark_buffer_dirty(leaf);
3111         btrfs_release_path(path);
3112 fail:
3113         if (ret) {
3114                 btrfs_abort_transaction(trans, root, ret);
3115                 return ret;
3116         }
3117         return 0;
3118
3119 }
3120
3121 static struct btrfs_block_group_cache *
3122 next_block_group(struct btrfs_root *root,
3123                  struct btrfs_block_group_cache *cache)
3124 {
3125         struct rb_node *node;
3126         spin_lock(&root->fs_info->block_group_cache_lock);
3127         node = rb_next(&cache->cache_node);
3128         btrfs_put_block_group(cache);
3129         if (node) {
3130                 cache = rb_entry(node, struct btrfs_block_group_cache,
3131                                  cache_node);
3132                 btrfs_get_block_group(cache);
3133         } else
3134                 cache = NULL;
3135         spin_unlock(&root->fs_info->block_group_cache_lock);
3136         return cache;
3137 }
3138
3139 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3140                             struct btrfs_trans_handle *trans,
3141                             struct btrfs_path *path)
3142 {
3143         struct btrfs_root *root = block_group->fs_info->tree_root;
3144         struct inode *inode = NULL;
3145         u64 alloc_hint = 0;
3146         int dcs = BTRFS_DC_ERROR;
3147         int num_pages = 0;
3148         int retries = 0;
3149         int ret = 0;
3150
3151         /*
3152          * If this block group is smaller than 100 megs don't bother caching the
3153          * block group.
3154          */
3155         if (block_group->key.offset < (100 * 1024 * 1024)) {
3156                 spin_lock(&block_group->lock);
3157                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3158                 spin_unlock(&block_group->lock);
3159                 return 0;
3160         }
3161
3162 again:
3163         inode = lookup_free_space_inode(root, block_group, path);
3164         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3165                 ret = PTR_ERR(inode);
3166                 btrfs_release_path(path);
3167                 goto out;
3168         }
3169
3170         if (IS_ERR(inode)) {
3171                 BUG_ON(retries);
3172                 retries++;
3173
3174                 if (block_group->ro)
3175                         goto out_free;
3176
3177                 ret = create_free_space_inode(root, trans, block_group, path);
3178                 if (ret)
3179                         goto out_free;
3180                 goto again;
3181         }
3182
3183         /* We've already setup this transaction, go ahead and exit */
3184         if (block_group->cache_generation == trans->transid &&
3185             i_size_read(inode)) {
3186                 dcs = BTRFS_DC_SETUP;
3187                 goto out_put;
3188         }
3189
3190         /*
3191          * We want to set the generation to 0, that way if anything goes wrong
3192          * from here on out we know not to trust this cache when we load up next
3193          * time.
3194          */
3195         BTRFS_I(inode)->generation = 0;
3196         ret = btrfs_update_inode(trans, root, inode);
3197         WARN_ON(ret);
3198
3199         if (i_size_read(inode) > 0) {
3200                 ret = btrfs_check_trunc_cache_free_space(root,
3201                                         &root->fs_info->global_block_rsv);
3202                 if (ret)
3203                         goto out_put;
3204
3205                 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3206                 if (ret)
3207                         goto out_put;
3208         }
3209
3210         spin_lock(&block_group->lock);
3211         if (block_group->cached != BTRFS_CACHE_FINISHED ||
3212             !btrfs_test_opt(root, SPACE_CACHE)) {
3213                 /*
3214                  * don't bother trying to write stuff out _if_
3215                  * a) we're not cached,
3216                  * b) we're with nospace_cache mount option.
3217                  */
3218                 dcs = BTRFS_DC_WRITTEN;
3219                 spin_unlock(&block_group->lock);
3220                 goto out_put;
3221         }
3222         spin_unlock(&block_group->lock);
3223
3224         /*
3225          * Try to preallocate enough space based on how big the block group is.
3226          * Keep in mind this has to include any pinned space which could end up
3227          * taking up quite a bit since it's not folded into the other space
3228          * cache.
3229          */
3230         num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3231         if (!num_pages)
3232                 num_pages = 1;
3233
3234         num_pages *= 16;
3235         num_pages *= PAGE_CACHE_SIZE;
3236
3237         ret = btrfs_check_data_free_space(inode, num_pages);
3238         if (ret)
3239                 goto out_put;
3240
3241         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3242                                               num_pages, num_pages,
3243                                               &alloc_hint);
3244         if (!ret)
3245                 dcs = BTRFS_DC_SETUP;
3246         btrfs_free_reserved_data_space(inode, num_pages);
3247
3248 out_put:
3249         iput(inode);
3250 out_free:
3251         btrfs_release_path(path);
3252 out:
3253         spin_lock(&block_group->lock);
3254         if (!ret && dcs == BTRFS_DC_SETUP)
3255                 block_group->cache_generation = trans->transid;
3256         block_group->disk_cache_state = dcs;
3257         spin_unlock(&block_group->lock);
3258
3259         return ret;
3260 }
3261
3262 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3263                                    struct btrfs_root *root)
3264 {
3265         struct btrfs_block_group_cache *cache;
3266         int err = 0;
3267         struct btrfs_path *path;
3268         u64 last = 0;
3269
3270         path = btrfs_alloc_path();
3271         if (!path)
3272                 return -ENOMEM;
3273
3274 again:
3275         while (1) {
3276                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3277                 while (cache) {
3278                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3279                                 break;
3280                         cache = next_block_group(root, cache);
3281                 }
3282                 if (!cache) {
3283                         if (last == 0)
3284                                 break;
3285                         last = 0;
3286                         continue;
3287                 }
3288                 err = cache_save_setup(cache, trans, path);
3289                 last = cache->key.objectid + cache->key.offset;
3290                 btrfs_put_block_group(cache);
3291         }
3292
3293         while (1) {
3294                 if (last == 0) {
3295                         err = btrfs_run_delayed_refs(trans, root,
3296                                                      (unsigned long)-1);
3297                         if (err) /* File system offline */
3298                                 goto out;
3299                 }
3300
3301                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3302                 while (cache) {
3303                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3304                                 btrfs_put_block_group(cache);
3305                                 goto again;
3306                         }
3307
3308                         if (cache->dirty)
3309                                 break;
3310                         cache = next_block_group(root, cache);
3311                 }
3312                 if (!cache) {
3313                         if (last == 0)
3314                                 break;
3315                         last = 0;
3316                         continue;
3317                 }
3318
3319                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3320                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3321                 cache->dirty = 0;
3322                 last = cache->key.objectid + cache->key.offset;
3323
3324                 err = write_one_cache_group(trans, root, path, cache);
3325                 btrfs_put_block_group(cache);
3326                 if (err) /* File system offline */
3327                         goto out;
3328         }
3329
3330         while (1) {
3331                 /*
3332                  * I don't think this is needed since we're just marking our
3333                  * preallocated extent as written, but just in case it can't
3334                  * hurt.
3335                  */
3336                 if (last == 0) {
3337                         err = btrfs_run_delayed_refs(trans, root,
3338                                                      (unsigned long)-1);
3339                         if (err) /* File system offline */
3340                                 goto out;
3341                 }
3342
3343                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3344                 while (cache) {
3345                         /*
3346                          * Really this shouldn't happen, but it could if we
3347                          * couldn't write the entire preallocated extent and
3348                          * splitting the extent resulted in a new block.
3349                          */
3350                         if (cache->dirty) {
3351                                 btrfs_put_block_group(cache);
3352                                 goto again;
3353                         }
3354                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3355                                 break;
3356                         cache = next_block_group(root, cache);
3357                 }
3358                 if (!cache) {
3359                         if (last == 0)
3360                                 break;
3361                         last = 0;
3362                         continue;
3363                 }
3364
3365                 err = btrfs_write_out_cache(root, trans, cache, path);
3366
3367                 /*
3368                  * If we didn't have an error then the cache state is still
3369                  * NEED_WRITE, so we can set it to WRITTEN.
3370                  */
3371                 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3372                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
3373                 last = cache->key.objectid + cache->key.offset;
3374                 btrfs_put_block_group(cache);
3375         }
3376 out:
3377
3378         btrfs_free_path(path);
3379         return err;
3380 }
3381
3382 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3383 {
3384         struct btrfs_block_group_cache *block_group;
3385         int readonly = 0;
3386
3387         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3388         if (!block_group || block_group->ro)
3389                 readonly = 1;
3390         if (block_group)
3391                 btrfs_put_block_group(block_group);
3392         return readonly;
3393 }
3394
3395 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3396                              u64 total_bytes, u64 bytes_used,
3397                              struct btrfs_space_info **space_info)
3398 {
3399         struct btrfs_space_info *found;
3400         int i;
3401         int factor;
3402         int ret;
3403
3404         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3405                      BTRFS_BLOCK_GROUP_RAID10))
3406                 factor = 2;
3407         else
3408                 factor = 1;
3409
3410         found = __find_space_info(info, flags);
3411         if (found) {
3412                 spin_lock(&found->lock);
3413                 found->total_bytes += total_bytes;
3414                 found->disk_total += total_bytes * factor;
3415                 found->bytes_used += bytes_used;
3416                 found->disk_used += bytes_used * factor;
3417                 found->full = 0;
3418                 spin_unlock(&found->lock);
3419                 *space_info = found;
3420                 return 0;
3421         }
3422         found = kzalloc(sizeof(*found), GFP_NOFS);
3423         if (!found)
3424                 return -ENOMEM;
3425
3426         ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3427         if (ret) {
3428                 kfree(found);
3429                 return ret;
3430         }
3431
3432         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3433                 INIT_LIST_HEAD(&found->block_groups[i]);
3434         init_rwsem(&found->groups_sem);
3435         spin_lock_init(&found->lock);
3436         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3437         found->total_bytes = total_bytes;
3438         found->disk_total = total_bytes * factor;
3439         found->bytes_used = bytes_used;
3440         found->disk_used = bytes_used * factor;
3441         found->bytes_pinned = 0;
3442         found->bytes_reserved = 0;
3443         found->bytes_readonly = 0;
3444         found->bytes_may_use = 0;
3445         found->full = 0;
3446         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3447         found->chunk_alloc = 0;
3448         found->flush = 0;
3449         init_waitqueue_head(&found->wait);
3450         *space_info = found;
3451         list_add_rcu(&found->list, &info->space_info);
3452         if (flags & BTRFS_BLOCK_GROUP_DATA)
3453                 info->data_sinfo = found;
3454         return 0;
3455 }
3456
3457 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3458 {
3459         u64 extra_flags = chunk_to_extended(flags) &
3460                                 BTRFS_EXTENDED_PROFILE_MASK;
3461
3462         write_seqlock(&fs_info->profiles_lock);
3463         if (flags & BTRFS_BLOCK_GROUP_DATA)
3464                 fs_info->avail_data_alloc_bits |= extra_flags;
3465         if (flags & BTRFS_BLOCK_GROUP_METADATA)
3466                 fs_info->avail_metadata_alloc_bits |= extra_flags;
3467         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3468                 fs_info->avail_system_alloc_bits |= extra_flags;
3469         write_sequnlock(&fs_info->profiles_lock);
3470 }
3471
3472 /*
3473  * returns target flags in extended format or 0 if restripe for this
3474  * chunk_type is not in progress
3475  *
3476  * should be called with either volume_mutex or balance_lock held
3477  */
3478 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3479 {
3480         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3481         u64 target = 0;
3482
3483         if (!bctl)
3484                 return 0;
3485
3486         if (flags & BTRFS_BLOCK_GROUP_DATA &&
3487             bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3488                 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3489         } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3490                    bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3491                 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3492         } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3493                    bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3494                 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3495         }
3496
3497         return target;
3498 }
3499
3500 /*
3501  * @flags: available profiles in extended format (see ctree.h)
3502  *
3503  * Returns reduced profile in chunk format.  If profile changing is in
3504  * progress (either running or paused) picks the target profile (if it's
3505  * already available), otherwise falls back to plain reducing.
3506  */
3507 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3508 {
3509         /*
3510          * we add in the count of missing devices because we want
3511          * to make sure that any RAID levels on a degraded FS
3512          * continue to be honored.
3513          */
3514         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3515                 root->fs_info->fs_devices->missing_devices;
3516         u64 target;
3517         u64 tmp;
3518
3519         /*
3520          * see if restripe for this chunk_type is in progress, if so
3521          * try to reduce to the target profile
3522          */
3523         spin_lock(&root->fs_info->balance_lock);
3524         target = get_restripe_target(root->fs_info, flags);
3525         if (target) {
3526                 /* pick target profile only if it's already available */
3527                 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3528                         spin_unlock(&root->fs_info->balance_lock);
3529                         return extended_to_chunk(target);
3530                 }
3531         }
3532         spin_unlock(&root->fs_info->balance_lock);
3533
3534         /* First, mask out the RAID levels which aren't possible */
3535         if (num_devices == 1)
3536                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3537                            BTRFS_BLOCK_GROUP_RAID5);
3538         if (num_devices < 3)
3539                 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3540         if (num_devices < 4)
3541                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3542
3543         tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3544                        BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3545                        BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3546         flags &= ~tmp;
3547
3548         if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3549                 tmp = BTRFS_BLOCK_GROUP_RAID6;
3550         else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3551                 tmp = BTRFS_BLOCK_GROUP_RAID5;
3552         else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3553                 tmp = BTRFS_BLOCK_GROUP_RAID10;
3554         else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3555                 tmp = BTRFS_BLOCK_GROUP_RAID1;
3556         else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3557                 tmp = BTRFS_BLOCK_GROUP_RAID0;
3558
3559         return extended_to_chunk(flags | tmp);
3560 }
3561
3562 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3563 {
3564         unsigned seq;
3565
3566         do {
3567                 seq = read_seqbegin(&root->fs_info->profiles_lock);
3568
3569                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3570                         flags |= root->fs_info->avail_data_alloc_bits;
3571                 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3572                         flags |= root->fs_info->avail_system_alloc_bits;
3573                 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3574                         flags |= root->fs_info->avail_metadata_alloc_bits;
3575         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3576
3577         return btrfs_reduce_alloc_profile(root, flags);
3578 }
3579
3580 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3581 {
3582         u64 flags;
3583         u64 ret;
3584
3585         if (data)
3586                 flags = BTRFS_BLOCK_GROUP_DATA;
3587         else if (root == root->fs_info->chunk_root)
3588                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3589         else
3590                 flags = BTRFS_BLOCK_GROUP_METADATA;
3591
3592         ret = get_alloc_profile(root, flags);
3593         return ret;
3594 }
3595
3596 /*
3597  * This will check the space that the inode allocates from to make sure we have
3598  * enough space for bytes.
3599  */
3600 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3601 {
3602         struct btrfs_space_info *data_sinfo;
3603         struct btrfs_root *root = BTRFS_I(inode)->root;
3604         struct btrfs_fs_info *fs_info = root->fs_info;
3605         u64 used;
3606         int ret = 0, committed = 0, alloc_chunk = 1;
3607
3608         /* make sure bytes are sectorsize aligned */
3609         bytes = ALIGN(bytes, root->sectorsize);
3610
3611         if (root == root->fs_info->tree_root ||
3612             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3613                 alloc_chunk = 0;
3614                 committed = 1;
3615         }
3616
3617         data_sinfo = fs_info->data_sinfo;
3618         if (!data_sinfo)
3619                 goto alloc;
3620
3621 again:
3622         /* make sure we have enough space to handle the data first */
3623         spin_lock(&data_sinfo->lock);
3624         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3625                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3626                 data_sinfo->bytes_may_use;
3627
3628         if (used + bytes > data_sinfo->total_bytes) {
3629                 struct btrfs_trans_handle *trans;
3630
3631                 /*
3632                  * if we don't have enough free bytes in this space then we need
3633                  * to alloc a new chunk.
3634                  */
3635                 if (!data_sinfo->full && alloc_chunk) {
3636                         u64 alloc_target;
3637
3638                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3639                         spin_unlock(&data_sinfo->lock);
3640 alloc:
3641                         alloc_target = btrfs_get_alloc_profile(root, 1);
3642                         trans = btrfs_join_transaction(root);
3643                         if (IS_ERR(trans))
3644                                 return PTR_ERR(trans);
3645
3646                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3647                                              alloc_target,
3648                                              CHUNK_ALLOC_NO_FORCE);
3649                         btrfs_end_transaction(trans, root);
3650                         if (ret < 0) {
3651                                 if (ret != -ENOSPC)
3652                                         return ret;
3653                                 else
3654                                         goto commit_trans;
3655                         }
3656
3657                         if (!data_sinfo)
3658                                 data_sinfo = fs_info->data_sinfo;
3659
3660                         goto again;
3661                 }
3662
3663                 /*
3664                  * If we don't have enough pinned space to deal with this
3665                  * allocation don't bother committing the transaction.
3666                  */
3667                 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3668                                            bytes) < 0)
3669                         committed = 1;
3670                 spin_unlock(&data_sinfo->lock);
3671
3672                 /* commit the current transaction and try again */
3673 commit_trans:
3674                 if (!committed &&
3675                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3676                         committed = 1;
3677
3678                         trans = btrfs_join_transaction(root);
3679                         if (IS_ERR(trans))
3680                                 return PTR_ERR(trans);
3681                         ret = btrfs_commit_transaction(trans, root);
3682                         if (ret)
3683                                 return ret;
3684                         goto again;
3685                 }
3686
3687                 return -ENOSPC;
3688         }
3689         data_sinfo->bytes_may_use += bytes;
3690         trace_btrfs_space_reservation(root->fs_info, "space_info",
3691                                       data_sinfo->flags, bytes, 1);
3692         spin_unlock(&data_sinfo->lock);
3693
3694         return 0;
3695 }
3696
3697 /*
3698  * Called if we need to clear a data reservation for this inode.
3699  */
3700 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3701 {
3702         struct btrfs_root *root = BTRFS_I(inode)->root;
3703         struct btrfs_space_info *data_sinfo;
3704
3705         /* make sure bytes are sectorsize aligned */
3706         bytes = ALIGN(bytes, root->sectorsize);
3707
3708         data_sinfo = root->fs_info->data_sinfo;
3709         spin_lock(&data_sinfo->lock);
3710         WARN_ON(data_sinfo->bytes_may_use < bytes);
3711         data_sinfo->bytes_may_use -= bytes;
3712         trace_btrfs_space_reservation(root->fs_info, "space_info",
3713                                       data_sinfo->flags, bytes, 0);
3714         spin_unlock(&data_sinfo->lock);
3715 }
3716
3717 static void force_metadata_allocation(struct btrfs_fs_info *info)
3718 {
3719         struct list_head *head = &info->space_info;
3720         struct btrfs_space_info *found;
3721
3722         rcu_read_lock();
3723         list_for_each_entry_rcu(found, head, list) {
3724                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3725                         found->force_alloc = CHUNK_ALLOC_FORCE;
3726         }
3727         rcu_read_unlock();
3728 }
3729
3730 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3731 {
3732         return (global->size << 1);
3733 }
3734
3735 static int should_alloc_chunk(struct btrfs_root *root,
3736                               struct btrfs_space_info *sinfo, int force)
3737 {
3738         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3739         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3740         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3741         u64 thresh;
3742
3743         if (force == CHUNK_ALLOC_FORCE)
3744                 return 1;
3745
3746         /*
3747          * We need to take into account the global rsv because for all intents
3748          * and purposes it's used space.  Don't worry about locking the
3749          * global_rsv, it doesn't change except when the transaction commits.
3750          */
3751         if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3752                 num_allocated += calc_global_rsv_need_space(global_rsv);
3753
3754         /*
3755          * in limited mode, we want to have some free space up to
3756          * about 1% of the FS size.
3757          */
3758         if (force == CHUNK_ALLOC_LIMITED) {
3759                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3760                 thresh = max_t(u64, 64 * 1024 * 1024,
3761                                div_factor_fine(thresh, 1));
3762
3763                 if (num_bytes - num_allocated < thresh)
3764                         return 1;
3765         }
3766
3767         if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3768                 return 0;
3769         return 1;
3770 }
3771
3772 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3773 {
3774         u64 num_dev;
3775
3776         if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3777                     BTRFS_BLOCK_GROUP_RAID0 |
3778                     BTRFS_BLOCK_GROUP_RAID5 |
3779                     BTRFS_BLOCK_GROUP_RAID6))
3780                 num_dev = root->fs_info->fs_devices->rw_devices;
3781         else if (type & BTRFS_BLOCK_GROUP_RAID1)
3782                 num_dev = 2;
3783         else
3784                 num_dev = 1;    /* DUP or single */
3785
3786         /* metadata for updaing devices and chunk tree */
3787         return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3788 }
3789
3790 static void check_system_chunk(struct btrfs_trans_handle *trans,
3791                                struct btrfs_root *root, u64 type)
3792 {
3793         struct btrfs_space_info *info;
3794         u64 left;
3795         u64 thresh;
3796
3797         info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3798         spin_lock(&info->lock);
3799         left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3800                 info->bytes_reserved - info->bytes_readonly;
3801         spin_unlock(&info->lock);
3802
3803         thresh = get_system_chunk_thresh(root, type);
3804         if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3805                 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3806                         left, thresh, type);
3807                 dump_space_info(info, 0, 0);
3808         }
3809
3810         if (left < thresh) {
3811                 u64 flags;
3812
3813                 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3814                 btrfs_alloc_chunk(trans, root, flags);
3815         }
3816 }
3817
3818 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3819                           struct btrfs_root *extent_root, u64 flags, int force)
3820 {
3821         struct btrfs_space_info *space_info;
3822         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3823         int wait_for_alloc = 0;
3824         int ret = 0;
3825
3826         /* Don't re-enter if we're already allocating a chunk */
3827         if (trans->allocating_chunk)
3828                 return -ENOSPC;
3829
3830         space_info = __find_space_info(extent_root->fs_info, flags);
3831         if (!space_info) {
3832                 ret = update_space_info(extent_root->fs_info, flags,
3833                                         0, 0, &space_info);
3834                 BUG_ON(ret); /* -ENOMEM */
3835         }
3836         BUG_ON(!space_info); /* Logic error */
3837
3838 again:
3839         spin_lock(&space_info->lock);
3840         if (force < space_info->force_alloc)
3841                 force = space_info->force_alloc;
3842         if (space_info->full) {
3843                 if (should_alloc_chunk(extent_root, space_info, force))
3844                         ret = -ENOSPC;
3845                 else
3846                         ret = 0;
3847                 spin_unlock(&space_info->lock);
3848                 return ret;
3849         }
3850
3851         if (!should_alloc_chunk(extent_root, space_info, force)) {
3852                 spin_unlock(&space_info->lock);
3853                 return 0;
3854         } else if (space_info->chunk_alloc) {
3855                 wait_for_alloc = 1;
3856         } else {
3857                 space_info->chunk_alloc = 1;
3858         }
3859
3860         spin_unlock(&space_info->lock);
3861
3862         mutex_lock(&fs_info->chunk_mutex);
3863
3864         /*
3865          * The chunk_mutex is held throughout the entirety of a chunk
3866          * allocation, so once we've acquired the chunk_mutex we know that the
3867          * other guy is done and we need to recheck and see if we should
3868          * allocate.
3869          */
3870         if (wait_for_alloc) {
3871                 mutex_unlock(&fs_info->chunk_mutex);
3872                 wait_for_alloc = 0;
3873                 goto again;
3874         }
3875
3876         trans->allocating_chunk = true;
3877
3878         /*
3879          * If we have mixed data/metadata chunks we want to make sure we keep
3880          * allocating mixed chunks instead of individual chunks.
3881          */
3882         if (btrfs_mixed_space_info(space_info))
3883                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3884
3885         /*
3886          * if we're doing a data chunk, go ahead and make sure that
3887          * we keep a reasonable number of metadata chunks allocated in the
3888          * FS as well.
3889          */
3890         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3891                 fs_info->data_chunk_allocations++;
3892                 if (!(fs_info->data_chunk_allocations %
3893                       fs_info->metadata_ratio))
3894                         force_metadata_allocation(fs_info);
3895         }
3896
3897         /*
3898          * Check if we have enough space in SYSTEM chunk because we may need
3899          * to update devices.
3900          */
3901         check_system_chunk(trans, extent_root, flags);
3902
3903         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3904         trans->allocating_chunk = false;
3905
3906         spin_lock(&space_info->lock);
3907         if (ret < 0 && ret != -ENOSPC)
3908                 goto out;
3909         if (ret)
3910                 space_info->full = 1;
3911         else
3912                 ret = 1;
3913
3914         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3915 out:
3916         space_info->chunk_alloc = 0;
3917         spin_unlock(&space_info->lock);
3918         mutex_unlock(&fs_info->chunk_mutex);
3919         return ret;
3920 }
3921
3922 static int can_overcommit(struct btrfs_root *root,
3923                           struct btrfs_space_info *space_info, u64 bytes,
3924                           enum btrfs_reserve_flush_enum flush)
3925 {
3926         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3927         u64 profile = btrfs_get_alloc_profile(root, 0);
3928         u64 space_size;
3929         u64 avail;
3930         u64 used;
3931
3932         used = space_info->bytes_used + space_info->bytes_reserved +
3933                 space_info->bytes_pinned + space_info->bytes_readonly;
3934
3935         /*
3936          * We only want to allow over committing if we have lots of actual space
3937          * free, but if we don't have enough space to handle the global reserve
3938          * space then we could end up having a real enospc problem when trying
3939          * to allocate a chunk or some other such important allocation.
3940          */
3941         spin_lock(&global_rsv->lock);
3942         space_size = calc_global_rsv_need_space(global_rsv);
3943         spin_unlock(&global_rsv->lock);
3944         if (used + space_size >= space_info->total_bytes)
3945                 return 0;
3946
3947         used += space_info->bytes_may_use;
3948
3949         spin_lock(&root->fs_info->free_chunk_lock);
3950         avail = root->fs_info->free_chunk_space;
3951         spin_unlock(&root->fs_info->free_chunk_lock);
3952
3953         /*
3954          * If we have dup, raid1 or raid10 then only half of the free
3955          * space is actually useable.  For raid56, the space info used
3956          * doesn't include the parity drive, so we don't have to
3957          * change the math
3958          */
3959         if (profile & (BTRFS_BLOCK_GROUP_DUP |
3960                        BTRFS_BLOCK_GROUP_RAID1 |
3961                        BTRFS_BLOCK_GROUP_RAID10))
3962                 avail >>= 1;
3963
3964         /*
3965          * If we aren't flushing all things, let us overcommit up to
3966          * 1/2th of the space. If we can flush, don't let us overcommit
3967          * too much, let it overcommit up to 1/8 of the space.
3968          */
3969         if (flush == BTRFS_RESERVE_FLUSH_ALL)
3970                 avail >>= 3;
3971         else
3972                 avail >>= 1;
3973
3974         if (used + bytes < space_info->total_bytes + avail)
3975                 return 1;
3976         return 0;
3977 }
3978
3979 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3980                                          unsigned long nr_pages)
3981 {
3982         struct super_block *sb = root->fs_info->sb;
3983
3984         if (down_read_trylock(&sb->s_umount)) {
3985                 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3986                 up_read(&sb->s_umount);
3987         } else {
3988                 /*
3989                  * We needn't worry the filesystem going from r/w to r/o though
3990                  * we don't acquire ->s_umount mutex, because the filesystem
3991                  * should guarantee the delalloc inodes list be empty after
3992                  * the filesystem is readonly(all dirty pages are written to
3993                  * the disk).
3994                  */
3995                 btrfs_start_all_delalloc_inodes(root->fs_info, 0);
3996                 if (!current->journal_info)
3997                         btrfs_wait_all_ordered_extents(root->fs_info);
3998         }
3999 }
4000
4001 /*
4002  * shrink metadata reservation for delalloc
4003  */
4004 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4005                             bool wait_ordered)
4006 {
4007         struct btrfs_block_rsv *block_rsv;
4008         struct btrfs_space_info *space_info;
4009         struct btrfs_trans_handle *trans;
4010         u64 delalloc_bytes;
4011         u64 max_reclaim;
4012         long time_left;
4013         unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
4014         int loops = 0;
4015         enum btrfs_reserve_flush_enum flush;
4016
4017         trans = (struct btrfs_trans_handle *)current->journal_info;
4018         block_rsv = &root->fs_info->delalloc_block_rsv;
4019         space_info = block_rsv->space_info;
4020
4021         smp_mb();
4022         delalloc_bytes = percpu_counter_sum_positive(
4023                                                 &root->fs_info->delalloc_bytes);
4024         if (delalloc_bytes == 0) {
4025                 if (trans)
4026                         return;
4027                 btrfs_wait_all_ordered_extents(root->fs_info);
4028                 return;
4029         }
4030
4031         while (delalloc_bytes && loops < 3) {
4032                 max_reclaim = min(delalloc_bytes, to_reclaim);
4033                 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4034                 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4035                 /*
4036                  * We need to wait for the async pages to actually start before
4037                  * we do anything.
4038                  */
4039                 wait_event(root->fs_info->async_submit_wait,
4040                            !atomic_read(&root->fs_info->async_delalloc_pages));
4041
4042                 if (!trans)
4043                         flush = BTRFS_RESERVE_FLUSH_ALL;
4044                 else
4045                         flush = BTRFS_RESERVE_NO_FLUSH;
4046                 spin_lock(&space_info->lock);
4047                 if (can_overcommit(root, space_info, orig, flush)) {
4048                         spin_unlock(&space_info->lock);
4049                         break;
4050                 }
4051                 spin_unlock(&space_info->lock);
4052
4053                 loops++;
4054                 if (wait_ordered && !trans) {
4055                         btrfs_wait_all_ordered_extents(root->fs_info);
4056                 } else {
4057                         time_left = schedule_timeout_killable(1);
4058                         if (time_left)
4059                                 break;
4060                 }
4061                 smp_mb();
4062                 delalloc_bytes = percpu_counter_sum_positive(
4063                                                 &root->fs_info->delalloc_bytes);
4064         }
4065 }
4066
4067 /**
4068  * maybe_commit_transaction - possibly commit the transaction if its ok to
4069  * @root - the root we're allocating for
4070  * @bytes - the number of bytes we want to reserve
4071  * @force - force the commit
4072  *
4073  * This will check to make sure that committing the transaction will actually
4074  * get us somewhere and then commit the transaction if it does.  Otherwise it
4075  * will return -ENOSPC.
4076  */
4077 static int may_commit_transaction(struct btrfs_root *root,
4078                                   struct btrfs_space_info *space_info,
4079                                   u64 bytes, int force)
4080 {
4081         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4082         struct btrfs_trans_handle *trans;
4083
4084         trans = (struct btrfs_trans_handle *)current->journal_info;
4085         if (trans)
4086                 return -EAGAIN;
4087
4088         if (force)
4089                 goto commit;
4090
4091         /* See if there is enough pinned space to make this reservation */
4092         spin_lock(&space_info->lock);
4093         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4094                                    bytes) >= 0) {
4095                 spin_unlock(&space_info->lock);
4096                 goto commit;
4097         }
4098         spin_unlock(&space_info->lock);
4099
4100         /*
4101          * See if there is some space in the delayed insertion reservation for
4102          * this reservation.
4103          */
4104         if (space_info != delayed_rsv->space_info)
4105                 return -ENOSPC;
4106
4107         spin_lock(&space_info->lock);
4108         spin_lock(&delayed_rsv->lock);
4109         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4110                                    bytes - delayed_rsv->size) >= 0) {
4111                 spin_unlock(&delayed_rsv->lock);
4112                 spin_unlock(&space_info->lock);
4113                 return -ENOSPC;
4114         }
4115         spin_unlock(&delayed_rsv->lock);
4116         spin_unlock(&space_info->lock);
4117
4118 commit:
4119         trans = btrfs_join_transaction(root);
4120         if (IS_ERR(trans))
4121                 return -ENOSPC;
4122
4123         return btrfs_commit_transaction(trans, root);
4124 }
4125
4126 enum flush_state {
4127         FLUSH_DELAYED_ITEMS_NR  =       1,
4128         FLUSH_DELAYED_ITEMS     =       2,
4129         FLUSH_DELALLOC          =       3,
4130         FLUSH_DELALLOC_WAIT     =       4,
4131         ALLOC_CHUNK             =       5,
4132         COMMIT_TRANS            =       6,
4133 };
4134
4135 static int flush_space(struct btrfs_root *root,
4136                        struct btrfs_space_info *space_info, u64 num_bytes,
4137                        u64 orig_bytes, int state)
4138 {
4139         struct btrfs_trans_handle *trans;
4140         int nr;
4141         int ret = 0;
4142
4143         switch (state) {
4144         case FLUSH_DELAYED_ITEMS_NR:
4145         case FLUSH_DELAYED_ITEMS:
4146                 if (state == FLUSH_DELAYED_ITEMS_NR) {
4147                         u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4148
4149                         nr = (int)div64_u64(num_bytes, bytes);
4150                         if (!nr)
4151                                 nr = 1;
4152                         nr *= 2;
4153                 } else {
4154                         nr = -1;
4155                 }
4156                 trans = btrfs_join_transaction(root);
4157                 if (IS_ERR(trans)) {
4158                         ret = PTR_ERR(trans);
4159                         break;
4160                 }
4161                 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4162                 btrfs_end_transaction(trans, root);
4163                 break;
4164         case FLUSH_DELALLOC:
4165         case FLUSH_DELALLOC_WAIT:
4166                 shrink_delalloc(root, num_bytes, orig_bytes,
4167                                 state == FLUSH_DELALLOC_WAIT);
4168                 break;
4169         case ALLOC_CHUNK:
4170                 trans = btrfs_join_transaction(root);
4171                 if (IS_ERR(trans)) {
4172                         ret = PTR_ERR(trans);
4173                         break;
4174                 }
4175                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4176                                      btrfs_get_alloc_profile(root, 0),
4177                                      CHUNK_ALLOC_NO_FORCE);
4178                 btrfs_end_transaction(trans, root);
4179                 if (ret == -ENOSPC)
4180                         ret = 0;
4181                 break;
4182         case COMMIT_TRANS:
4183                 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4184                 break;
4185         default:
4186                 ret = -ENOSPC;
4187                 break;
4188         }
4189
4190         return ret;
4191 }
4192 /**
4193  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4194  * @root - the root we're allocating for
4195  * @block_rsv - the block_rsv we're allocating for
4196  * @orig_bytes - the number of bytes we want
4197  * @flush - whether or not we can flush to make our reservation
4198  *
4199  * This will reserve orgi_bytes number of bytes from the space info associated
4200  * with the block_rsv.  If there is not enough space it will make an attempt to
4201  * flush out space to make room.  It will do this by flushing delalloc if
4202  * possible or committing the transaction.  If flush is 0 then no attempts to
4203  * regain reservations will be made and this will fail if there is not enough
4204  * space already.
4205  */
4206 static int reserve_metadata_bytes(struct btrfs_root *root,
4207                                   struct btrfs_block_rsv *block_rsv,
4208                                   u64 orig_bytes,
4209                                   enum btrfs_reserve_flush_enum flush)
4210 {
4211         struct btrfs_space_info *space_info = block_rsv->space_info;
4212         u64 used;
4213         u64 num_bytes = orig_bytes;
4214         int flush_state = FLUSH_DELAYED_ITEMS_NR;
4215         int ret = 0;
4216         bool flushing = false;
4217
4218 again:
4219         ret = 0;
4220         spin_lock(&space_info->lock);
4221         /*
4222          * We only want to wait if somebody other than us is flushing and we
4223          * are actually allowed to flush all things.
4224          */
4225         while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4226                space_info->flush) {
4227                 spin_unlock(&space_info->lock);
4228                 /*
4229                  * If we have a trans handle we can't wait because the flusher
4230                  * may have to commit the transaction, which would mean we would
4231                  * deadlock since we are waiting for the flusher to finish, but
4232                  * hold the current transaction open.
4233                  */
4234                 if (current->journal_info)
4235                         return -EAGAIN;
4236                 ret = wait_event_killable(space_info->wait, !space_info->flush);
4237                 /* Must have been killed, return */
4238                 if (ret)
4239                         return -EINTR;
4240
4241                 spin_lock(&space_info->lock);
4242         }
4243
4244         ret = -ENOSPC;
4245         used = space_info->bytes_used + space_info->bytes_reserved +
4246                 space_info->bytes_pinned + space_info->bytes_readonly +
4247                 space_info->bytes_may_use;
4248
4249         /*
4250          * The idea here is that we've not already over-reserved the block group
4251          * then we can go ahead and save our reservation first and then start
4252          * flushing if we need to.  Otherwise if we've already overcommitted
4253          * lets start flushing stuff first and then come back and try to make
4254          * our reservation.
4255          */
4256         if (used <= space_info->total_bytes) {
4257                 if (used + orig_bytes <= space_info->total_bytes) {
4258                         space_info->bytes_may_use += orig_bytes;
4259                         trace_btrfs_space_reservation(root->fs_info,
4260                                 "space_info", space_info->flags, orig_bytes, 1);
4261                         ret = 0;
4262                 } else {
4263                         /*
4264                          * Ok set num_bytes to orig_bytes since we aren't
4265                          * overocmmitted, this way we only try and reclaim what
4266                          * we need.
4267                          */
4268                         num_bytes = orig_bytes;
4269                 }
4270         } else {
4271                 /*
4272                  * Ok we're over committed, set num_bytes to the overcommitted
4273                  * amount plus the amount of bytes that we need for this
4274                  * reservation.
4275                  */
4276                 num_bytes = used - space_info->total_bytes +
4277                         (orig_bytes * 2);
4278         }
4279
4280         if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4281                 space_info->bytes_may_use += orig_bytes;
4282                 trace_btrfs_space_reservation(root->fs_info, "space_info",
4283                                               space_info->flags, orig_bytes,
4284                                               1);
4285                 ret = 0;
4286         }
4287
4288         /*
4289          * Couldn't make our reservation, save our place so while we're trying
4290          * to reclaim space we can actually use it instead of somebody else
4291          * stealing it from us.
4292          *
4293          * We make the other tasks wait for the flush only when we can flush
4294          * all things.
4295          */
4296         if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4297                 flushing = true;
4298                 space_info->flush = 1;
4299         }
4300
4301         spin_unlock(&space_info->lock);
4302
4303         if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4304                 goto out;
4305
4306         ret = flush_space(root, space_info, num_bytes, orig_bytes,
4307                           flush_state);
4308         flush_state++;
4309
4310         /*
4311          * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4312          * would happen. So skip delalloc flush.
4313          */
4314         if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4315             (flush_state == FLUSH_DELALLOC ||
4316              flush_state == FLUSH_DELALLOC_WAIT))
4317                 flush_state = ALLOC_CHUNK;
4318
4319         if (!ret)
4320                 goto again;
4321         else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4322                  flush_state < COMMIT_TRANS)
4323                 goto again;
4324         else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4325                  flush_state <= COMMIT_TRANS)
4326                 goto again;
4327
4328 out:
4329         if (ret == -ENOSPC &&
4330             unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4331                 struct btrfs_block_rsv *global_rsv =
4332                         &root->fs_info->global_block_rsv;
4333
4334                 if (block_rsv != global_rsv &&
4335                     !block_rsv_use_bytes(global_rsv, orig_bytes))
4336                         ret = 0;
4337         }
4338         if (flushing) {
4339                 spin_lock(&space_info->lock);
4340                 space_info->flush = 0;
4341                 wake_up_all(&space_info->wait);
4342                 spin_unlock(&space_info->lock);
4343         }
4344         return ret;
4345 }
4346
4347 static struct btrfs_block_rsv *get_block_rsv(
4348                                         const struct btrfs_trans_handle *trans,
4349                                         const struct btrfs_root *root)
4350 {
4351         struct btrfs_block_rsv *block_rsv = NULL;
4352
4353         if (root->ref_cows)
4354                 block_rsv = trans->block_rsv;
4355
4356         if (root == root->fs_info->csum_root && trans->adding_csums)
4357                 block_rsv = trans->block_rsv;
4358
4359         if (root == root->fs_info->uuid_root)
4360                 block_rsv = trans->block_rsv;
4361
4362         if (!block_rsv)
4363                 block_rsv = root->block_rsv;
4364
4365         if (!block_rsv)
4366                 block_rsv = &root->fs_info->empty_block_rsv;
4367
4368         return block_rsv;
4369 }
4370
4371 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4372                                u64 num_bytes)
4373 {
4374         int ret = -ENOSPC;
4375         spin_lock(&block_rsv->lock);
4376         if (block_rsv->reserved >= num_bytes) {
4377                 block_rsv->reserved -= num_bytes;
4378                 if (block_rsv->reserved < block_rsv->size)
4379                         block_rsv->full = 0;
4380                 ret = 0;
4381         }
4382         spin_unlock(&block_rsv->lock);
4383         return ret;
4384 }
4385
4386 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4387                                 u64 num_bytes, int update_size)
4388 {
4389         spin_lock(&block_rsv->lock);
4390         block_rsv->reserved += num_bytes;
4391         if (update_size)
4392                 block_rsv->size += num_bytes;
4393         else if (block_rsv->reserved >= block_rsv->size)
4394                 block_rsv->full = 1;
4395         spin_unlock(&block_rsv->lock);
4396 }
4397
4398 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4399                              struct btrfs_block_rsv *dest, u64 num_bytes,
4400                              int min_factor)
4401 {
4402         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4403         u64 min_bytes;
4404
4405         if (global_rsv->space_info != dest->space_info)
4406                 return -ENOSPC;
4407
4408         spin_lock(&global_rsv->lock);
4409         min_bytes = div_factor(global_rsv->size, min_factor);
4410         if (global_rsv->reserved < min_bytes + num_bytes) {
4411                 spin_unlock(&global_rsv->lock);
4412                 return -ENOSPC;
4413         }
4414         global_rsv->reserved -= num_bytes;
4415         if (global_rsv->reserved < global_rsv->size)
4416                 global_rsv->full = 0;
4417         spin_unlock(&global_rsv->lock);
4418
4419         block_rsv_add_bytes(dest, num_bytes, 1);
4420         return 0;
4421 }
4422
4423 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4424                                     struct btrfs_block_rsv *block_rsv,
4425                                     struct btrfs_block_rsv *dest, u64 num_bytes)
4426 {
4427         struct btrfs_space_info *space_info = block_rsv->space_info;
4428
4429         spin_lock(&block_rsv->lock);
4430         if (num_bytes == (u64)-1)
4431                 num_bytes = block_rsv->size;
4432         block_rsv->size -= num_bytes;
4433         if (block_rsv->reserved >= block_rsv->size) {
4434                 num_bytes = block_rsv->reserved - block_rsv->size;
4435                 block_rsv->reserved = block_rsv->size;
4436                 block_rsv->full = 1;
4437         } else {
4438                 num_bytes = 0;
4439         }
4440         spin_unlock(&block_rsv->lock);
4441
4442         if (num_bytes > 0) {
4443                 if (dest) {
4444                         spin_lock(&dest->lock);
4445                         if (!dest->full) {
4446                                 u64 bytes_to_add;
4447
4448                                 bytes_to_add = dest->size - dest->reserved;
4449                                 bytes_to_add = min(num_bytes, bytes_to_add);
4450                                 dest->reserved += bytes_to_add;
4451                                 if (dest->reserved >= dest->size)
4452                                         dest->full = 1;
4453                                 num_bytes -= bytes_to_add;
4454                         }
4455                         spin_unlock(&dest->lock);
4456                 }
4457                 if (num_bytes) {
4458                         spin_lock(&space_info->lock);
4459                         space_info->bytes_may_use -= num_bytes;
4460                         trace_btrfs_space_reservation(fs_info, "space_info",
4461                                         space_info->flags, num_bytes, 0);
4462                         spin_unlock(&space_info->lock);
4463                 }
4464         }
4465 }
4466
4467 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4468                                    struct btrfs_block_rsv *dst, u64 num_bytes)
4469 {
4470         int ret;
4471
4472         ret = block_rsv_use_bytes(src, num_bytes);
4473         if (ret)
4474                 return ret;
4475
4476         block_rsv_add_bytes(dst, num_bytes, 1);
4477         return 0;
4478 }
4479
4480 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4481 {
4482         memset(rsv, 0, sizeof(*rsv));
4483         spin_lock_init(&rsv->lock);
4484         rsv->type = type;
4485 }
4486
4487 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4488                                               unsigned short type)
4489 {
4490         struct btrfs_block_rsv *block_rsv;
4491         struct btrfs_fs_info *fs_info = root->fs_info;
4492
4493         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4494         if (!block_rsv)
4495                 return NULL;
4496
4497         btrfs_init_block_rsv(block_rsv, type);
4498         block_rsv->space_info = __find_space_info(fs_info,
4499                                                   BTRFS_BLOCK_GROUP_METADATA);
4500         return block_rsv;
4501 }
4502
4503 void btrfs_free_block_rsv(struct btrfs_root *root,
4504                           struct btrfs_block_rsv *rsv)
4505 {
4506         if (!rsv)
4507                 return;
4508         btrfs_block_rsv_release(root, rsv, (u64)-1);
4509         kfree(rsv);
4510 }
4511
4512 int btrfs_block_rsv_add(struct btrfs_root *root,
4513                         struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4514                         enum btrfs_reserve_flush_enum flush)
4515 {
4516         int ret;
4517
4518         if (num_bytes == 0)
4519                 return 0;
4520
4521         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4522         if (!ret) {
4523                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4524                 return 0;
4525         }
4526
4527         return ret;
4528 }
4529
4530 int btrfs_block_rsv_check(struct btrfs_root *root,
4531                           struct btrfs_block_rsv *block_rsv, int min_factor)
4532 {
4533         u64 num_bytes = 0;
4534         int ret = -ENOSPC;
4535
4536         if (!block_rsv)
4537                 return 0;
4538
4539         spin_lock(&block_rsv->lock);
4540         num_bytes = div_factor(block_rsv->size, min_factor);
4541         if (block_rsv->reserved >= num_bytes)
4542                 ret = 0;
4543         spin_unlock(&block_rsv->lock);
4544
4545         return ret;
4546 }
4547
4548 int btrfs_block_rsv_refill(struct btrfs_root *root,
4549                            struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4550                            enum btrfs_reserve_flush_enum flush)
4551 {
4552         u64 num_bytes = 0;
4553         int ret = -ENOSPC;
4554
4555         if (!block_rsv)
4556                 return 0;
4557
4558         spin_lock(&block_rsv->lock);
4559         num_bytes = min_reserved;
4560         if (block_rsv->reserved >= num_bytes)
4561                 ret = 0;
4562         else
4563                 num_bytes -= block_rsv->reserved;
4564         spin_unlock(&block_rsv->lock);
4565
4566         if (!ret)
4567                 return 0;
4568
4569         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4570         if (!ret) {
4571                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4572                 return 0;
4573         }
4574
4575         return ret;
4576 }
4577
4578 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4579                             struct btrfs_block_rsv *dst_rsv,
4580                             u64 num_bytes)
4581 {
4582         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4583 }
4584
4585 void btrfs_block_rsv_release(struct btrfs_root *root,
4586                              struct btrfs_block_rsv *block_rsv,
4587                              u64 num_bytes)
4588 {
4589         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4590         if (global_rsv->full || global_rsv == block_rsv ||
4591             block_rsv->space_info != global_rsv->space_info)
4592                 global_rsv = NULL;
4593         block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4594                                 num_bytes);
4595 }
4596
4597 /*
4598  * helper to calculate size of global block reservation.
4599  * the desired value is sum of space used by extent tree,
4600  * checksum tree and root tree
4601  */
4602 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4603 {
4604         struct btrfs_space_info *sinfo;
4605         u64 num_bytes;
4606         u64 meta_used;
4607         u64 data_used;
4608         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4609
4610         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4611         spin_lock(&sinfo->lock);
4612         data_used = sinfo->bytes_used;
4613         spin_unlock(&sinfo->lock);
4614
4615         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4616         spin_lock(&sinfo->lock);
4617         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4618                 data_used = 0;
4619         meta_used = sinfo->bytes_used;
4620         spin_unlock(&sinfo->lock);
4621
4622         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4623                     csum_size * 2;
4624         num_bytes += div64_u64(data_used + meta_used, 50);
4625
4626         if (num_bytes * 3 > meta_used)
4627                 num_bytes = div64_u64(meta_used, 3);
4628
4629         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4630 }
4631
4632 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4633 {
4634         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4635         struct btrfs_space_info *sinfo = block_rsv->space_info;
4636         u64 num_bytes;
4637
4638         num_bytes = calc_global_metadata_size(fs_info);
4639
4640         spin_lock(&sinfo->lock);
4641         spin_lock(&block_rsv->lock);
4642
4643         block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4644
4645         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4646                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4647                     sinfo->bytes_may_use;
4648
4649         if (sinfo->total_bytes > num_bytes) {
4650                 num_bytes = sinfo->total_bytes - num_bytes;
4651                 block_rsv->reserved += num_bytes;
4652                 sinfo->bytes_may_use += num_bytes;
4653                 trace_btrfs_space_reservation(fs_info, "space_info",
4654                                       sinfo->flags, num_bytes, 1);
4655         }
4656
4657         if (block_rsv->reserved >= block_rsv->size) {
4658                 num_bytes = block_rsv->reserved - block_rsv->size;
4659                 sinfo->bytes_may_use -= num_bytes;
4660                 trace_btrfs_space_reservation(fs_info, "space_info",
4661                                       sinfo->flags, num_bytes, 0);
4662                 block_rsv->reserved = block_rsv->size;
4663                 block_rsv->full = 1;
4664         }
4665
4666         spin_unlock(&block_rsv->lock);
4667         spin_unlock(&sinfo->lock);
4668 }
4669
4670 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4671 {
4672         struct btrfs_space_info *space_info;
4673
4674         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4675         fs_info->chunk_block_rsv.space_info = space_info;
4676
4677         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4678         fs_info->global_block_rsv.space_info = space_info;
4679         fs_info->delalloc_block_rsv.space_info = space_info;
4680         fs_info->trans_block_rsv.space_info = space_info;
4681         fs_info->empty_block_rsv.space_info = space_info;
4682         fs_info->delayed_block_rsv.space_info = space_info;
4683
4684         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4685         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4686         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4687         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4688         if (fs_info->quota_root)
4689                 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4690         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4691
4692         update_global_block_rsv(fs_info);
4693 }
4694
4695 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4696 {
4697         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4698                                 (u64)-1);
4699         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4700         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4701         WARN_ON(fs_info->trans_block_rsv.size > 0);
4702         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4703         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4704         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4705         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4706         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4707 }
4708
4709 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4710                                   struct btrfs_root *root)
4711 {
4712         if (!trans->block_rsv)
4713                 return;
4714
4715         if (!trans->bytes_reserved)
4716                 return;
4717
4718         trace_btrfs_space_reservation(root->fs_info, "transaction",
4719                                       trans->transid, trans->bytes_reserved, 0);
4720         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4721         trans->bytes_reserved = 0;
4722 }
4723
4724 /* Can only return 0 or -ENOSPC */
4725 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4726                                   struct inode *inode)
4727 {
4728         struct btrfs_root *root = BTRFS_I(inode)->root;
4729         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4730         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4731
4732         /*
4733          * We need to hold space in order to delete our orphan item once we've
4734          * added it, so this takes the reservation so we can release it later
4735          * when we are truly done with the orphan item.
4736          */
4737         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4738         trace_btrfs_space_reservation(root->fs_info, "orphan",
4739                                       btrfs_ino(inode), num_bytes, 1);
4740         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4741 }
4742
4743 void btrfs_orphan_release_metadata(struct inode *inode)
4744 {
4745         struct btrfs_root *root = BTRFS_I(inode)->root;
4746         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4747         trace_btrfs_space_reservation(root->fs_info, "orphan",
4748                                       btrfs_ino(inode), num_bytes, 0);
4749         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4750 }
4751
4752 /*
4753  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4754  * root: the root of the parent directory
4755  * rsv: block reservation
4756  * items: the number of items that we need do reservation
4757  * qgroup_reserved: used to return the reserved size in qgroup
4758  *
4759  * This function is used to reserve the space for snapshot/subvolume
4760  * creation and deletion. Those operations are different with the
4761  * common file/directory operations, they change two fs/file trees
4762  * and root tree, the number of items that the qgroup reserves is
4763  * different with the free space reservation. So we can not use
4764  * the space reseravtion mechanism in start_transaction().
4765  */
4766 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4767                                      struct btrfs_block_rsv *rsv,
4768                                      int items,
4769                                      u64 *qgroup_reserved,
4770                                      bool use_global_rsv)
4771 {
4772         u64 num_bytes;
4773         int ret;
4774         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4775
4776         if (root->fs_info->quota_enabled) {
4777                 /* One for parent inode, two for dir entries */
4778                 num_bytes = 3 * root->leafsize;
4779                 ret = btrfs_qgroup_reserve(root, num_bytes);
4780                 if (ret)
4781                         return ret;
4782         } else {
4783                 num_bytes = 0;
4784         }
4785
4786         *qgroup_reserved = num_bytes;
4787
4788         num_bytes = btrfs_calc_trans_metadata_size(root, items);
4789         rsv->space_info = __find_space_info(root->fs_info,
4790                                             BTRFS_BLOCK_GROUP_METADATA);
4791         ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4792                                   BTRFS_RESERVE_FLUSH_ALL);
4793
4794         if (ret == -ENOSPC && use_global_rsv)
4795                 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4796
4797         if (ret) {
4798                 if (*qgroup_reserved)
4799                         btrfs_qgroup_free(root, *qgroup_reserved);
4800         }
4801
4802         return ret;
4803 }
4804
4805 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4806                                       struct btrfs_block_rsv *rsv,
4807                                       u64 qgroup_reserved)
4808 {
4809         btrfs_block_rsv_release(root, rsv, (u64)-1);
4810         if (qgroup_reserved)
4811                 btrfs_qgroup_free(root, qgroup_reserved);
4812 }
4813
4814 /**
4815  * drop_outstanding_extent - drop an outstanding extent
4816  * @inode: the inode we're dropping the extent for
4817  *
4818  * This is called when we are freeing up an outstanding extent, either called
4819  * after an error or after an extent is written.  This will return the number of
4820  * reserved extents that need to be freed.  This must be called with
4821  * BTRFS_I(inode)->lock held.
4822  */
4823 static unsigned drop_outstanding_extent(struct inode *inode)
4824 {
4825         unsigned drop_inode_space = 0;
4826         unsigned dropped_extents = 0;
4827
4828         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4829         BTRFS_I(inode)->outstanding_extents--;
4830
4831         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4832             test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4833                                &BTRFS_I(inode)->runtime_flags))
4834                 drop_inode_space = 1;
4835
4836         /*
4837          * If we have more or the same amount of outsanding extents than we have
4838          * reserved then we need to leave the reserved extents count alone.
4839          */
4840         if (BTRFS_I(inode)->outstanding_extents >=
4841             BTRFS_I(inode)->reserved_extents)
4842                 return drop_inode_space;
4843
4844         dropped_extents = BTRFS_I(inode)->reserved_extents -
4845                 BTRFS_I(inode)->outstanding_extents;
4846         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4847         return dropped_extents + drop_inode_space;
4848 }
4849
4850 /**
4851  * calc_csum_metadata_size - return the amount of metada space that must be
4852  *      reserved/free'd for the given bytes.
4853  * @inode: the inode we're manipulating
4854  * @num_bytes: the number of bytes in question
4855  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4856  *
4857  * This adjusts the number of csum_bytes in the inode and then returns the
4858  * correct amount of metadata that must either be reserved or freed.  We
4859  * calculate how many checksums we can fit into one leaf and then divide the
4860  * number of bytes that will need to be checksumed by this value to figure out
4861  * how many checksums will be required.  If we are adding bytes then the number
4862  * may go up and we will return the number of additional bytes that must be
4863  * reserved.  If it is going down we will return the number of bytes that must
4864  * be freed.
4865  *
4866  * This must be called with BTRFS_I(inode)->lock held.
4867  */
4868 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4869                                    int reserve)
4870 {
4871         struct btrfs_root *root = BTRFS_I(inode)->root;
4872         u64 csum_size;
4873         int num_csums_per_leaf;
4874         int num_csums;
4875         int old_csums;
4876
4877         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4878             BTRFS_I(inode)->csum_bytes == 0)
4879                 return 0;
4880
4881         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4882         if (reserve)
4883                 BTRFS_I(inode)->csum_bytes += num_bytes;
4884         else
4885                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4886         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4887         num_csums_per_leaf = (int)div64_u64(csum_size,
4888                                             sizeof(struct btrfs_csum_item) +
4889                                             sizeof(struct btrfs_disk_key));
4890         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4891         num_csums = num_csums + num_csums_per_leaf - 1;
4892         num_csums = num_csums / num_csums_per_leaf;
4893
4894         old_csums = old_csums + num_csums_per_leaf - 1;
4895         old_csums = old_csums / num_csums_per_leaf;
4896
4897         /* No change, no need to reserve more */
4898         if (old_csums == num_csums)
4899                 return 0;
4900
4901         if (reserve)
4902                 return btrfs_calc_trans_metadata_size(root,
4903                                                       num_csums - old_csums);
4904
4905         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4906 }
4907
4908 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4909 {
4910         struct btrfs_root *root = BTRFS_I(inode)->root;
4911         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4912         u64 to_reserve = 0;
4913         u64 csum_bytes;
4914         unsigned nr_extents = 0;
4915         int extra_reserve = 0;
4916         enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4917         int ret = 0;
4918         bool delalloc_lock = true;
4919         u64 to_free = 0;
4920         unsigned dropped;
4921
4922         /* If we are a free space inode we need to not flush since we will be in
4923          * the middle of a transaction commit.  We also don't need the delalloc
4924          * mutex since we won't race with anybody.  We need this mostly to make
4925          * lockdep shut its filthy mouth.
4926          */
4927         if (btrfs_is_free_space_inode(inode)) {
4928                 flush = BTRFS_RESERVE_NO_FLUSH;
4929                 delalloc_lock = false;
4930         }
4931
4932         if (flush != BTRFS_RESERVE_NO_FLUSH &&
4933             btrfs_transaction_in_commit(root->fs_info))
4934                 schedule_timeout(1);
4935
4936         if (delalloc_lock)
4937                 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4938
4939         num_bytes = ALIGN(num_bytes, root->sectorsize);
4940
4941         spin_lock(&BTRFS_I(inode)->lock);
4942         BTRFS_I(inode)->outstanding_extents++;
4943
4944         if (BTRFS_I(inode)->outstanding_extents >
4945             BTRFS_I(inode)->reserved_extents)
4946                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4947                         BTRFS_I(inode)->reserved_extents;
4948
4949         /*
4950          * Add an item to reserve for updating the inode when we complete the
4951          * delalloc io.
4952          */
4953         if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4954                       &BTRFS_I(inode)->runtime_flags)) {
4955                 nr_extents++;
4956                 extra_reserve = 1;
4957         }
4958
4959         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4960         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4961         csum_bytes = BTRFS_I(inode)->csum_bytes;
4962         spin_unlock(&BTRFS_I(inode)->lock);
4963
4964         if (root->fs_info->quota_enabled) {
4965                 ret = btrfs_qgroup_reserve(root, num_bytes +
4966                                            nr_extents * root->leafsize);
4967                 if (ret)
4968                         goto out_fail;
4969         }
4970
4971         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4972         if (unlikely(ret)) {
4973                 if (root->fs_info->quota_enabled)
4974                         btrfs_qgroup_free(root, num_bytes +
4975                                                 nr_extents * root->leafsize);
4976                 goto out_fail;
4977         }
4978
4979         spin_lock(&BTRFS_I(inode)->lock);
4980         if (extra_reserve) {
4981                 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4982                         &BTRFS_I(inode)->runtime_flags);
4983                 nr_extents--;
4984         }
4985         BTRFS_I(inode)->reserved_extents += nr_extents;
4986         spin_unlock(&BTRFS_I(inode)->lock);
4987
4988         if (delalloc_lock)
4989                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4990
4991         if (to_reserve)
4992                 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4993                                               btrfs_ino(inode), to_reserve, 1);
4994         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4995
4996         return 0;
4997
4998 out_fail:
4999         spin_lock(&BTRFS_I(inode)->lock);
5000         dropped = drop_outstanding_extent(inode);
5001         /*
5002          * If the inodes csum_bytes is the same as the original
5003          * csum_bytes then we know we haven't raced with any free()ers
5004          * so we can just reduce our inodes csum bytes and carry on.
5005          */
5006         if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5007                 calc_csum_metadata_size(inode, num_bytes, 0);
5008         } else {
5009                 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5010                 u64 bytes;
5011
5012                 /*
5013                  * This is tricky, but first we need to figure out how much we
5014                  * free'd from any free-ers that occured during this
5015                  * reservation, so we reset ->csum_bytes to the csum_bytes
5016                  * before we dropped our lock, and then call the free for the
5017                  * number of bytes that were freed while we were trying our
5018                  * reservation.
5019                  */
5020                 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5021                 BTRFS_I(inode)->csum_bytes = csum_bytes;
5022                 to_free = calc_csum_metadata_size(inode, bytes, 0);
5023
5024
5025                 /*
5026                  * Now we need to see how much we would have freed had we not
5027                  * been making this reservation and our ->csum_bytes were not
5028                  * artificially inflated.
5029                  */
5030                 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5031                 bytes = csum_bytes - orig_csum_bytes;
5032                 bytes = calc_csum_metadata_size(inode, bytes, 0);
5033
5034                 /*
5035                  * Now reset ->csum_bytes to what it should be.  If bytes is
5036                  * more than to_free then we would have free'd more space had we
5037                  * not had an artificially high ->csum_bytes, so we need to free
5038                  * the remainder.  If bytes is the same or less then we don't
5039                  * need to do anything, the other free-ers did the correct
5040                  * thing.
5041                  */
5042                 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5043                 if (bytes > to_free)
5044                         to_free = bytes - to_free;
5045                 else
5046                         to_free = 0;
5047         }
5048         spin_unlock(&BTRFS_I(inode)->lock);
5049         if (dropped)
5050                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5051
5052         if (to_free) {
5053                 btrfs_block_rsv_release(root, block_rsv, to_free);
5054                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5055                                               btrfs_ino(inode), to_free, 0);
5056         }
5057         if (delalloc_lock)
5058                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5059         return ret;
5060 }
5061
5062 /**
5063  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5064  * @inode: the inode to release the reservation for
5065  * @num_bytes: the number of bytes we're releasing
5066  *
5067  * This will release the metadata reservation for an inode.  This can be called
5068  * once we complete IO for a given set of bytes to release their metadata
5069  * reservations.
5070  */
5071 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5072 {
5073         struct btrfs_root *root = BTRFS_I(inode)->root;
5074         u64 to_free = 0;
5075         unsigned dropped;
5076
5077         num_bytes = ALIGN(num_bytes, root->sectorsize);
5078         spin_lock(&BTRFS_I(inode)->lock);
5079         dropped = drop_outstanding_extent(inode);
5080
5081         if (num_bytes)
5082                 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5083         spin_unlock(&BTRFS_I(inode)->lock);
5084         if (dropped > 0)
5085                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5086
5087         trace_btrfs_space_reservation(root->fs_info, "delalloc",
5088                                       btrfs_ino(inode), to_free, 0);
5089         if (root->fs_info->quota_enabled) {
5090                 btrfs_qgroup_free(root, num_bytes +
5091                                         dropped * root->leafsize);
5092         }
5093
5094         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5095                                 to_free);
5096 }
5097
5098 /**
5099  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5100  * @inode: inode we're writing to
5101  * @num_bytes: the number of bytes we want to allocate
5102  *
5103  * This will do the following things
5104  *
5105  * o reserve space in the data space info for num_bytes
5106  * o reserve space in the metadata space info based on number of outstanding
5107  *   extents and how much csums will be needed
5108  * o add to the inodes ->delalloc_bytes
5109  * o add it to the fs_info's delalloc inodes list.
5110  *
5111  * This will return 0 for success and -ENOSPC if there is no space left.
5112  */
5113 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5114 {
5115         int ret;
5116
5117         ret = btrfs_check_data_free_space(inode, num_bytes);
5118         if (ret)
5119                 return ret;
5120
5121         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5122         if (ret) {
5123                 btrfs_free_reserved_data_space(inode, num_bytes);
5124                 return ret;
5125         }
5126
5127         return 0;
5128 }
5129
5130 /**
5131  * btrfs_delalloc_release_space - release data and metadata space for delalloc
5132  * @inode: inode we're releasing space for
5133  * @num_bytes: the number of bytes we want to free up
5134  *
5135  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
5136  * called in the case that we don't need the metadata AND data reservations
5137  * anymore.  So if there is an error or we insert an inline extent.
5138  *
5139  * This function will release the metadata space that was not used and will
5140  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5141  * list if there are no delalloc bytes left.
5142  */
5143 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5144 {
5145         btrfs_delalloc_release_metadata(inode, num_bytes);
5146         btrfs_free_reserved_data_space(inode, num_bytes);
5147 }
5148
5149 static int update_block_group(struct btrfs_root *root,
5150                               u64 bytenr, u64 num_bytes, int alloc)
5151 {
5152         struct btrfs_block_group_cache *cache = NULL;
5153         struct btrfs_fs_info *info = root->fs_info;
5154         u64 total = num_bytes;
5155         u64 old_val;
5156         u64 byte_in_group;
5157         int factor;
5158
5159         /* block accounting for super block */
5160         spin_lock(&info->delalloc_root_lock);
5161         old_val = btrfs_super_bytes_used(info->super_copy);
5162         if (alloc)
5163                 old_val += num_bytes;
5164         else
5165                 old_val -= num_bytes;
5166         btrfs_set_super_bytes_used(info->super_copy, old_val);
5167         spin_unlock(&info->delalloc_root_lock);
5168
5169         while (total) {
5170                 cache = btrfs_lookup_block_group(info, bytenr);
5171                 if (!cache)
5172                         return -ENOENT;
5173                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5174                                     BTRFS_BLOCK_GROUP_RAID1 |
5175                                     BTRFS_BLOCK_GROUP_RAID10))
5176                         factor = 2;
5177                 else
5178                         factor = 1;
5179                 /*
5180                  * If this block group has free space cache written out, we
5181                  * need to make sure to load it if we are removing space.  This
5182                  * is because we need the unpinning stage to actually add the
5183                  * space back to the block group, otherwise we will leak space.
5184                  */
5185                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5186                         cache_block_group(cache, 1);
5187
5188                 byte_in_group = bytenr - cache->key.objectid;
5189                 WARN_ON(byte_in_group > cache->key.offset);
5190
5191                 spin_lock(&cache->space_info->lock);
5192                 spin_lock(&cache->lock);
5193
5194                 if (btrfs_test_opt(root, SPACE_CACHE) &&
5195                     cache->disk_cache_state < BTRFS_DC_CLEAR)
5196                         cache->disk_cache_state = BTRFS_DC_CLEAR;
5197
5198                 cache->dirty = 1;
5199                 old_val = btrfs_block_group_used(&cache->item);
5200                 num_bytes = min(total, cache->key.offset - byte_in_group);
5201                 if (alloc) {
5202                         old_val += num_bytes;
5203                         btrfs_set_block_group_used(&cache->item, old_val);
5204                         cache->reserved -= num_bytes;
5205                         cache->space_info->bytes_reserved -= num_bytes;
5206                         cache->space_info->bytes_used += num_bytes;
5207                         cache->space_info->disk_used += num_bytes * factor;
5208                         spin_unlock(&cache->lock);
5209                         spin_unlock(&cache->space_info->lock);
5210                 } else {
5211                         old_val -= num_bytes;
5212                         btrfs_set_block_group_used(&cache->item, old_val);
5213                         cache->pinned += num_bytes;
5214                         cache->space_info->bytes_pinned += num_bytes;
5215                         cache->space_info->bytes_used -= num_bytes;
5216                         cache->space_info->disk_used -= num_bytes * factor;
5217                         spin_unlock(&cache->lock);
5218                         spin_unlock(&cache->space_info->lock);
5219
5220                         set_extent_dirty(info->pinned_extents,
5221                                          bytenr, bytenr + num_bytes - 1,
5222                                          GFP_NOFS | __GFP_NOFAIL);
5223                 }
5224                 btrfs_put_block_group(cache);
5225                 total -= num_bytes;
5226                 bytenr += num_bytes;
5227         }
5228         return 0;
5229 }
5230
5231 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5232 {
5233         struct btrfs_block_group_cache *cache;
5234         u64 bytenr;
5235
5236         spin_lock(&root->fs_info->block_group_cache_lock);
5237         bytenr = root->fs_info->first_logical_byte;
5238         spin_unlock(&root->fs_info->block_group_cache_lock);
5239
5240         if (bytenr < (u64)-1)
5241                 return bytenr;
5242
5243         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5244         if (!cache)
5245                 return 0;
5246
5247         bytenr = cache->key.objectid;
5248         btrfs_put_block_group(cache);
5249
5250         return bytenr;
5251 }
5252
5253 static int pin_down_extent(struct btrfs_root *root,
5254                            struct btrfs_block_group_cache *cache,
5255                            u64 bytenr, u64 num_bytes, int reserved)
5256 {
5257         spin_lock(&cache->space_info->lock);
5258         spin_lock(&cache->lock);
5259         cache->pinned += num_bytes;
5260         cache->space_info->bytes_pinned += num_bytes;
5261         if (reserved) {
5262                 cache->reserved -= num_bytes;
5263                 cache->space_info->bytes_reserved -= num_bytes;
5264         }
5265         spin_unlock(&cache->lock);
5266         spin_unlock(&cache->space_info->lock);
5267
5268         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5269                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5270         if (reserved)
5271                 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5272         return 0;
5273 }
5274
5275 /*
5276  * this function must be called within transaction
5277  */
5278 int btrfs_pin_extent(struct btrfs_root *root,
5279                      u64 bytenr, u64 num_bytes, int reserved)
5280 {
5281         struct btrfs_block_group_cache *cache;
5282
5283         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5284         BUG_ON(!cache); /* Logic error */
5285
5286         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5287
5288         btrfs_put_block_group(cache);
5289         return 0;
5290 }
5291
5292 /*
5293  * this function must be called within transaction
5294  */
5295 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5296                                     u64 bytenr, u64 num_bytes)
5297 {
5298         struct btrfs_block_group_cache *cache;
5299         int ret;
5300
5301         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5302         if (!cache)
5303                 return -EINVAL;
5304
5305         /*
5306          * pull in the free space cache (if any) so that our pin
5307          * removes the free space from the cache.  We have load_only set
5308          * to one because the slow code to read in the free extents does check
5309          * the pinned extents.
5310          */
5311         cache_block_group(cache, 1);
5312
5313         pin_down_extent(root, cache, bytenr, num_bytes, 0);
5314
5315         /* remove us from the free space cache (if we're there at all) */
5316         ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5317         btrfs_put_block_group(cache);
5318         return ret;
5319 }
5320
5321 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5322 {
5323         int ret;
5324         struct btrfs_block_group_cache *block_group;
5325         struct btrfs_caching_control *caching_ctl;
5326
5327         block_group = btrfs_lookup_block_group(root->fs_info, start);
5328         if (!block_group)
5329                 return -EINVAL;
5330
5331         cache_block_group(block_group, 0);
5332         caching_ctl = get_caching_control(block_group);
5333
5334         if (!caching_ctl) {
5335                 /* Logic error */
5336                 BUG_ON(!block_group_cache_done(block_group));
5337                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5338         } else {
5339                 mutex_lock(&caching_ctl->mutex);
5340
5341                 if (start >= caching_ctl->progress) {
5342                         ret = add_excluded_extent(root, start, num_bytes);
5343                 } else if (start + num_bytes <= caching_ctl->progress) {
5344                         ret = btrfs_remove_free_space(block_group,
5345                                                       start, num_bytes);
5346                 } else {
5347                         num_bytes = caching_ctl->progress - start;
5348                         ret = btrfs_remove_free_space(block_group,
5349                                                       start, num_bytes);
5350                         if (ret)
5351                                 goto out_lock;
5352
5353                         num_bytes = (start + num_bytes) -
5354                                 caching_ctl->progress;
5355                         start = caching_ctl->progress;
5356                         ret = add_excluded_extent(root, start, num_bytes);
5357                 }
5358 out_lock:
5359                 mutex_unlock(&caching_ctl->mutex);
5360                 put_caching_control(caching_ctl);
5361         }
5362         btrfs_put_block_group(block_group);
5363         return ret;
5364 }
5365
5366 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5367                                  struct extent_buffer *eb)
5368 {
5369         struct btrfs_file_extent_item *item;
5370         struct btrfs_key key;
5371         int found_type;
5372         int i;
5373
5374         if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5375                 return 0;
5376
5377         for (i = 0; i < btrfs_header_nritems(eb); i++) {
5378                 btrfs_item_key_to_cpu(eb, &key, i);
5379                 if (key.type != BTRFS_EXTENT_DATA_KEY)
5380                         continue;
5381                 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5382                 found_type = btrfs_file_extent_type(eb, item);
5383                 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5384                         continue;
5385                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5386                         continue;
5387                 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5388                 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5389                 __exclude_logged_extent(log, key.objectid, key.offset);
5390         }
5391
5392         return 0;
5393 }
5394
5395 /**
5396  * btrfs_update_reserved_bytes - update the block_group and space info counters
5397  * @cache:      The cache we are manipulating
5398  * @num_bytes:  The number of bytes in question
5399  * @reserve:    One of the reservation enums
5400  *
5401  * This is called by the allocator when it reserves space, or by somebody who is
5402  * freeing space that was never actually used on disk.  For example if you
5403  * reserve some space for a new leaf in transaction A and before transaction A
5404  * commits you free that leaf, you call this with reserve set to 0 in order to
5405  * clear the reservation.
5406  *
5407  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5408  * ENOSPC accounting.  For data we handle the reservation through clearing the
5409  * delalloc bits in the io_tree.  We have to do this since we could end up
5410  * allocating less disk space for the amount of data we have reserved in the
5411  * case of compression.
5412  *
5413  * If this is a reservation and the block group has become read only we cannot
5414  * make the reservation and return -EAGAIN, otherwise this function always
5415  * succeeds.
5416  */
5417 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5418                                        u64 num_bytes, int reserve)
5419 {
5420         struct btrfs_space_info *space_info = cache->space_info;
5421         int ret = 0;
5422
5423         spin_lock(&space_info->lock);
5424         spin_lock(&cache->lock);
5425         if (reserve != RESERVE_FREE) {
5426                 if (cache->ro) {
5427                         ret = -EAGAIN;
5428                 } else {
5429                         cache->reserved += num_bytes;
5430                         space_info->bytes_reserved += num_bytes;
5431                         if (reserve == RESERVE_ALLOC) {
5432                                 trace_btrfs_space_reservation(cache->fs_info,
5433                                                 "space_info", space_info->flags,
5434                                                 num_bytes, 0);
5435                                 space_info->bytes_may_use -= num_bytes;
5436                         }
5437                 }
5438         } else {
5439                 if (cache->ro)
5440                         space_info->bytes_readonly += num_bytes;
5441                 cache->reserved -= num_bytes;
5442                 space_info->bytes_reserved -= num_bytes;
5443         }
5444         spin_unlock(&cache->lock);
5445         spin_unlock(&space_info->lock);
5446         return ret;
5447 }
5448
5449 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5450                                 struct btrfs_root *root)
5451 {
5452         struct btrfs_fs_info *fs_info = root->fs_info;
5453         struct btrfs_caching_control *next;
5454         struct btrfs_caching_control *caching_ctl;
5455         struct btrfs_block_group_cache *cache;
5456         struct btrfs_space_info *space_info;
5457
5458         down_write(&fs_info->extent_commit_sem);
5459
5460         list_for_each_entry_safe(caching_ctl, next,
5461                                  &fs_info->caching_block_groups, list) {
5462                 cache = caching_ctl->block_group;
5463                 if (block_group_cache_done(cache)) {
5464                         cache->last_byte_to_unpin = (u64)-1;
5465                         list_del_init(&caching_ctl->list);
5466                         put_caching_control(caching_ctl);
5467                 } else {
5468                         cache->last_byte_to_unpin = caching_ctl->progress;
5469                 }
5470         }
5471
5472         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5473                 fs_info->pinned_extents = &fs_info->freed_extents[1];
5474         else
5475                 fs_info->pinned_extents = &fs_info->freed_extents[0];
5476
5477         up_write(&fs_info->extent_commit_sem);
5478
5479         list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5480                 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5481
5482         update_global_block_rsv(fs_info);
5483 }
5484
5485 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5486 {
5487         struct btrfs_fs_info *fs_info = root->fs_info;
5488         struct btrfs_block_group_cache *cache = NULL;
5489         struct btrfs_space_info *space_info;
5490         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5491         u64 len;
5492         bool readonly;
5493
5494         while (start <= end) {
5495                 readonly = false;
5496                 if (!cache ||
5497                     start >= cache->key.objectid + cache->key.offset) {
5498                         if (cache)
5499                                 btrfs_put_block_group(cache);
5500                         cache = btrfs_lookup_block_group(fs_info, start);
5501                         BUG_ON(!cache); /* Logic error */
5502                 }
5503
5504                 len = cache->key.objectid + cache->key.offset - start;
5505                 len = min(len, end + 1 - start);
5506
5507                 if (start < cache->last_byte_to_unpin) {
5508                         len = min(len, cache->last_byte_to_unpin - start);
5509                         btrfs_add_free_space(cache, start, len);
5510                 }
5511
5512                 start += len;
5513                 space_info = cache->space_info;
5514
5515                 spin_lock(&space_info->lock);
5516                 spin_lock(&cache->lock);
5517                 cache->pinned -= len;
5518                 space_info->bytes_pinned -= len;
5519                 if (cache->ro) {
5520                         space_info->bytes_readonly += len;
5521                         readonly = true;
5522                 }
5523                 spin_unlock(&cache->lock);
5524                 if (!readonly && global_rsv->space_info == space_info) {
5525                         spin_lock(&global_rsv->lock);
5526                         if (!global_rsv->full) {
5527                                 len = min(len, global_rsv->size -
5528                                           global_rsv->reserved);
5529                                 global_rsv->reserved += len;
5530                                 space_info->bytes_may_use += len;
5531                                 if (global_rsv->reserved >= global_rsv->size)
5532                                         global_rsv->full = 1;
5533                         }
5534                         spin_unlock(&global_rsv->lock);
5535                 }
5536                 spin_unlock(&space_info->lock);
5537         }
5538
5539         if (cache)
5540                 btrfs_put_block_group(cache);
5541         return 0;
5542 }
5543
5544 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5545                                struct btrfs_root *root)
5546 {
5547         struct btrfs_fs_info *fs_info = root->fs_info;
5548         struct extent_io_tree *unpin;
5549         u64 start;
5550         u64 end;
5551         int ret;
5552
5553         if (trans->aborted)
5554                 return 0;
5555
5556         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5557                 unpin = &fs_info->freed_extents[1];
5558         else
5559                 unpin = &fs_info->freed_extents[0];
5560
5561         while (1) {
5562                 ret = find_first_extent_bit(unpin, 0, &start, &end,
5563                                             EXTENT_DIRTY, NULL);
5564                 if (ret)
5565                         break;
5566
5567                 if (btrfs_test_opt(root, DISCARD))
5568                         ret = btrfs_discard_extent(root, start,
5569                                                    end + 1 - start, NULL);
5570
5571                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5572                 unpin_extent_range(root, start, end);
5573                 cond_resched();
5574         }
5575
5576         return 0;
5577 }
5578
5579 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5580                              u64 owner, u64 root_objectid)
5581 {
5582         struct btrfs_space_info *space_info;
5583         u64 flags;
5584
5585         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5586                 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5587                         flags = BTRFS_BLOCK_GROUP_SYSTEM;
5588                 else
5589                         flags = BTRFS_BLOCK_GROUP_METADATA;
5590         } else {
5591                 flags = BTRFS_BLOCK_GROUP_DATA;
5592         }
5593
5594         space_info = __find_space_info(fs_info, flags);
5595         BUG_ON(!space_info); /* Logic bug */
5596         percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5597 }
5598
5599
5600 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5601                                 struct btrfs_root *root,
5602                                 u64 bytenr, u64 num_bytes, u64 parent,
5603                                 u64 root_objectid, u64 owner_objectid,
5604                                 u64 owner_offset, int refs_to_drop,
5605                                 struct btrfs_delayed_extent_op *extent_op)
5606 {
5607         struct btrfs_key key;
5608         struct btrfs_path *path;
5609         struct btrfs_fs_info *info = root->fs_info;
5610         struct btrfs_root *extent_root = info->extent_root;
5611         struct extent_buffer *leaf;
5612         struct btrfs_extent_item *ei;
5613         struct btrfs_extent_inline_ref *iref;
5614         int ret;
5615         int is_data;
5616         int extent_slot = 0;
5617         int found_extent = 0;
5618         int num_to_del = 1;
5619         u32 item_size;
5620         u64 refs;
5621         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5622                                                  SKINNY_METADATA);
5623
5624         path = btrfs_alloc_path();
5625         if (!path)
5626                 return -ENOMEM;
5627
5628         path->reada = 1;
5629         path->leave_spinning = 1;
5630
5631         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5632         BUG_ON(!is_data && refs_to_drop != 1);
5633
5634         if (is_data)
5635                 skinny_metadata = 0;
5636
5637         ret = lookup_extent_backref(trans, extent_root, path, &iref,
5638                                     bytenr, num_bytes, parent,
5639                                     root_objectid, owner_objectid,
5640                                     owner_offset);
5641         if (ret == 0) {
5642                 extent_slot = path->slots[0];
5643                 while (extent_slot >= 0) {
5644                         btrfs_item_key_to_cpu(path->nodes[0], &key,
5645                                               extent_slot);
5646                         if (key.objectid != bytenr)
5647                                 break;
5648                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5649                             key.offset == num_bytes) {
5650                                 found_extent = 1;
5651                                 break;
5652                         }
5653                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
5654                             key.offset == owner_objectid) {
5655                                 found_extent = 1;
5656                                 break;
5657                         }
5658                         if (path->slots[0] - extent_slot > 5)
5659                                 break;
5660                         extent_slot--;
5661                 }
5662 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5663                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5664                 if (found_extent && item_size < sizeof(*ei))
5665                         found_extent = 0;
5666 #endif
5667                 if (!found_extent) {
5668                         BUG_ON(iref);
5669                         ret = remove_extent_backref(trans, extent_root, path,
5670                                                     NULL, refs_to_drop,
5671                                                     is_data);
5672                         if (ret) {
5673                                 btrfs_abort_transaction(trans, extent_root, ret);
5674                                 goto out;
5675                         }
5676                         btrfs_release_path(path);
5677                         path->leave_spinning = 1;
5678
5679                         key.objectid = bytenr;
5680                         key.type = BTRFS_EXTENT_ITEM_KEY;
5681                         key.offset = num_bytes;
5682
5683                         if (!is_data && skinny_metadata) {
5684                                 key.type = BTRFS_METADATA_ITEM_KEY;
5685                                 key.offset = owner_objectid;
5686                         }
5687
5688                         ret = btrfs_search_slot(trans, extent_root,
5689                                                 &key, path, -1, 1);
5690                         if (ret > 0 && skinny_metadata && path->slots[0]) {
5691                                 /*
5692                                  * Couldn't find our skinny metadata item,
5693                                  * see if we have ye olde extent item.
5694                                  */
5695                                 path->slots[0]--;
5696                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
5697                                                       path->slots[0]);
5698                                 if (key.objectid == bytenr &&
5699                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
5700                                     key.offset == num_bytes)
5701                                         ret = 0;
5702                         }
5703
5704                         if (ret > 0 && skinny_metadata) {
5705                                 skinny_metadata = false;
5706                                 key.type = BTRFS_EXTENT_ITEM_KEY;
5707                                 key.offset = num_bytes;
5708                                 btrfs_release_path(path);
5709                                 ret = btrfs_search_slot(trans, extent_root,
5710                                                         &key, path, -1, 1);
5711                         }
5712
5713                         if (ret) {
5714                                 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5715                                         ret, bytenr);
5716                                 if (ret > 0)
5717                                         btrfs_print_leaf(extent_root,
5718                                                          path->nodes[0]);
5719                         }
5720                         if (ret < 0) {
5721                                 btrfs_abort_transaction(trans, extent_root, ret);
5722                                 goto out;
5723                         }
5724                         extent_slot = path->slots[0];
5725                 }
5726         } else if (ret == -ENOENT) {
5727                 btrfs_print_leaf(extent_root, path->nodes[0]);
5728                 WARN_ON(1);
5729                 btrfs_err(info,
5730                         "unable to find ref byte nr %llu parent %llu root %llu  owner %llu offset %llu",
5731                         bytenr, parent, root_objectid, owner_objectid,
5732                         owner_offset);
5733         } else {
5734                 btrfs_abort_transaction(trans, extent_root, ret);
5735                 goto out;
5736         }
5737
5738         leaf = path->nodes[0];
5739         item_size = btrfs_item_size_nr(leaf, extent_slot);
5740 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5741         if (item_size < sizeof(*ei)) {
5742                 BUG_ON(found_extent || extent_slot != path->slots[0]);
5743                 ret = convert_extent_item_v0(trans, extent_root, path,
5744                                              owner_objectid, 0);
5745                 if (ret < 0) {
5746                         btrfs_abort_transaction(trans, extent_root, ret);
5747                         goto out;
5748                 }
5749
5750                 btrfs_release_path(path);
5751                 path->leave_spinning = 1;
5752
5753                 key.objectid = bytenr;
5754                 key.type = BTRFS_EXTENT_ITEM_KEY;
5755                 key.offset = num_bytes;
5756
5757                 ret = btrfs_search_slot(trans, extent_root, &key, path,
5758                                         -1, 1);
5759                 if (ret) {
5760                         btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5761                                 ret, bytenr);
5762                         btrfs_print_leaf(extent_root, path->nodes[0]);
5763                 }
5764                 if (ret < 0) {
5765                         btrfs_abort_transaction(trans, extent_root, ret);
5766                         goto out;
5767                 }
5768
5769                 extent_slot = path->slots[0];
5770                 leaf = path->nodes[0];
5771                 item_size = btrfs_item_size_nr(leaf, extent_slot);
5772         }
5773 #endif
5774         BUG_ON(item_size < sizeof(*ei));
5775         ei = btrfs_item_ptr(leaf, extent_slot,
5776                             struct btrfs_extent_item);
5777         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5778             key.type == BTRFS_EXTENT_ITEM_KEY) {
5779                 struct btrfs_tree_block_info *bi;
5780                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5781                 bi = (struct btrfs_tree_block_info *)(ei + 1);
5782                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5783         }
5784
5785         refs = btrfs_extent_refs(leaf, ei);
5786         if (refs < refs_to_drop) {
5787                 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5788                           "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5789                 ret = -EINVAL;
5790                 btrfs_abort_transaction(trans, extent_root, ret);
5791                 goto out;
5792         }
5793         refs -= refs_to_drop;
5794
5795         if (refs > 0) {
5796                 if (extent_op)
5797                         __run_delayed_extent_op(extent_op, leaf, ei);
5798                 /*
5799                  * In the case of inline back ref, reference count will
5800                  * be updated by remove_extent_backref
5801                  */
5802                 if (iref) {
5803                         BUG_ON(!found_extent);
5804                 } else {
5805                         btrfs_set_extent_refs(leaf, ei, refs);
5806                         btrfs_mark_buffer_dirty(leaf);
5807                 }
5808                 if (found_extent) {
5809                         ret = remove_extent_backref(trans, extent_root, path,
5810                                                     iref, refs_to_drop,
5811                                                     is_data);
5812                         if (ret) {
5813                                 btrfs_abort_transaction(trans, extent_root, ret);
5814                                 goto out;
5815                         }
5816                 }
5817                 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5818                                  root_objectid);
5819         } else {
5820                 if (found_extent) {
5821                         BUG_ON(is_data && refs_to_drop !=
5822                                extent_data_ref_count(root, path, iref));
5823                         if (iref) {
5824                                 BUG_ON(path->slots[0] != extent_slot);
5825                         } else {
5826                                 BUG_ON(path->slots[0] != extent_slot + 1);
5827                                 path->slots[0] = extent_slot;
5828                                 num_to_del = 2;
5829                         }
5830                 }
5831
5832                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5833                                       num_to_del);
5834                 if (ret) {
5835                         btrfs_abort_transaction(trans, extent_root, ret);
5836                         goto out;
5837                 }
5838                 btrfs_release_path(path);
5839
5840                 if (is_data) {
5841                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5842                         if (ret) {
5843                                 btrfs_abort_transaction(trans, extent_root, ret);
5844                                 goto out;
5845                         }
5846                 }
5847
5848                 ret = update_block_group(root, bytenr, num_bytes, 0);
5849                 if (ret) {
5850                         btrfs_abort_transaction(trans, extent_root, ret);
5851                         goto out;
5852                 }
5853         }
5854 out:
5855         btrfs_free_path(path);
5856         return ret;
5857 }
5858
5859 /*
5860  * when we free an block, it is possible (and likely) that we free the last
5861  * delayed ref for that extent as well.  This searches the delayed ref tree for
5862  * a given extent, and if there are no other delayed refs to be processed, it
5863  * removes it from the tree.
5864  */
5865 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5866                                       struct btrfs_root *root, u64 bytenr)
5867 {
5868         struct btrfs_delayed_ref_head *head;
5869         struct btrfs_delayed_ref_root *delayed_refs;
5870         struct btrfs_delayed_ref_node *ref;
5871         struct rb_node *node;
5872         int ret = 0;
5873
5874         delayed_refs = &trans->transaction->delayed_refs;
5875         spin_lock(&delayed_refs->lock);
5876         head = btrfs_find_delayed_ref_head(trans, bytenr);
5877         if (!head)
5878                 goto out;
5879
5880         node = rb_prev(&head->node.rb_node);
5881         if (!node)
5882                 goto out;
5883
5884         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5885
5886         /* there are still entries for this ref, we can't drop it */
5887         if (ref->bytenr == bytenr)
5888                 goto out;
5889
5890         if (head->extent_op) {
5891                 if (!head->must_insert_reserved)
5892                         goto out;
5893                 btrfs_free_delayed_extent_op(head->extent_op);
5894                 head->extent_op = NULL;
5895         }
5896
5897         /*
5898          * waiting for the lock here would deadlock.  If someone else has it
5899          * locked they are already in the process of dropping it anyway
5900          */
5901         if (!mutex_trylock(&head->mutex))
5902                 goto out;
5903
5904         /*
5905          * at this point we have a head with no other entries.  Go
5906          * ahead and process it.
5907          */
5908         head->node.in_tree = 0;
5909         rb_erase(&head->node.rb_node, &delayed_refs->root);
5910
5911         delayed_refs->num_entries--;
5912
5913         /*
5914          * we don't take a ref on the node because we're removing it from the
5915          * tree, so we just steal the ref the tree was holding.
5916          */
5917         delayed_refs->num_heads--;
5918         if (list_empty(&head->cluster))
5919                 delayed_refs->num_heads_ready--;
5920
5921         list_del_init(&head->cluster);
5922         spin_unlock(&delayed_refs->lock);
5923
5924         BUG_ON(head->extent_op);
5925         if (head->must_insert_reserved)
5926                 ret = 1;
5927
5928         mutex_unlock(&head->mutex);
5929         btrfs_put_delayed_ref(&head->node);
5930         return ret;
5931 out:
5932         spin_unlock(&delayed_refs->lock);
5933         return 0;
5934 }
5935
5936 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5937                            struct btrfs_root *root,
5938                            struct extent_buffer *buf,
5939                            u64 parent, int last_ref)
5940 {
5941         struct btrfs_block_group_cache *cache = NULL;
5942         int pin = 1;
5943         int ret;
5944
5945         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5946                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5947                                         buf->start, buf->len,
5948                                         parent, root->root_key.objectid,
5949                                         btrfs_header_level(buf),
5950                                         BTRFS_DROP_DELAYED_REF, NULL, 0);
5951                 BUG_ON(ret); /* -ENOMEM */
5952         }
5953
5954         if (!last_ref)
5955                 return;
5956
5957         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5958
5959         if (btrfs_header_generation(buf) == trans->transid) {
5960                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5961                         ret = check_ref_cleanup(trans, root, buf->start);
5962                         if (!ret)
5963                                 goto out;
5964                 }
5965
5966                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5967                         pin_down_extent(root, cache, buf->start, buf->len, 1);
5968                         goto out;
5969                 }
5970
5971                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5972
5973                 btrfs_add_free_space(cache, buf->start, buf->len);
5974                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5975                 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
5976                 pin = 0;
5977         }
5978 out:
5979         if (pin)
5980                 add_pinned_bytes(root->fs_info, buf->len,
5981                                  btrfs_header_level(buf),
5982                                  root->root_key.objectid);
5983
5984         /*
5985          * Deleting the buffer, clear the corrupt flag since it doesn't matter
5986          * anymore.
5987          */
5988         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5989         btrfs_put_block_group(cache);
5990 }
5991
5992 /* Can return -ENOMEM */
5993 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5994                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5995                       u64 owner, u64 offset, int for_cow)
5996 {
5997         int ret;
5998         struct btrfs_fs_info *fs_info = root->fs_info;
5999
6000         add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6001
6002         /*
6003          * tree log blocks never actually go into the extent allocation
6004          * tree, just update pinning info and exit early.
6005          */
6006         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6007                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6008                 /* unlocks the pinned mutex */
6009                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6010                 ret = 0;
6011         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6012                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6013                                         num_bytes,
6014                                         parent, root_objectid, (int)owner,
6015                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6016         } else {
6017                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6018                                                 num_bytes,
6019                                                 parent, root_objectid, owner,
6020                                                 offset, BTRFS_DROP_DELAYED_REF,
6021                                                 NULL, for_cow);
6022         }
6023         return ret;
6024 }
6025
6026 static u64 stripe_align(struct btrfs_root *root,
6027                         struct btrfs_block_group_cache *cache,
6028                         u64 val, u64 num_bytes)
6029 {
6030         u64 ret = ALIGN(val, root->stripesize);
6031         return ret;
6032 }
6033
6034 /*
6035  * when we wait for progress in the block group caching, its because
6036  * our allocation attempt failed at least once.  So, we must sleep
6037  * and let some progress happen before we try again.
6038  *
6039  * This function will sleep at least once waiting for new free space to
6040  * show up, and then it will check the block group free space numbers
6041  * for our min num_bytes.  Another option is to have it go ahead
6042  * and look in the rbtree for a free extent of a given size, but this
6043  * is a good start.
6044  *
6045  * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6046  * any of the information in this block group.
6047  */
6048 static noinline void
6049 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6050                                 u64 num_bytes)
6051 {
6052         struct btrfs_caching_control *caching_ctl;
6053
6054         caching_ctl = get_caching_control(cache);
6055         if (!caching_ctl)
6056                 return;
6057
6058         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6059                    (cache->free_space_ctl->free_space >= num_bytes));
6060
6061         put_caching_control(caching_ctl);
6062 }
6063
6064 static noinline int
6065 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6066 {
6067         struct btrfs_caching_control *caching_ctl;
6068         int ret = 0;
6069
6070         caching_ctl = get_caching_control(cache);
6071         if (!caching_ctl)
6072                 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6073
6074         wait_event(caching_ctl->wait, block_group_cache_done(cache));
6075         if (cache->cached == BTRFS_CACHE_ERROR)
6076                 ret = -EIO;
6077         put_caching_control(caching_ctl);
6078         return ret;
6079 }
6080
6081 int __get_raid_index(u64 flags)
6082 {
6083         if (flags & BTRFS_BLOCK_GROUP_RAID10)
6084                 return BTRFS_RAID_RAID10;
6085         else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6086                 return BTRFS_RAID_RAID1;
6087         else if (flags & BTRFS_BLOCK_GROUP_DUP)
6088                 return BTRFS_RAID_DUP;
6089         else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6090                 return BTRFS_RAID_RAID0;
6091         else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6092                 return BTRFS_RAID_RAID5;
6093         else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6094                 return BTRFS_RAID_RAID6;
6095
6096         return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6097 }
6098
6099 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6100 {
6101         return __get_raid_index(cache->flags);
6102 }
6103
6104 enum btrfs_loop_type {
6105         LOOP_CACHING_NOWAIT = 0,
6106         LOOP_CACHING_WAIT = 1,
6107         LOOP_ALLOC_CHUNK = 2,
6108         LOOP_NO_EMPTY_SIZE = 3,
6109 };
6110
6111 /*
6112  * walks the btree of allocated extents and find a hole of a given size.
6113  * The key ins is changed to record the hole:
6114  * ins->objectid == start position
6115  * ins->flags = BTRFS_EXTENT_ITEM_KEY
6116  * ins->offset == the size of the hole.
6117  * Any available blocks before search_start are skipped.
6118  *
6119  * If there is no suitable free space, we will record the max size of
6120  * the free space extent currently.
6121  */
6122 static noinline int find_free_extent(struct btrfs_root *orig_root,
6123                                      u64 num_bytes, u64 empty_size,
6124                                      u64 hint_byte, struct btrfs_key *ins,
6125                                      u64 flags)
6126 {
6127         int ret = 0;
6128         struct btrfs_root *root = orig_root->fs_info->extent_root;
6129         struct btrfs_free_cluster *last_ptr = NULL;
6130         struct btrfs_block_group_cache *block_group = NULL;
6131         struct btrfs_block_group_cache *used_block_group;
6132         u64 search_start = 0;
6133         u64 max_extent_size = 0;
6134         int empty_cluster = 2 * 1024 * 1024;
6135         struct btrfs_space_info *space_info;
6136         int loop = 0;
6137         int index = __get_raid_index(flags);
6138         int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6139                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6140         bool found_uncached_bg = false;
6141         bool failed_cluster_refill = false;
6142         bool failed_alloc = false;
6143         bool use_cluster = true;
6144         bool have_caching_bg = false;
6145
6146         WARN_ON(num_bytes < root->sectorsize);
6147         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6148         ins->objectid = 0;
6149         ins->offset = 0;
6150
6151         trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6152
6153         space_info = __find_space_info(root->fs_info, flags);
6154         if (!space_info) {
6155                 btrfs_err(root->fs_info, "No space info for %llu", flags);
6156                 return -ENOSPC;
6157         }
6158
6159         /*
6160          * If the space info is for both data and metadata it means we have a
6161          * small filesystem and we can't use the clustering stuff.
6162          */
6163         if (btrfs_mixed_space_info(space_info))
6164                 use_cluster = false;
6165
6166         if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6167                 last_ptr = &root->fs_info->meta_alloc_cluster;
6168                 if (!btrfs_test_opt(root, SSD))
6169                         empty_cluster = 64 * 1024;
6170         }
6171
6172         if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6173             btrfs_test_opt(root, SSD)) {
6174                 last_ptr = &root->fs_info->data_alloc_cluster;
6175         }
6176
6177         if (last_ptr) {
6178                 spin_lock(&last_ptr->lock);
6179                 if (last_ptr->block_group)
6180                         hint_byte = last_ptr->window_start;
6181                 spin_unlock(&last_ptr->lock);
6182         }
6183
6184         search_start = max(search_start, first_logical_byte(root, 0));
6185         search_start = max(search_start, hint_byte);
6186
6187         if (!last_ptr)
6188                 empty_cluster = 0;
6189
6190         if (search_start == hint_byte) {
6191                 block_group = btrfs_lookup_block_group(root->fs_info,
6192                                                        search_start);
6193                 used_block_group = block_group;
6194                 /*
6195                  * we don't want to use the block group if it doesn't match our
6196                  * allocation bits, or if its not cached.
6197                  *
6198                  * However if we are re-searching with an ideal block group
6199                  * picked out then we don't care that the block group is cached.
6200                  */
6201                 if (block_group && block_group_bits(block_group, flags) &&
6202                     block_group->cached != BTRFS_CACHE_NO) {
6203                         down_read(&space_info->groups_sem);
6204                         if (list_empty(&block_group->list) ||
6205                             block_group->ro) {
6206                                 /*
6207                                  * someone is removing this block group,
6208                                  * we can't jump into the have_block_group
6209                                  * target because our list pointers are not
6210                                  * valid
6211                                  */
6212                                 btrfs_put_block_group(block_group);
6213                                 up_read(&space_info->groups_sem);
6214                         } else {
6215                                 index = get_block_group_index(block_group);
6216                                 goto have_block_group;
6217                         }
6218                 } else if (block_group) {
6219                         btrfs_put_block_group(block_group);
6220                 }
6221         }
6222 search:
6223         have_caching_bg = false;
6224         down_read(&space_info->groups_sem);
6225         list_for_each_entry(block_group, &space_info->block_groups[index],
6226                             list) {
6227                 u64 offset;
6228                 int cached;
6229
6230                 used_block_group = block_group;
6231                 btrfs_get_block_group(block_group);
6232                 search_start = block_group->key.objectid;
6233
6234                 /*
6235                  * this can happen if we end up cycling through all the
6236                  * raid types, but we want to make sure we only allocate
6237                  * for the proper type.
6238                  */
6239                 if (!block_group_bits(block_group, flags)) {
6240                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
6241                                 BTRFS_BLOCK_GROUP_RAID1 |
6242                                 BTRFS_BLOCK_GROUP_RAID5 |
6243                                 BTRFS_BLOCK_GROUP_RAID6 |
6244                                 BTRFS_BLOCK_GROUP_RAID10;
6245
6246                         /*
6247                          * if they asked for extra copies and this block group
6248                          * doesn't provide them, bail.  This does allow us to
6249                          * fill raid0 from raid1.
6250                          */
6251                         if ((flags & extra) && !(block_group->flags & extra))
6252                                 goto loop;
6253                 }
6254
6255 have_block_group:
6256                 cached = block_group_cache_done(block_group);
6257                 if (unlikely(!cached)) {
6258                         found_uncached_bg = true;
6259                         ret = cache_block_group(block_group, 0);
6260                         BUG_ON(ret < 0);
6261                         ret = 0;
6262                 }
6263
6264                 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6265                         goto loop;
6266                 if (unlikely(block_group->ro))
6267                         goto loop;
6268
6269                 /*
6270                  * Ok we want to try and use the cluster allocator, so
6271                  * lets look there
6272                  */
6273                 if (last_ptr) {
6274                         unsigned long aligned_cluster;
6275                         /*
6276                          * the refill lock keeps out other
6277                          * people trying to start a new cluster
6278                          */
6279                         spin_lock(&last_ptr->refill_lock);
6280                         used_block_group = last_ptr->block_group;
6281                         if (used_block_group != block_group &&
6282                             (!used_block_group ||
6283                              used_block_group->ro ||
6284                              !block_group_bits(used_block_group, flags))) {
6285                                 used_block_group = block_group;
6286                                 goto refill_cluster;
6287                         }
6288
6289                         if (used_block_group != block_group)
6290                                 btrfs_get_block_group(used_block_group);
6291
6292                         offset = btrfs_alloc_from_cluster(used_block_group,
6293                                                 last_ptr,
6294                                                 num_bytes,
6295                                                 used_block_group->key.objectid,
6296                                                 &max_extent_size);
6297                         if (offset) {
6298                                 /* we have a block, we're done */
6299                                 spin_unlock(&last_ptr->refill_lock);
6300                                 trace_btrfs_reserve_extent_cluster(root,
6301                                         block_group, search_start, num_bytes);
6302                                 goto checks;
6303                         }
6304
6305                         WARN_ON(last_ptr->block_group != used_block_group);
6306                         if (used_block_group != block_group) {
6307                                 btrfs_put_block_group(used_block_group);
6308                                 used_block_group = block_group;
6309                         }
6310 refill_cluster:
6311                         BUG_ON(used_block_group != block_group);
6312                         /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6313                          * set up a new clusters, so lets just skip it
6314                          * and let the allocator find whatever block
6315                          * it can find.  If we reach this point, we
6316                          * will have tried the cluster allocator
6317                          * plenty of times and not have found
6318                          * anything, so we are likely way too
6319                          * fragmented for the clustering stuff to find
6320                          * anything.
6321                          *
6322                          * However, if the cluster is taken from the
6323                          * current block group, release the cluster
6324                          * first, so that we stand a better chance of
6325                          * succeeding in the unclustered
6326                          * allocation.  */
6327                         if (loop >= LOOP_NO_EMPTY_SIZE &&
6328                             last_ptr->block_group != block_group) {
6329                                 spin_unlock(&last_ptr->refill_lock);
6330                                 goto unclustered_alloc;
6331                         }
6332
6333                         /*
6334                          * this cluster didn't work out, free it and
6335                          * start over
6336                          */
6337                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
6338
6339                         if (loop >= LOOP_NO_EMPTY_SIZE) {
6340                                 spin_unlock(&last_ptr->refill_lock);
6341                                 goto unclustered_alloc;
6342                         }
6343
6344                         aligned_cluster = max_t(unsigned long,
6345                                                 empty_cluster + empty_size,
6346                                               block_group->full_stripe_len);
6347
6348                         /* allocate a cluster in this block group */
6349                         ret = btrfs_find_space_cluster(root, block_group,
6350                                                        last_ptr, search_start,
6351                                                        num_bytes,
6352                                                        aligned_cluster);
6353                         if (ret == 0) {
6354                                 /*
6355                                  * now pull our allocation out of this
6356                                  * cluster
6357                                  */
6358                                 offset = btrfs_alloc_from_cluster(block_group,
6359                                                         last_ptr,
6360                                                         num_bytes,
6361                                                         search_start,
6362                                                         &max_extent_size);
6363                                 if (offset) {
6364                                         /* we found one, proceed */
6365                                         spin_unlock(&last_ptr->refill_lock);
6366                                         trace_btrfs_reserve_extent_cluster(root,
6367                                                 block_group, search_start,
6368                                                 num_bytes);
6369                                         goto checks;
6370                                 }
6371                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
6372                                    && !failed_cluster_refill) {
6373                                 spin_unlock(&last_ptr->refill_lock);
6374
6375                                 failed_cluster_refill = true;
6376                                 wait_block_group_cache_progress(block_group,
6377                                        num_bytes + empty_cluster + empty_size);
6378                                 goto have_block_group;
6379                         }
6380
6381                         /*
6382                          * at this point we either didn't find a cluster
6383                          * or we weren't able to allocate a block from our
6384                          * cluster.  Free the cluster we've been trying
6385                          * to use, and go to the next block group
6386                          */
6387                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
6388                         spin_unlock(&last_ptr->refill_lock);
6389                         goto loop;
6390                 }
6391
6392 unclustered_alloc:
6393                 spin_lock(&block_group->free_space_ctl->tree_lock);
6394                 if (cached &&
6395                     block_group->free_space_ctl->free_space <
6396                     num_bytes + empty_cluster + empty_size) {
6397                         if (block_group->free_space_ctl->free_space >
6398                             max_extent_size)
6399                                 max_extent_size =
6400                                         block_group->free_space_ctl->free_space;
6401                         spin_unlock(&block_group->free_space_ctl->tree_lock);
6402                         goto loop;
6403                 }
6404                 spin_unlock(&block_group->free_space_ctl->tree_lock);
6405
6406                 offset = btrfs_find_space_for_alloc(block_group, search_start,
6407                                                     num_bytes, empty_size,
6408                                                     &max_extent_size);
6409                 /*
6410                  * If we didn't find a chunk, and we haven't failed on this
6411                  * block group before, and this block group is in the middle of
6412                  * caching and we are ok with waiting, then go ahead and wait
6413                  * for progress to be made, and set failed_alloc to true.
6414                  *
6415                  * If failed_alloc is true then we've already waited on this
6416                  * block group once and should move on to the next block group.
6417                  */
6418                 if (!offset && !failed_alloc && !cached &&
6419                     loop > LOOP_CACHING_NOWAIT) {
6420                         wait_block_group_cache_progress(block_group,
6421                                                 num_bytes + empty_size);
6422                         failed_alloc = true;
6423                         goto have_block_group;
6424                 } else if (!offset) {
6425                         if (!cached)
6426                                 have_caching_bg = true;
6427                         goto loop;
6428                 }
6429 checks:
6430                 search_start = stripe_align(root, used_block_group,
6431                                             offset, num_bytes);
6432
6433                 /* move on to the next group */
6434                 if (search_start + num_bytes >
6435                     used_block_group->key.objectid + used_block_group->key.offset) {
6436                         btrfs_add_free_space(used_block_group, offset, num_bytes);
6437                         goto loop;
6438                 }
6439
6440                 if (offset < search_start)
6441                         btrfs_add_free_space(used_block_group, offset,
6442                                              search_start - offset);
6443                 BUG_ON(offset > search_start);
6444
6445                 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6446                                                   alloc_type);
6447                 if (ret == -EAGAIN) {
6448                         btrfs_add_free_space(used_block_group, offset, num_bytes);
6449                         goto loop;
6450                 }
6451
6452                 /* we are all good, lets return */
6453                 ins->objectid = search_start;
6454                 ins->offset = num_bytes;
6455
6456                 trace_btrfs_reserve_extent(orig_root, block_group,
6457                                            search_start, num_bytes);
6458                 if (used_block_group != block_group)
6459                         btrfs_put_block_group(used_block_group);
6460                 btrfs_put_block_group(block_group);
6461                 break;
6462 loop:
6463                 failed_cluster_refill = false;
6464                 failed_alloc = false;
6465                 BUG_ON(index != get_block_group_index(block_group));
6466                 if (used_block_group != block_group)
6467                         btrfs_put_block_group(used_block_group);
6468                 btrfs_put_block_group(block_group);
6469         }
6470         up_read(&space_info->groups_sem);
6471
6472         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6473                 goto search;
6474
6475         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6476                 goto search;
6477
6478         /*
6479          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6480          *                      caching kthreads as we move along
6481          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6482          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6483          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6484          *                      again
6485          */
6486         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6487                 index = 0;
6488                 loop++;
6489                 if (loop == LOOP_ALLOC_CHUNK) {
6490                         struct btrfs_trans_handle *trans;
6491
6492                         trans = btrfs_join_transaction(root);
6493                         if (IS_ERR(trans)) {
6494                                 ret = PTR_ERR(trans);
6495                                 goto out;
6496                         }
6497
6498                         ret = do_chunk_alloc(trans, root, flags,
6499                                              CHUNK_ALLOC_FORCE);
6500                         /*
6501                          * Do not bail out on ENOSPC since we
6502                          * can do more things.
6503                          */
6504                         if (ret < 0 && ret != -ENOSPC)
6505                                 btrfs_abort_transaction(trans,
6506                                                         root, ret);
6507                         else
6508                                 ret = 0;
6509                         btrfs_end_transaction(trans, root);
6510                         if (ret)
6511                                 goto out;
6512                 }
6513
6514                 if (loop == LOOP_NO_EMPTY_SIZE) {
6515                         empty_size = 0;
6516                         empty_cluster = 0;
6517                 }
6518
6519                 goto search;
6520         } else if (!ins->objectid) {
6521                 ret = -ENOSPC;
6522         } else if (ins->objectid) {
6523                 ret = 0;
6524         }
6525 out:
6526         if (ret == -ENOSPC)
6527                 ins->offset = max_extent_size;
6528         return ret;
6529 }
6530
6531 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6532                             int dump_block_groups)
6533 {
6534         struct btrfs_block_group_cache *cache;
6535         int index = 0;
6536
6537         spin_lock(&info->lock);
6538         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6539                info->flags,
6540                info->total_bytes - info->bytes_used - info->bytes_pinned -
6541                info->bytes_reserved - info->bytes_readonly,
6542                (info->full) ? "" : "not ");
6543         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6544                "reserved=%llu, may_use=%llu, readonly=%llu\n",
6545                info->total_bytes, info->bytes_used, info->bytes_pinned,
6546                info->bytes_reserved, info->bytes_may_use,
6547                info->bytes_readonly);
6548         spin_unlock(&info->lock);
6549
6550         if (!dump_block_groups)
6551                 return;
6552
6553         down_read(&info->groups_sem);
6554 again:
6555         list_for_each_entry(cache, &info->block_groups[index], list) {
6556                 spin_lock(&cache->lock);
6557                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6558                        cache->key.objectid, cache->key.offset,
6559                        btrfs_block_group_used(&cache->item), cache->pinned,
6560                        cache->reserved, cache->ro ? "[readonly]" : "");
6561                 btrfs_dump_free_space(cache, bytes);
6562                 spin_unlock(&cache->lock);
6563         }
6564         if (++index < BTRFS_NR_RAID_TYPES)
6565                 goto again;
6566         up_read(&info->groups_sem);
6567 }
6568
6569 int btrfs_reserve_extent(struct btrfs_root *root,
6570                          u64 num_bytes, u64 min_alloc_size,
6571                          u64 empty_size, u64 hint_byte,
6572                          struct btrfs_key *ins, int is_data)
6573 {
6574         bool final_tried = false;
6575         u64 flags;
6576         int ret;
6577
6578         flags = btrfs_get_alloc_profile(root, is_data);
6579 again:
6580         WARN_ON(num_bytes < root->sectorsize);
6581         ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6582                                flags);
6583
6584         if (ret == -ENOSPC) {
6585                 if (!final_tried && ins->offset) {
6586                         num_bytes = min(num_bytes >> 1, ins->offset);
6587                         num_bytes = round_down(num_bytes, root->sectorsize);
6588                         num_bytes = max(num_bytes, min_alloc_size);
6589                         if (num_bytes == min_alloc_size)
6590                                 final_tried = true;
6591                         goto again;
6592                 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6593                         struct btrfs_space_info *sinfo;
6594
6595                         sinfo = __find_space_info(root->fs_info, flags);
6596                         btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6597                                 flags, num_bytes);
6598                         if (sinfo)
6599                                 dump_space_info(sinfo, num_bytes, 1);
6600                 }
6601         }
6602
6603         return ret;
6604 }
6605
6606 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6607                                         u64 start, u64 len, int pin)
6608 {
6609         struct btrfs_block_group_cache *cache;
6610         int ret = 0;
6611
6612         cache = btrfs_lookup_block_group(root->fs_info, start);
6613         if (!cache) {
6614                 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6615                         start);
6616                 return -ENOSPC;
6617         }
6618
6619         if (btrfs_test_opt(root, DISCARD))
6620                 ret = btrfs_discard_extent(root, start, len, NULL);
6621
6622         if (pin)
6623                 pin_down_extent(root, cache, start, len, 1);
6624         else {
6625                 btrfs_add_free_space(cache, start, len);
6626                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6627         }
6628         btrfs_put_block_group(cache);
6629
6630         trace_btrfs_reserved_extent_free(root, start, len);
6631
6632         return ret;
6633 }
6634
6635 int btrfs_free_reserved_extent(struct btrfs_root *root,
6636                                         u64 start, u64 len)
6637 {
6638         return __btrfs_free_reserved_extent(root, start, len, 0);
6639 }
6640
6641 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6642                                        u64 start, u64 len)
6643 {
6644         return __btrfs_free_reserved_extent(root, start, len, 1);
6645 }
6646
6647 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6648                                       struct btrfs_root *root,
6649                                       u64 parent, u64 root_objectid,
6650                                       u64 flags, u64 owner, u64 offset,
6651                                       struct btrfs_key *ins, int ref_mod)
6652 {
6653         int ret;
6654         struct btrfs_fs_info *fs_info = root->fs_info;
6655         struct btrfs_extent_item *extent_item;
6656         struct btrfs_extent_inline_ref *iref;
6657         struct btrfs_path *path;
6658         struct extent_buffer *leaf;
6659         int type;
6660         u32 size;
6661
6662         if (parent > 0)
6663                 type = BTRFS_SHARED_DATA_REF_KEY;
6664         else
6665                 type = BTRFS_EXTENT_DATA_REF_KEY;
6666
6667         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6668
6669         path = btrfs_alloc_path();
6670         if (!path)
6671                 return -ENOMEM;
6672
6673         path->leave_spinning = 1;
6674         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6675                                       ins, size);
6676         if (ret) {
6677                 btrfs_free_path(path);
6678                 return ret;
6679         }
6680
6681         leaf = path->nodes[0];
6682         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6683                                      struct btrfs_extent_item);
6684         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6685         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6686         btrfs_set_extent_flags(leaf, extent_item,
6687                                flags | BTRFS_EXTENT_FLAG_DATA);
6688
6689         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6690         btrfs_set_extent_inline_ref_type(leaf, iref, type);
6691         if (parent > 0) {
6692                 struct btrfs_shared_data_ref *ref;
6693                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6694                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6695                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6696         } else {
6697                 struct btrfs_extent_data_ref *ref;
6698                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6699                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6700                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6701                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6702                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6703         }
6704
6705         btrfs_mark_buffer_dirty(path->nodes[0]);
6706         btrfs_free_path(path);
6707
6708         ret = update_block_group(root, ins->objectid, ins->offset, 1);
6709         if (ret) { /* -ENOENT, logic error */
6710                 btrfs_err(fs_info, "update block group failed for %llu %llu",
6711                         ins->objectid, ins->offset);
6712                 BUG();
6713         }
6714         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6715         return ret;
6716 }
6717
6718 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6719                                      struct btrfs_root *root,
6720                                      u64 parent, u64 root_objectid,
6721                                      u64 flags, struct btrfs_disk_key *key,
6722                                      int level, struct btrfs_key *ins)
6723 {
6724         int ret;
6725         struct btrfs_fs_info *fs_info = root->fs_info;
6726         struct btrfs_extent_item *extent_item;
6727         struct btrfs_tree_block_info *block_info;
6728         struct btrfs_extent_inline_ref *iref;
6729         struct btrfs_path *path;
6730         struct extent_buffer *leaf;
6731         u32 size = sizeof(*extent_item) + sizeof(*iref);
6732         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6733                                                  SKINNY_METADATA);
6734
6735         if (!skinny_metadata)
6736                 size += sizeof(*block_info);
6737
6738         path = btrfs_alloc_path();
6739         if (!path) {
6740                 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6741                                                    root->leafsize);
6742                 return -ENOMEM;
6743         }
6744
6745         path->leave_spinning = 1;
6746         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6747                                       ins, size);
6748         if (ret) {
6749                 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6750                                                    root->leafsize);
6751                 btrfs_free_path(path);
6752                 return ret;
6753         }
6754
6755         leaf = path->nodes[0];
6756         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6757                                      struct btrfs_extent_item);
6758         btrfs_set_extent_refs(leaf, extent_item, 1);
6759         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6760         btrfs_set_extent_flags(leaf, extent_item,
6761                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6762
6763         if (skinny_metadata) {
6764                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6765         } else {
6766                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6767                 btrfs_set_tree_block_key(leaf, block_info, key);
6768                 btrfs_set_tree_block_level(leaf, block_info, level);
6769                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6770         }
6771
6772         if (parent > 0) {
6773                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6774                 btrfs_set_extent_inline_ref_type(leaf, iref,
6775                                                  BTRFS_SHARED_BLOCK_REF_KEY);
6776                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6777         } else {
6778                 btrfs_set_extent_inline_ref_type(leaf, iref,
6779                                                  BTRFS_TREE_BLOCK_REF_KEY);
6780                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6781         }
6782
6783         btrfs_mark_buffer_dirty(leaf);
6784         btrfs_free_path(path);
6785
6786         ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6787         if (ret) { /* -ENOENT, logic error */
6788                 btrfs_err(fs_info, "update block group failed for %llu %llu",
6789                         ins->objectid, ins->offset);
6790                 BUG();
6791         }
6792
6793         trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6794         return ret;
6795 }
6796
6797 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6798                                      struct btrfs_root *root,
6799                                      u64 root_objectid, u64 owner,
6800                                      u64 offset, struct btrfs_key *ins)
6801 {
6802         int ret;
6803
6804         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6805
6806         ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6807                                          ins->offset, 0,
6808                                          root_objectid, owner, offset,
6809                                          BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6810         return ret;
6811 }
6812
6813 /*
6814  * this is used by the tree logging recovery code.  It records that
6815  * an extent has been allocated and makes sure to clear the free
6816  * space cache bits as well
6817  */
6818 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6819                                    struct btrfs_root *root,
6820                                    u64 root_objectid, u64 owner, u64 offset,
6821                                    struct btrfs_key *ins)
6822 {
6823         int ret;
6824         struct btrfs_block_group_cache *block_group;
6825
6826         /*
6827          * Mixed block groups will exclude before processing the log so we only
6828          * need to do the exlude dance if this fs isn't mixed.
6829          */
6830         if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6831                 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6832                 if (ret)
6833                         return ret;
6834         }
6835
6836         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6837         if (!block_group)
6838                 return -EINVAL;
6839
6840         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6841                                           RESERVE_ALLOC_NO_ACCOUNT);
6842         BUG_ON(ret); /* logic error */
6843         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6844                                          0, owner, offset, ins, 1);
6845         btrfs_put_block_group(block_group);
6846         return ret;
6847 }
6848
6849 static struct extent_buffer *
6850 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6851                       u64 bytenr, u32 blocksize, int level)
6852 {
6853         struct extent_buffer *buf;
6854
6855         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6856         if (!buf)
6857                 return ERR_PTR(-ENOMEM);
6858         btrfs_set_header_generation(buf, trans->transid);
6859         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6860         btrfs_tree_lock(buf);
6861         clean_tree_block(trans, root, buf);
6862         clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6863
6864         btrfs_set_lock_blocking(buf);
6865         btrfs_set_buffer_uptodate(buf);
6866
6867         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6868                 /*
6869                  * we allow two log transactions at a time, use different
6870                  * EXENT bit to differentiate dirty pages.
6871                  */
6872                 if (root->log_transid % 2 == 0)
6873                         set_extent_dirty(&root->dirty_log_pages, buf->start,
6874                                         buf->start + buf->len - 1, GFP_NOFS);
6875                 else
6876                         set_extent_new(&root->dirty_log_pages, buf->start,
6877                                         buf->start + buf->len - 1, GFP_NOFS);
6878         } else {
6879                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6880                          buf->start + buf->len - 1, GFP_NOFS);
6881         }
6882         trans->blocks_used++;
6883         /* this returns a buffer locked for blocking */
6884         return buf;
6885 }
6886
6887 static struct btrfs_block_rsv *
6888 use_block_rsv(struct btrfs_trans_handle *trans,
6889               struct btrfs_root *root, u32 blocksize)
6890 {
6891         struct btrfs_block_rsv *block_rsv;
6892         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6893         int ret;
6894         bool global_updated = false;
6895
6896         block_rsv = get_block_rsv(trans, root);
6897
6898         if (unlikely(block_rsv->size == 0))
6899                 goto try_reserve;
6900 again:
6901         ret = block_rsv_use_bytes(block_rsv, blocksize);
6902         if (!ret)
6903                 return block_rsv;
6904
6905         if (block_rsv->failfast)
6906                 return ERR_PTR(ret);
6907
6908         if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6909                 global_updated = true;
6910                 update_global_block_rsv(root->fs_info);
6911                 goto again;
6912         }
6913
6914         if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6915                 static DEFINE_RATELIMIT_STATE(_rs,
6916                                 DEFAULT_RATELIMIT_INTERVAL * 10,
6917                                 /*DEFAULT_RATELIMIT_BURST*/ 1);
6918                 if (__ratelimit(&_rs))
6919                         WARN(1, KERN_DEBUG
6920                                 "btrfs: block rsv returned %d\n", ret);
6921         }
6922 try_reserve:
6923         ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6924                                      BTRFS_RESERVE_NO_FLUSH);
6925         if (!ret)
6926                 return block_rsv;
6927         /*
6928          * If we couldn't reserve metadata bytes try and use some from
6929          * the global reserve if its space type is the same as the global
6930          * reservation.
6931          */
6932         if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6933             block_rsv->space_info == global_rsv->space_info) {
6934                 ret = block_rsv_use_bytes(global_rsv, blocksize);
6935                 if (!ret)
6936                         return global_rsv;
6937         }
6938         return ERR_PTR(ret);
6939 }
6940
6941 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6942                             struct btrfs_block_rsv *block_rsv, u32 blocksize)
6943 {
6944         block_rsv_add_bytes(block_rsv, blocksize, 0);
6945         block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6946 }
6947
6948 /*
6949  * finds a free extent and does all the dirty work required for allocation
6950  * returns the key for the extent through ins, and a tree buffer for
6951  * the first block of the extent through buf.
6952  *
6953  * returns the tree buffer or NULL.
6954  */
6955 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6956                                         struct btrfs_root *root, u32 blocksize,
6957                                         u64 parent, u64 root_objectid,
6958                                         struct btrfs_disk_key *key, int level,
6959                                         u64 hint, u64 empty_size)
6960 {
6961         struct btrfs_key ins;
6962         struct btrfs_block_rsv *block_rsv;
6963         struct extent_buffer *buf;
6964         u64 flags = 0;
6965         int ret;
6966         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6967                                                  SKINNY_METADATA);
6968
6969         block_rsv = use_block_rsv(trans, root, blocksize);
6970         if (IS_ERR(block_rsv))
6971                 return ERR_CAST(block_rsv);
6972
6973         ret = btrfs_reserve_extent(root, blocksize, blocksize,
6974                                    empty_size, hint, &ins, 0);
6975         if (ret) {
6976                 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6977                 return ERR_PTR(ret);
6978         }
6979
6980         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6981                                     blocksize, level);
6982         BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6983
6984         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6985                 if (parent == 0)
6986                         parent = ins.objectid;
6987                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6988         } else
6989                 BUG_ON(parent > 0);
6990
6991         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6992                 struct btrfs_delayed_extent_op *extent_op;
6993                 extent_op = btrfs_alloc_delayed_extent_op();
6994                 BUG_ON(!extent_op); /* -ENOMEM */
6995                 if (key)
6996                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
6997                 else
6998                         memset(&extent_op->key, 0, sizeof(extent_op->key));
6999                 extent_op->flags_to_set = flags;
7000                 if (skinny_metadata)
7001                         extent_op->update_key = 0;
7002                 else
7003                         extent_op->update_key = 1;
7004                 extent_op->update_flags = 1;
7005                 extent_op->is_data = 0;
7006                 extent_op->level = level;
7007
7008                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7009                                         ins.objectid,
7010                                         ins.offset, parent, root_objectid,
7011                                         level, BTRFS_ADD_DELAYED_EXTENT,
7012                                         extent_op, 0);
7013                 BUG_ON(ret); /* -ENOMEM */
7014         }
7015         return buf;
7016 }
7017
7018 struct walk_control {
7019         u64 refs[BTRFS_MAX_LEVEL];
7020         u64 flags[BTRFS_MAX_LEVEL];
7021         struct btrfs_key update_progress;
7022         int stage;
7023         int level;
7024         int shared_level;
7025         int update_ref;
7026         int keep_locks;
7027         int reada_slot;
7028         int reada_count;
7029         int for_reloc;
7030 };
7031
7032 #define DROP_REFERENCE  1
7033 #define UPDATE_BACKREF  2
7034
7035 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7036                                      struct btrfs_root *root,
7037                                      struct walk_control *wc,
7038                                      struct btrfs_path *path)
7039 {
7040         u64 bytenr;
7041         u64 generation;
7042         u64 refs;
7043         u64 flags;
7044         u32 nritems;
7045         u32 blocksize;
7046         struct btrfs_key key;
7047         struct extent_buffer *eb;
7048         int ret;
7049         int slot;
7050         int nread = 0;
7051
7052         if (path->slots[wc->level] < wc->reada_slot) {
7053                 wc->reada_count = wc->reada_count * 2 / 3;
7054                 wc->reada_count = max(wc->reada_count, 2);
7055         } else {
7056                 wc->reada_count = wc->reada_count * 3 / 2;
7057                 wc->reada_count = min_t(int, wc->reada_count,
7058                                         BTRFS_NODEPTRS_PER_BLOCK(root));
7059         }
7060
7061         eb = path->nodes[wc->level];
7062         nritems = btrfs_header_nritems(eb);
7063         blocksize = btrfs_level_size(root, wc->level - 1);
7064
7065         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7066                 if (nread >= wc->reada_count)
7067                         break;
7068
7069                 cond_resched();
7070                 bytenr = btrfs_node_blockptr(eb, slot);
7071                 generation = btrfs_node_ptr_generation(eb, slot);
7072
7073                 if (slot == path->slots[wc->level])
7074                         goto reada;
7075
7076                 if (wc->stage == UPDATE_BACKREF &&
7077                     generation <= root->root_key.offset)
7078                         continue;
7079
7080                 /* We don't lock the tree block, it's OK to be racy here */
7081                 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7082                                                wc->level - 1, 1, &refs,
7083                                                &flags);
7084                 /* We don't care about errors in readahead. */
7085                 if (ret < 0)
7086                         continue;
7087                 BUG_ON(refs == 0);
7088
7089                 if (wc->stage == DROP_REFERENCE) {
7090                         if (refs == 1)
7091                                 goto reada;
7092
7093                         if (wc->level == 1 &&
7094                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7095                                 continue;
7096                         if (!wc->update_ref ||
7097                             generation <= root->root_key.offset)
7098                                 continue;
7099                         btrfs_node_key_to_cpu(eb, &key, slot);
7100                         ret = btrfs_comp_cpu_keys(&key,
7101                                                   &wc->update_progress);
7102                         if (ret < 0)
7103                                 continue;
7104                 } else {
7105                         if (wc->level == 1 &&
7106                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7107                                 continue;
7108                 }
7109 reada:
7110                 ret = readahead_tree_block(root, bytenr, blocksize,
7111                                            generation);
7112                 if (ret)
7113                         break;
7114                 nread++;
7115         }
7116         wc->reada_slot = slot;
7117 }
7118
7119 /*
7120  * helper to process tree block while walking down the tree.
7121  *
7122  * when wc->stage == UPDATE_BACKREF, this function updates
7123  * back refs for pointers in the block.
7124  *
7125  * NOTE: return value 1 means we should stop walking down.
7126  */
7127 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7128                                    struct btrfs_root *root,
7129                                    struct btrfs_path *path,
7130                                    struct walk_control *wc, int lookup_info)
7131 {
7132         int level = wc->level;
7133         struct extent_buffer *eb = path->nodes[level];
7134         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7135         int ret;
7136
7137         if (wc->stage == UPDATE_BACKREF &&
7138             btrfs_header_owner(eb) != root->root_key.objectid)
7139                 return 1;
7140
7141         /*
7142          * when reference count of tree block is 1, it won't increase
7143          * again. once full backref flag is set, we never clear it.
7144          */
7145         if (lookup_info &&
7146             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7147              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7148                 BUG_ON(!path->locks[level]);
7149                 ret = btrfs_lookup_extent_info(trans, root,
7150                                                eb->start, level, 1,
7151                                                &wc->refs[level],
7152                                                &wc->flags[level]);
7153                 BUG_ON(ret == -ENOMEM);
7154                 if (ret)
7155                         return ret;
7156                 BUG_ON(wc->refs[level] == 0);
7157         }
7158
7159         if (wc->stage == DROP_REFERENCE) {
7160                 if (wc->refs[level] > 1)
7161                         return 1;
7162
7163                 if (path->locks[level] && !wc->keep_locks) {
7164                         btrfs_tree_unlock_rw(eb, path->locks[level]);
7165                         path->locks[level] = 0;
7166                 }
7167                 return 0;
7168         }
7169
7170         /* wc->stage == UPDATE_BACKREF */
7171         if (!(wc->flags[level] & flag)) {
7172                 BUG_ON(!path->locks[level]);
7173                 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7174                 BUG_ON(ret); /* -ENOMEM */
7175                 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7176                 BUG_ON(ret); /* -ENOMEM */
7177                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7178                                                   eb->len, flag,
7179                                                   btrfs_header_level(eb), 0);
7180                 BUG_ON(ret); /* -ENOMEM */
7181                 wc->flags[level] |= flag;
7182         }
7183
7184         /*
7185          * the block is shared by multiple trees, so it's not good to
7186          * keep the tree lock
7187          */
7188         if (path->locks[level] && level > 0) {
7189                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7190                 path->locks[level] = 0;
7191         }
7192         return 0;
7193 }
7194
7195 /*
7196  * helper to process tree block pointer.
7197  *
7198  * when wc->stage == DROP_REFERENCE, this function checks
7199  * reference count of the block pointed to. if the block
7200  * is shared and we need update back refs for the subtree
7201  * rooted at the block, this function changes wc->stage to
7202  * UPDATE_BACKREF. if the block is shared and there is no
7203  * need to update back, this function drops the reference
7204  * to the block.
7205  *
7206  * NOTE: return value 1 means we should stop walking down.
7207  */
7208 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7209                                  struct btrfs_root *root,
7210                                  struct btrfs_path *path,
7211                                  struct walk_control *wc, int *lookup_info)
7212 {
7213         u64 bytenr;
7214         u64 generation;
7215         u64 parent;
7216         u32 blocksize;
7217         struct btrfs_key key;
7218         struct extent_buffer *next;
7219         int level = wc->level;
7220         int reada = 0;
7221         int ret = 0;
7222
7223         generation = btrfs_node_ptr_generation(path->nodes[level],
7224                                                path->slots[level]);
7225         /*
7226          * if the lower level block was created before the snapshot
7227          * was created, we know there is no need to update back refs
7228          * for the subtree
7229          */
7230         if (wc->stage == UPDATE_BACKREF &&
7231             generation <= root->root_key.offset) {
7232                 *lookup_info = 1;
7233                 return 1;
7234         }
7235
7236         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7237         blocksize = btrfs_level_size(root, level - 1);
7238
7239         next = btrfs_find_tree_block(root, bytenr, blocksize);
7240         if (!next) {
7241                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7242                 if (!next)
7243                         return -ENOMEM;
7244                 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7245                                                level - 1);
7246                 reada = 1;
7247         }
7248         btrfs_tree_lock(next);
7249         btrfs_set_lock_blocking(next);
7250
7251         ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7252                                        &wc->refs[level - 1],
7253                                        &wc->flags[level - 1]);
7254         if (ret < 0) {
7255                 btrfs_tree_unlock(next);
7256                 return ret;
7257         }
7258
7259         if (unlikely(wc->refs[level - 1] == 0)) {
7260                 btrfs_err(root->fs_info, "Missing references.");
7261                 BUG();
7262         }
7263         *lookup_info = 0;
7264
7265         if (wc->stage == DROP_REFERENCE) {
7266                 if (wc->refs[level - 1] > 1) {
7267                         if (level == 1 &&
7268                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7269                                 goto skip;
7270
7271                         if (!wc->update_ref ||
7272                             generation <= root->root_key.offset)
7273                                 goto skip;
7274
7275                         btrfs_node_key_to_cpu(path->nodes[level], &key,
7276                                               path->slots[level]);
7277                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7278                         if (ret < 0)
7279                                 goto skip;
7280
7281                         wc->stage = UPDATE_BACKREF;
7282                         wc->shared_level = level - 1;
7283                 }
7284         } else {
7285                 if (level == 1 &&
7286                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7287                         goto skip;
7288         }
7289
7290         if (!btrfs_buffer_uptodate(next, generation, 0)) {
7291                 btrfs_tree_unlock(next);
7292                 free_extent_buffer(next);
7293                 next = NULL;
7294                 *lookup_info = 1;
7295         }
7296
7297         if (!next) {
7298                 if (reada && level == 1)
7299                         reada_walk_down(trans, root, wc, path);
7300                 next = read_tree_block(root, bytenr, blocksize, generation);
7301                 if (!next || !extent_buffer_uptodate(next)) {
7302                         free_extent_buffer(next);
7303                         return -EIO;
7304                 }
7305                 btrfs_tree_lock(next);
7306                 btrfs_set_lock_blocking(next);
7307         }
7308
7309         level--;
7310         BUG_ON(level != btrfs_header_level(next));
7311         path->nodes[level] = next;
7312         path->slots[level] = 0;
7313         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7314         wc->level = level;
7315         if (wc->level == 1)
7316                 wc->reada_slot = 0;
7317         return 0;
7318 skip:
7319         wc->refs[level - 1] = 0;
7320         wc->flags[level - 1] = 0;
7321         if (wc->stage == DROP_REFERENCE) {
7322                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7323                         parent = path->nodes[level]->start;
7324                 } else {
7325                         BUG_ON(root->root_key.objectid !=
7326                                btrfs_header_owner(path->nodes[level]));
7327                         parent = 0;
7328                 }
7329
7330                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7331                                 root->root_key.objectid, level - 1, 0, 0);
7332                 BUG_ON(ret); /* -ENOMEM */
7333         }
7334         btrfs_tree_unlock(next);
7335         free_extent_buffer(next);
7336         *lookup_info = 1;
7337         return 1;
7338 }
7339
7340 /*
7341  * helper to process tree block while walking up the tree.
7342  *
7343  * when wc->stage == DROP_REFERENCE, this function drops
7344  * reference count on the block.
7345  *
7346  * when wc->stage == UPDATE_BACKREF, this function changes
7347  * wc->stage back to DROP_REFERENCE if we changed wc->stage
7348  * to UPDATE_BACKREF previously while processing the block.
7349  *
7350  * NOTE: return value 1 means we should stop walking up.
7351  */
7352 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7353                                  struct btrfs_root *root,
7354                                  struct btrfs_path *path,
7355                                  struct walk_control *wc)
7356 {
7357         int ret;
7358         int level = wc->level;
7359         struct extent_buffer *eb = path->nodes[level];
7360         u64 parent = 0;
7361
7362         if (wc->stage == UPDATE_BACKREF) {
7363                 BUG_ON(wc->shared_level < level);
7364                 if (level < wc->shared_level)
7365                         goto out;
7366
7367                 ret = find_next_key(path, level + 1, &wc->update_progress);
7368                 if (ret > 0)
7369                         wc->update_ref = 0;
7370
7371                 wc->stage = DROP_REFERENCE;
7372                 wc->shared_level = -1;
7373                 path->slots[level] = 0;
7374
7375                 /*
7376                  * check reference count again if the block isn't locked.
7377                  * we should start walking down the tree again if reference
7378                  * count is one.
7379                  */
7380                 if (!path->locks[level]) {
7381                         BUG_ON(level == 0);
7382                         btrfs_tree_lock(eb);
7383                         btrfs_set_lock_blocking(eb);
7384                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7385
7386                         ret = btrfs_lookup_extent_info(trans, root,
7387                                                        eb->start, level, 1,
7388                                                        &wc->refs[level],
7389                                                        &wc->flags[level]);
7390                         if (ret < 0) {
7391                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7392                                 path->locks[level] = 0;
7393                                 return ret;
7394                         }
7395                         BUG_ON(wc->refs[level] == 0);
7396                         if (wc->refs[level] == 1) {
7397                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7398                                 path->locks[level] = 0;
7399                                 return 1;
7400                         }
7401                 }
7402         }
7403
7404         /* wc->stage == DROP_REFERENCE */
7405         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7406
7407         if (wc->refs[level] == 1) {
7408                 if (level == 0) {
7409                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7410                                 ret = btrfs_dec_ref(trans, root, eb, 1,
7411                                                     wc->for_reloc);
7412                         else
7413                                 ret = btrfs_dec_ref(trans, root, eb, 0,
7414                                                     wc->for_reloc);
7415                         BUG_ON(ret); /* -ENOMEM */
7416                 }
7417                 /* make block locked assertion in clean_tree_block happy */
7418                 if (!path->locks[level] &&
7419                     btrfs_header_generation(eb) == trans->transid) {
7420                         btrfs_tree_lock(eb);
7421                         btrfs_set_lock_blocking(eb);
7422                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7423                 }
7424                 clean_tree_block(trans, root, eb);
7425         }
7426
7427         if (eb == root->node) {
7428                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7429                         parent = eb->start;
7430                 else
7431                         BUG_ON(root->root_key.objectid !=
7432                                btrfs_header_owner(eb));
7433         } else {
7434                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7435                         parent = path->nodes[level + 1]->start;
7436                 else
7437                         BUG_ON(root->root_key.objectid !=
7438                                btrfs_header_owner(path->nodes[level + 1]));
7439         }
7440
7441         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7442 out:
7443         wc->refs[level] = 0;
7444         wc->flags[level] = 0;
7445         return 0;
7446 }
7447
7448 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7449                                    struct btrfs_root *root,
7450                                    struct btrfs_path *path,
7451                                    struct walk_control *wc)
7452 {
7453         int level = wc->level;
7454         int lookup_info = 1;
7455         int ret;
7456
7457         while (level >= 0) {
7458                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7459                 if (ret > 0)
7460                         break;
7461
7462                 if (level == 0)
7463                         break;
7464
7465                 if (path->slots[level] >=
7466                     btrfs_header_nritems(path->nodes[level]))
7467                         break;
7468
7469                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7470                 if (ret > 0) {
7471                         path->slots[level]++;
7472                         continue;
7473                 } else if (ret < 0)
7474                         return ret;
7475                 level = wc->level;
7476         }
7477         return 0;
7478 }
7479
7480 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7481                                  struct btrfs_root *root,
7482                                  struct btrfs_path *path,
7483                                  struct walk_control *wc, int max_level)
7484 {
7485         int level = wc->level;
7486         int ret;
7487
7488         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7489         while (level < max_level && path->nodes[level]) {
7490                 wc->level = level;
7491                 if (path->slots[level] + 1 <
7492                     btrfs_header_nritems(path->nodes[level])) {
7493                         path->slots[level]++;
7494                         return 0;
7495                 } else {
7496                         ret = walk_up_proc(trans, root, path, wc);
7497                         if (ret > 0)
7498                                 return 0;
7499
7500                         if (path->locks[level]) {
7501                                 btrfs_tree_unlock_rw(path->nodes[level],
7502                                                      path->locks[level]);
7503                                 path->locks[level] = 0;
7504                         }
7505                         free_extent_buffer(path->nodes[level]);
7506                         path->nodes[level] = NULL;
7507                         level++;
7508                 }
7509         }
7510         return 1;
7511 }
7512
7513 /*
7514  * drop a subvolume tree.
7515  *
7516  * this function traverses the tree freeing any blocks that only
7517  * referenced by the tree.
7518  *
7519  * when a shared tree block is found. this function decreases its
7520  * reference count by one. if update_ref is true, this function
7521  * also make sure backrefs for the shared block and all lower level
7522  * blocks are properly updated.
7523  *
7524  * If called with for_reloc == 0, may exit early with -EAGAIN
7525  */
7526 int btrfs_drop_snapshot(struct btrfs_root *root,
7527                          struct btrfs_block_rsv *block_rsv, int update_ref,
7528                          int for_reloc)
7529 {
7530         struct btrfs_path *path;
7531         struct btrfs_trans_handle *trans;
7532         struct btrfs_root *tree_root = root->fs_info->tree_root;
7533         struct btrfs_root_item *root_item = &root->root_item;
7534         struct walk_control *wc;
7535         struct btrfs_key key;
7536         int err = 0;
7537         int ret;
7538         int level;
7539         bool root_dropped = false;
7540
7541         path = btrfs_alloc_path();
7542         if (!path) {
7543                 err = -ENOMEM;
7544                 goto out;
7545         }
7546
7547         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7548         if (!wc) {
7549                 btrfs_free_path(path);
7550                 err = -ENOMEM;
7551                 goto out;
7552         }
7553
7554         trans = btrfs_start_transaction(tree_root, 0);
7555         if (IS_ERR(trans)) {
7556                 err = PTR_ERR(trans);
7557                 goto out_free;
7558         }
7559
7560         if (block_rsv)
7561                 trans->block_rsv = block_rsv;
7562
7563         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7564                 level = btrfs_header_level(root->node);
7565                 path->nodes[level] = btrfs_lock_root_node(root);
7566                 btrfs_set_lock_blocking(path->nodes[level]);
7567                 path->slots[level] = 0;
7568                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7569                 memset(&wc->update_progress, 0,
7570                        sizeof(wc->update_progress));
7571         } else {
7572                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7573                 memcpy(&wc->update_progress, &key,
7574                        sizeof(wc->update_progress));
7575
7576                 level = root_item->drop_level;
7577                 BUG_ON(level == 0);
7578                 path->lowest_level = level;
7579                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7580                 path->lowest_level = 0;
7581                 if (ret < 0) {
7582                         err = ret;
7583                         goto out_end_trans;
7584                 }
7585                 WARN_ON(ret > 0);
7586
7587                 /*
7588                  * unlock our path, this is safe because only this
7589                  * function is allowed to delete this snapshot
7590                  */
7591                 btrfs_unlock_up_safe(path, 0);
7592
7593                 level = btrfs_header_level(root->node);
7594                 while (1) {
7595                         btrfs_tree_lock(path->nodes[level]);
7596                         btrfs_set_lock_blocking(path->nodes[level]);
7597                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7598
7599                         ret = btrfs_lookup_extent_info(trans, root,
7600                                                 path->nodes[level]->start,
7601                                                 level, 1, &wc->refs[level],
7602                                                 &wc->flags[level]);
7603                         if (ret < 0) {
7604                                 err = ret;
7605                                 goto out_end_trans;
7606                         }
7607                         BUG_ON(wc->refs[level] == 0);
7608
7609                         if (level == root_item->drop_level)
7610                                 break;
7611
7612                         btrfs_tree_unlock(path->nodes[level]);
7613                         path->locks[level] = 0;
7614                         WARN_ON(wc->refs[level] != 1);
7615                         level--;
7616                 }
7617         }
7618
7619         wc->level = level;
7620         wc->shared_level = -1;
7621         wc->stage = DROP_REFERENCE;
7622         wc->update_ref = update_ref;
7623         wc->keep_locks = 0;
7624         wc->for_reloc = for_reloc;
7625         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7626
7627         while (1) {
7628
7629                 ret = walk_down_tree(trans, root, path, wc);
7630                 if (ret < 0) {
7631                         err = ret;
7632                         break;
7633                 }
7634
7635                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7636                 if (ret < 0) {
7637                         err = ret;
7638                         break;
7639                 }
7640
7641                 if (ret > 0) {
7642                         BUG_ON(wc->stage != DROP_REFERENCE);
7643                         break;
7644                 }
7645
7646                 if (wc->stage == DROP_REFERENCE) {
7647                         level = wc->level;
7648                         btrfs_node_key(path->nodes[level],
7649                                        &root_item->drop_progress,
7650                                        path->slots[level]);
7651                         root_item->drop_level = level;
7652                 }
7653
7654                 BUG_ON(wc->level == 0);
7655                 if (btrfs_should_end_transaction(trans, tree_root) ||
7656                     (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7657                         ret = btrfs_update_root(trans, tree_root,
7658                                                 &root->root_key,
7659                                                 root_item);
7660                         if (ret) {
7661                                 btrfs_abort_transaction(trans, tree_root, ret);
7662                                 err = ret;
7663                                 goto out_end_trans;
7664                         }
7665
7666                         btrfs_end_transaction_throttle(trans, tree_root);
7667                         if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7668                                 pr_debug("btrfs: drop snapshot early exit\n");
7669                                 err = -EAGAIN;
7670                                 goto out_free;
7671                         }
7672
7673                         trans = btrfs_start_transaction(tree_root, 0);
7674                         if (IS_ERR(trans)) {
7675                                 err = PTR_ERR(trans);
7676                                 goto out_free;
7677                         }
7678                         if (block_rsv)
7679                                 trans->block_rsv = block_rsv;
7680                 }
7681         }
7682         btrfs_release_path(path);
7683         if (err)
7684                 goto out_end_trans;
7685
7686         ret = btrfs_del_root(trans, tree_root, &root->root_key);
7687         if (ret) {
7688                 btrfs_abort_transaction(trans, tree_root, ret);
7689                 goto out_end_trans;
7690         }
7691
7692         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7693                 ret = btrfs_find_root(tree_root, &root->root_key, path,
7694                                       NULL, NULL);
7695                 if (ret < 0) {
7696                         btrfs_abort_transaction(trans, tree_root, ret);
7697                         err = ret;
7698                         goto out_end_trans;
7699                 } else if (ret > 0) {
7700                         /* if we fail to delete the orphan item this time
7701                          * around, it'll get picked up the next time.
7702                          *
7703                          * The most common failure here is just -ENOENT.
7704                          */
7705                         btrfs_del_orphan_item(trans, tree_root,
7706                                               root->root_key.objectid);
7707                 }
7708         }
7709
7710         if (root->in_radix) {
7711                 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7712         } else {
7713                 free_extent_buffer(root->node);
7714                 free_extent_buffer(root->commit_root);
7715                 btrfs_put_fs_root(root);
7716         }
7717         root_dropped = true;
7718 out_end_trans:
7719         btrfs_end_transaction_throttle(trans, tree_root);
7720 out_free:
7721         kfree(wc);
7722         btrfs_free_path(path);
7723 out:
7724         /*
7725          * So if we need to stop dropping the snapshot for whatever reason we
7726          * need to make sure to add it back to the dead root list so that we
7727          * keep trying to do the work later.  This also cleans up roots if we
7728          * don't have it in the radix (like when we recover after a power fail
7729          * or unmount) so we don't leak memory.
7730          */
7731         if (!for_reloc && root_dropped == false)
7732                 btrfs_add_dead_root(root);
7733         if (err)
7734                 btrfs_std_error(root->fs_info, err);
7735         return err;
7736 }
7737
7738 /*
7739  * drop subtree rooted at tree block 'node'.
7740  *
7741  * NOTE: this function will unlock and release tree block 'node'
7742  * only used by relocation code
7743  */
7744 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7745                         struct btrfs_root *root,
7746                         struct extent_buffer *node,
7747                         struct extent_buffer *parent)
7748 {
7749         struct btrfs_path *path;
7750         struct walk_control *wc;
7751         int level;
7752         int parent_level;
7753         int ret = 0;
7754         int wret;
7755
7756         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7757
7758         path = btrfs_alloc_path();
7759         if (!path)
7760                 return -ENOMEM;
7761
7762         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7763         if (!wc) {
7764                 btrfs_free_path(path);
7765                 return -ENOMEM;
7766         }
7767
7768         btrfs_assert_tree_locked(parent);
7769         parent_level = btrfs_header_level(parent);
7770         extent_buffer_get(parent);
7771         path->nodes[parent_level] = parent;
7772         path->slots[parent_level] = btrfs_header_nritems(parent);
7773
7774         btrfs_assert_tree_locked(node);
7775         level = btrfs_header_level(node);
7776         path->nodes[level] = node;
7777         path->slots[level] = 0;
7778         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7779
7780         wc->refs[parent_level] = 1;
7781         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7782         wc->level = level;
7783         wc->shared_level = -1;
7784         wc->stage = DROP_REFERENCE;
7785         wc->update_ref = 0;
7786         wc->keep_locks = 1;
7787         wc->for_reloc = 1;
7788         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7789
7790         while (1) {
7791                 wret = walk_down_tree(trans, root, path, wc);
7792                 if (wret < 0) {
7793                         ret = wret;
7794                         break;
7795                 }
7796
7797                 wret = walk_up_tree(trans, root, path, wc, parent_level);
7798                 if (wret < 0)
7799                         ret = wret;
7800                 if (wret != 0)
7801                         break;
7802         }
7803
7804         kfree(wc);
7805         btrfs_free_path(path);
7806         return ret;
7807 }
7808
7809 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7810 {
7811         u64 num_devices;
7812         u64 stripped;
7813
7814         /*
7815          * if restripe for this chunk_type is on pick target profile and
7816          * return, otherwise do the usual balance
7817          */
7818         stripped = get_restripe_target(root->fs_info, flags);
7819         if (stripped)
7820                 return extended_to_chunk(stripped);
7821
7822         /*
7823          * we add in the count of missing devices because we want
7824          * to make sure that any RAID levels on a degraded FS
7825          * continue to be honored.
7826          */
7827         num_devices = root->fs_info->fs_devices->rw_devices +
7828                 root->fs_info->fs_devices->missing_devices;
7829
7830         stripped = BTRFS_BLOCK_GROUP_RAID0 |
7831                 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7832                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7833
7834         if (num_devices == 1) {
7835                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7836                 stripped = flags & ~stripped;
7837
7838                 /* turn raid0 into single device chunks */
7839                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7840                         return stripped;
7841
7842                 /* turn mirroring into duplication */
7843                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7844                              BTRFS_BLOCK_GROUP_RAID10))
7845                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7846         } else {
7847                 /* they already had raid on here, just return */
7848                 if (flags & stripped)
7849                         return flags;
7850
7851                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7852                 stripped = flags & ~stripped;
7853
7854                 /* switch duplicated blocks with raid1 */
7855                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7856                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7857
7858                 /* this is drive concat, leave it alone */
7859         }
7860
7861         return flags;
7862 }
7863
7864 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7865 {
7866         struct btrfs_space_info *sinfo = cache->space_info;
7867         u64 num_bytes;
7868         u64 min_allocable_bytes;
7869         int ret = -ENOSPC;
7870
7871
7872         /*
7873          * We need some metadata space and system metadata space for
7874          * allocating chunks in some corner cases until we force to set
7875          * it to be readonly.
7876          */
7877         if ((sinfo->flags &
7878              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7879             !force)
7880                 min_allocable_bytes = 1 * 1024 * 1024;
7881         else
7882                 min_allocable_bytes = 0;
7883
7884         spin_lock(&sinfo->lock);
7885         spin_lock(&cache->lock);
7886
7887         if (cache->ro) {
7888                 ret = 0;
7889                 goto out;
7890         }
7891
7892         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7893                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7894
7895         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7896             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7897             min_allocable_bytes <= sinfo->total_bytes) {
7898                 sinfo->bytes_readonly += num_bytes;
7899                 cache->ro = 1;
7900                 ret = 0;
7901         }
7902 out:
7903         spin_unlock(&cache->lock);
7904         spin_unlock(&sinfo->lock);
7905         return ret;
7906 }
7907
7908 int btrfs_set_block_group_ro(struct btrfs_root *root,
7909                              struct btrfs_block_group_cache *cache)
7910
7911 {
7912         struct btrfs_trans_handle *trans;
7913         u64 alloc_flags;
7914         int ret;
7915
7916         BUG_ON(cache->ro);
7917
7918         trans = btrfs_join_transaction(root);
7919         if (IS_ERR(trans))
7920                 return PTR_ERR(trans);
7921
7922         alloc_flags = update_block_group_flags(root, cache->flags);
7923         if (alloc_flags != cache->flags) {
7924                 ret = do_chunk_alloc(trans, root, alloc_flags,
7925                                      CHUNK_ALLOC_FORCE);
7926                 if (ret < 0)
7927                         goto out;
7928         }
7929
7930         ret = set_block_group_ro(cache, 0);
7931         if (!ret)
7932                 goto out;
7933         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7934         ret = do_chunk_alloc(trans, root, alloc_flags,
7935                              CHUNK_ALLOC_FORCE);
7936         if (ret < 0)
7937                 goto out;
7938         ret = set_block_group_ro(cache, 0);
7939 out:
7940         btrfs_end_transaction(trans, root);
7941         return ret;
7942 }
7943
7944 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7945                             struct btrfs_root *root, u64 type)
7946 {
7947         u64 alloc_flags = get_alloc_profile(root, type);
7948         return do_chunk_alloc(trans, root, alloc_flags,
7949                               CHUNK_ALLOC_FORCE);
7950 }
7951
7952 /*
7953  * helper to account the unused space of all the readonly block group in the
7954  * list. takes mirrors into account.
7955  */
7956 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7957 {
7958         struct btrfs_block_group_cache *block_group;
7959         u64 free_bytes = 0;
7960         int factor;
7961
7962         list_for_each_entry(block_group, groups_list, list) {
7963                 spin_lock(&block_group->lock);
7964
7965                 if (!block_group->ro) {
7966                         spin_unlock(&block_group->lock);
7967                         continue;
7968                 }
7969
7970                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7971                                           BTRFS_BLOCK_GROUP_RAID10 |
7972                                           BTRFS_BLOCK_GROUP_DUP))
7973                         factor = 2;
7974                 else
7975                         factor = 1;
7976
7977                 free_bytes += (block_group->key.offset -
7978                                btrfs_block_group_used(&block_group->item)) *
7979                                factor;
7980
7981                 spin_unlock(&block_group->lock);
7982         }
7983
7984         return free_bytes;
7985 }
7986
7987 /*
7988  * helper to account the unused space of all the readonly block group in the
7989  * space_info. takes mirrors into account.
7990  */
7991 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7992 {
7993         int i;
7994         u64 free_bytes = 0;
7995
7996         spin_lock(&sinfo->lock);
7997
7998         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7999                 if (!list_empty(&sinfo->block_groups[i]))
8000                         free_bytes += __btrfs_get_ro_block_group_free_space(
8001                                                 &sinfo->block_groups[i]);
8002
8003         spin_unlock(&sinfo->lock);
8004
8005         return free_bytes;
8006 }
8007
8008 void btrfs_set_block_group_rw(struct btrfs_root *root,
8009                               struct btrfs_block_group_cache *cache)
8010 {
8011         struct btrfs_space_info *sinfo = cache->space_info;
8012         u64 num_bytes;
8013
8014         BUG_ON(!cache->ro);
8015
8016         spin_lock(&sinfo->lock);
8017         spin_lock(&cache->lock);
8018         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8019                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8020         sinfo->bytes_readonly -= num_bytes;
8021         cache->ro = 0;
8022         spin_unlock(&cache->lock);
8023         spin_unlock(&sinfo->lock);
8024 }
8025
8026 /*
8027  * checks to see if its even possible to relocate this block group.
8028  *
8029  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8030  * ok to go ahead and try.
8031  */
8032 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8033 {
8034         struct btrfs_block_group_cache *block_group;
8035         struct btrfs_space_info *space_info;
8036         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8037         struct btrfs_device *device;
8038         struct btrfs_trans_handle *trans;
8039         u64 min_free;
8040         u64 dev_min = 1;
8041         u64 dev_nr = 0;
8042         u64 target;
8043         int index;
8044         int full = 0;
8045         int ret = 0;
8046
8047         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8048
8049         /* odd, couldn't find the block group, leave it alone */
8050         if (!block_group)
8051                 return -1;
8052
8053         min_free = btrfs_block_group_used(&block_group->item);
8054
8055         /* no bytes used, we're good */
8056         if (!min_free)
8057                 goto out;
8058
8059         space_info = block_group->space_info;
8060         spin_lock(&space_info->lock);
8061
8062         full = space_info->full;
8063
8064         /*
8065          * if this is the last block group we have in this space, we can't
8066          * relocate it unless we're able to allocate a new chunk below.
8067          *
8068          * Otherwise, we need to make sure we have room in the space to handle
8069          * all of the extents from this block group.  If we can, we're good
8070          */
8071         if ((space_info->total_bytes != block_group->key.offset) &&
8072             (space_info->bytes_used + space_info->bytes_reserved +
8073              space_info->bytes_pinned + space_info->bytes_readonly +
8074              min_free < space_info->total_bytes)) {
8075                 spin_unlock(&space_info->lock);
8076                 goto out;
8077         }
8078         spin_unlock(&space_info->lock);
8079
8080         /*
8081          * ok we don't have enough space, but maybe we have free space on our
8082          * devices to allocate new chunks for relocation, so loop through our
8083          * alloc devices and guess if we have enough space.  if this block
8084          * group is going to be restriped, run checks against the target
8085          * profile instead of the current one.
8086          */
8087         ret = -1;
8088
8089         /*
8090          * index:
8091          *      0: raid10
8092          *      1: raid1
8093          *      2: dup
8094          *      3: raid0
8095          *      4: single
8096          */
8097         target = get_restripe_target(root->fs_info, block_group->flags);
8098         if (target) {
8099                 index = __get_raid_index(extended_to_chunk(target));
8100         } else {
8101                 /*
8102                  * this is just a balance, so if we were marked as full
8103                  * we know there is no space for a new chunk
8104                  */
8105                 if (full)
8106                         goto out;
8107
8108                 index = get_block_group_index(block_group);
8109         }
8110
8111         if (index == BTRFS_RAID_RAID10) {
8112                 dev_min = 4;
8113                 /* Divide by 2 */
8114                 min_free >>= 1;
8115         } else if (index == BTRFS_RAID_RAID1) {
8116                 dev_min = 2;
8117         } else if (index == BTRFS_RAID_DUP) {
8118                 /* Multiply by 2 */
8119                 min_free <<= 1;
8120         } else if (index == BTRFS_RAID_RAID0) {
8121                 dev_min = fs_devices->rw_devices;
8122                 do_div(min_free, dev_min);
8123         }
8124
8125         /* We need to do this so that we can look at pending chunks */
8126         trans = btrfs_join_transaction(root);
8127         if (IS_ERR(trans)) {
8128                 ret = PTR_ERR(trans);
8129                 goto out;
8130         }
8131
8132         mutex_lock(&root->fs_info->chunk_mutex);
8133         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8134                 u64 dev_offset;
8135
8136                 /*
8137                  * check to make sure we can actually find a chunk with enough
8138                  * space to fit our block group in.
8139                  */
8140                 if (device->total_bytes > device->bytes_used + min_free &&
8141                     !device->is_tgtdev_for_dev_replace) {
8142                         ret = find_free_dev_extent(trans, device, min_free,
8143                                                    &dev_offset, NULL);
8144                         if (!ret)
8145                                 dev_nr++;
8146
8147                         if (dev_nr >= dev_min)
8148                                 break;
8149
8150                         ret = -1;
8151                 }
8152         }
8153         mutex_unlock(&root->fs_info->chunk_mutex);
8154         btrfs_end_transaction(trans, root);
8155 out:
8156         btrfs_put_block_group(block_group);
8157         return ret;
8158 }
8159
8160 static int find_first_block_group(struct btrfs_root *root,
8161                 struct btrfs_path *path, struct btrfs_key *key)
8162 {
8163         int ret = 0;
8164         struct btrfs_key found_key;
8165         struct extent_buffer *leaf;
8166         int slot;
8167
8168         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8169         if (ret < 0)
8170                 goto out;
8171
8172         while (1) {
8173                 slot = path->slots[0];
8174                 leaf = path->nodes[0];
8175                 if (slot >= btrfs_header_nritems(leaf)) {
8176                         ret = btrfs_next_leaf(root, path);
8177                         if (ret == 0)
8178                                 continue;
8179                         if (ret < 0)
8180                                 goto out;
8181                         break;
8182                 }
8183                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8184
8185                 if (found_key.objectid >= key->objectid &&
8186                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8187                         ret = 0;
8188                         goto out;
8189                 }
8190                 path->slots[0]++;
8191         }
8192 out:
8193         return ret;
8194 }
8195
8196 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8197 {
8198         struct btrfs_block_group_cache *block_group;
8199         u64 last = 0;
8200
8201         while (1) {
8202                 struct inode *inode;
8203
8204                 block_group = btrfs_lookup_first_block_group(info, last);
8205                 while (block_group) {
8206                         spin_lock(&block_group->lock);
8207                         if (block_group->iref)
8208                                 break;
8209                         spin_unlock(&block_group->lock);
8210                         block_group = next_block_group(info->tree_root,
8211                                                        block_group);
8212                 }
8213                 if (!block_group) {
8214                         if (last == 0)
8215                                 break;
8216                         last = 0;
8217                         continue;
8218                 }
8219
8220                 inode = block_group->inode;
8221                 block_group->iref = 0;
8222                 block_group->inode = NULL;
8223                 spin_unlock(&block_group->lock);
8224                 iput(inode);
8225                 last = block_group->key.objectid + block_group->key.offset;
8226                 btrfs_put_block_group(block_group);
8227         }
8228 }
8229
8230 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8231 {
8232         struct btrfs_block_group_cache *block_group;
8233         struct btrfs_space_info *space_info;
8234         struct btrfs_caching_control *caching_ctl;
8235         struct rb_node *n;
8236
8237         down_write(&info->extent_commit_sem);
8238         while (!list_empty(&info->caching_block_groups)) {
8239                 caching_ctl = list_entry(info->caching_block_groups.next,
8240                                          struct btrfs_caching_control, list);
8241                 list_del(&caching_ctl->list);
8242                 put_caching_control(caching_ctl);
8243         }
8244         up_write(&info->extent_commit_sem);
8245
8246         spin_lock(&info->block_group_cache_lock);
8247         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8248                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8249                                        cache_node);
8250                 rb_erase(&block_group->cache_node,
8251                          &info->block_group_cache_tree);
8252                 spin_unlock(&info->block_group_cache_lock);
8253
8254                 down_write(&block_group->space_info->groups_sem);
8255                 list_del(&block_group->list);
8256                 up_write(&block_group->space_info->groups_sem);
8257
8258                 if (block_group->cached == BTRFS_CACHE_STARTED)
8259                         wait_block_group_cache_done(block_group);
8260
8261                 /*
8262                  * We haven't cached this block group, which means we could
8263                  * possibly have excluded extents on this block group.
8264                  */
8265                 if (block_group->cached == BTRFS_CACHE_NO ||
8266                     block_group->cached == BTRFS_CACHE_ERROR)
8267                         free_excluded_extents(info->extent_root, block_group);
8268
8269                 btrfs_remove_free_space_cache(block_group);
8270                 btrfs_put_block_group(block_group);
8271
8272                 spin_lock(&info->block_group_cache_lock);
8273         }
8274         spin_unlock(&info->block_group_cache_lock);
8275
8276         /* now that all the block groups are freed, go through and
8277          * free all the space_info structs.  This is only called during
8278          * the final stages of unmount, and so we know nobody is
8279          * using them.  We call synchronize_rcu() once before we start,
8280          * just to be on the safe side.
8281          */
8282         synchronize_rcu();
8283
8284         release_global_block_rsv(info);
8285
8286         while(!list_empty(&info->space_info)) {
8287                 space_info = list_entry(info->space_info.next,
8288                                         struct btrfs_space_info,
8289                                         list);
8290                 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8291                         if (space_info->bytes_pinned > 0 ||
8292                             space_info->bytes_reserved > 0 ||
8293                             space_info->bytes_may_use > 0) {
8294                                 WARN_ON(1);
8295                                 dump_space_info(space_info, 0, 0);
8296                         }
8297                 }
8298                 percpu_counter_destroy(&space_info->total_bytes_pinned);
8299                 list_del(&space_info->list);
8300                 kfree(space_info);
8301         }
8302         return 0;
8303 }
8304
8305 static void __link_block_group(struct btrfs_space_info *space_info,
8306                                struct btrfs_block_group_cache *cache)
8307 {
8308         int index = get_block_group_index(cache);
8309
8310         down_write(&space_info->groups_sem);
8311         list_add_tail(&cache->list, &space_info->block_groups[index]);
8312         up_write(&space_info->groups_sem);
8313 }
8314
8315 int btrfs_read_block_groups(struct btrfs_root *root)
8316 {
8317         struct btrfs_path *path;
8318         int ret;
8319         struct btrfs_block_group_cache *cache;
8320         struct btrfs_fs_info *info = root->fs_info;
8321         struct btrfs_space_info *space_info;
8322         struct btrfs_key key;
8323         struct btrfs_key found_key;
8324         struct extent_buffer *leaf;
8325         int need_clear = 0;
8326         u64 cache_gen;
8327
8328         root = info->extent_root;
8329         key.objectid = 0;
8330         key.offset = 0;
8331         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8332         path = btrfs_alloc_path();
8333         if (!path)
8334                 return -ENOMEM;
8335         path->reada = 1;
8336
8337         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8338         if (btrfs_test_opt(root, SPACE_CACHE) &&
8339             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8340                 need_clear = 1;
8341         if (btrfs_test_opt(root, CLEAR_CACHE))
8342                 need_clear = 1;
8343
8344         while (1) {
8345                 ret = find_first_block_group(root, path, &key);
8346                 if (ret > 0)
8347                         break;
8348                 if (ret != 0)
8349                         goto error;
8350                 leaf = path->nodes[0];
8351                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8352                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8353                 if (!cache) {
8354                         ret = -ENOMEM;
8355                         goto error;
8356                 }
8357                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8358                                                 GFP_NOFS);
8359                 if (!cache->free_space_ctl) {
8360                         kfree(cache);
8361                         ret = -ENOMEM;
8362                         goto error;
8363                 }
8364
8365                 atomic_set(&cache->count, 1);
8366                 spin_lock_init(&cache->lock);
8367                 cache->fs_info = info;
8368                 INIT_LIST_HEAD(&cache->list);
8369                 INIT_LIST_HEAD(&cache->cluster_list);
8370
8371                 if (need_clear) {
8372                         /*
8373                          * When we mount with old space cache, we need to
8374                          * set BTRFS_DC_CLEAR and set dirty flag.
8375                          *
8376                          * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8377                          *    truncate the old free space cache inode and
8378                          *    setup a new one.
8379                          * b) Setting 'dirty flag' makes sure that we flush
8380                          *    the new space cache info onto disk.
8381                          */
8382                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8383                         if (btrfs_test_opt(root, SPACE_CACHE))
8384                                 cache->dirty = 1;
8385                 }
8386
8387                 read_extent_buffer(leaf, &cache->item,
8388                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8389                                    sizeof(cache->item));
8390                 memcpy(&cache->key, &found_key, sizeof(found_key));
8391
8392                 key.objectid = found_key.objectid + found_key.offset;
8393                 btrfs_release_path(path);
8394                 cache->flags = btrfs_block_group_flags(&cache->item);
8395                 cache->sectorsize = root->sectorsize;
8396                 cache->full_stripe_len = btrfs_full_stripe_len(root,
8397                                                &root->fs_info->mapping_tree,
8398                                                found_key.objectid);
8399                 btrfs_init_free_space_ctl(cache);
8400
8401                 /*
8402                  * We need to exclude the super stripes now so that the space
8403                  * info has super bytes accounted for, otherwise we'll think
8404                  * we have more space than we actually do.
8405                  */
8406                 ret = exclude_super_stripes(root, cache);
8407                 if (ret) {
8408                         /*
8409                          * We may have excluded something, so call this just in
8410                          * case.
8411                          */
8412                         free_excluded_extents(root, cache);
8413                         kfree(cache->free_space_ctl);
8414                         kfree(cache);
8415                         goto error;
8416                 }
8417
8418                 /*
8419                  * check for two cases, either we are full, and therefore
8420                  * don't need to bother with the caching work since we won't
8421                  * find any space, or we are empty, and we can just add all
8422                  * the space in and be done with it.  This saves us _alot_ of
8423                  * time, particularly in the full case.
8424                  */
8425                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8426                         cache->last_byte_to_unpin = (u64)-1;
8427                         cache->cached = BTRFS_CACHE_FINISHED;
8428                         free_excluded_extents(root, cache);
8429                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8430                         cache->last_byte_to_unpin = (u64)-1;
8431                         cache->cached = BTRFS_CACHE_FINISHED;
8432                         add_new_free_space(cache, root->fs_info,
8433                                            found_key.objectid,
8434                                            found_key.objectid +
8435                                            found_key.offset);
8436                         free_excluded_extents(root, cache);
8437                 }
8438
8439                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8440                 if (ret) {
8441                         btrfs_remove_free_space_cache(cache);
8442                         btrfs_put_block_group(cache);
8443                         goto error;
8444                 }
8445
8446                 ret = update_space_info(info, cache->flags, found_key.offset,
8447                                         btrfs_block_group_used(&cache->item),
8448                                         &space_info);
8449                 if (ret) {
8450                         btrfs_remove_free_space_cache(cache);
8451                         spin_lock(&info->block_group_cache_lock);
8452                         rb_erase(&cache->cache_node,
8453                                  &info->block_group_cache_tree);
8454                         spin_unlock(&info->block_group_cache_lock);
8455                         btrfs_put_block_group(cache);
8456                         goto error;
8457                 }
8458
8459                 cache->space_info = space_info;
8460                 spin_lock(&cache->space_info->lock);
8461                 cache->space_info->bytes_readonly += cache->bytes_super;
8462                 spin_unlock(&cache->space_info->lock);
8463
8464                 __link_block_group(space_info, cache);
8465
8466                 set_avail_alloc_bits(root->fs_info, cache->flags);
8467                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8468                         set_block_group_ro(cache, 1);
8469         }
8470
8471         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8472                 if (!(get_alloc_profile(root, space_info->flags) &
8473                       (BTRFS_BLOCK_GROUP_RAID10 |
8474                        BTRFS_BLOCK_GROUP_RAID1 |
8475                        BTRFS_BLOCK_GROUP_RAID5 |
8476                        BTRFS_BLOCK_GROUP_RAID6 |
8477                        BTRFS_BLOCK_GROUP_DUP)))
8478                         continue;
8479                 /*
8480                  * avoid allocating from un-mirrored block group if there are
8481                  * mirrored block groups.
8482                  */
8483                 list_for_each_entry(cache,
8484                                 &space_info->block_groups[BTRFS_RAID_RAID0],
8485                                 list)
8486                         set_block_group_ro(cache, 1);
8487                 list_for_each_entry(cache,
8488                                 &space_info->block_groups[BTRFS_RAID_SINGLE],
8489                                 list)
8490                         set_block_group_ro(cache, 1);
8491         }
8492
8493         init_global_block_rsv(info);
8494         ret = 0;
8495 error:
8496         btrfs_free_path(path);
8497         return ret;
8498 }
8499
8500 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8501                                        struct btrfs_root *root)
8502 {
8503         struct btrfs_block_group_cache *block_group, *tmp;
8504         struct btrfs_root *extent_root = root->fs_info->extent_root;
8505         struct btrfs_block_group_item item;
8506         struct btrfs_key key;
8507         int ret = 0;
8508
8509         list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8510                                  new_bg_list) {
8511                 list_del_init(&block_group->new_bg_list);
8512
8513                 if (ret)
8514                         continue;
8515
8516                 spin_lock(&block_group->lock);
8517                 memcpy(&item, &block_group->item, sizeof(item));
8518                 memcpy(&key, &block_group->key, sizeof(key));
8519                 spin_unlock(&block_group->lock);
8520
8521                 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8522                                         sizeof(item));
8523                 if (ret)
8524                         btrfs_abort_transaction(trans, extent_root, ret);
8525                 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8526                                                key.objectid, key.offset);
8527                 if (ret)
8528                         btrfs_abort_transaction(trans, extent_root, ret);
8529         }
8530 }
8531
8532 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8533                            struct btrfs_root *root, u64 bytes_used,
8534                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8535                            u64 size)
8536 {
8537         int ret;
8538         struct btrfs_root *extent_root;
8539         struct btrfs_block_group_cache *cache;
8540
8541         extent_root = root->fs_info->extent_root;
8542
8543         root->fs_info->last_trans_log_full_commit = trans->transid;
8544
8545         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8546         if (!cache)
8547                 return -ENOMEM;
8548         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8549                                         GFP_NOFS);
8550         if (!cache->free_space_ctl) {
8551                 kfree(cache);
8552                 return -ENOMEM;
8553         }
8554
8555         cache->key.objectid = chunk_offset;
8556         cache->key.offset = size;
8557         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8558         cache->sectorsize = root->sectorsize;
8559         cache->fs_info = root->fs_info;
8560         cache->full_stripe_len = btrfs_full_stripe_len(root,
8561                                                &root->fs_info->mapping_tree,
8562                                                chunk_offset);
8563
8564         atomic_set(&cache->count, 1);
8565         spin_lock_init(&cache->lock);
8566         INIT_LIST_HEAD(&cache->list);
8567         INIT_LIST_HEAD(&cache->cluster_list);
8568         INIT_LIST_HEAD(&cache->new_bg_list);
8569
8570         btrfs_init_free_space_ctl(cache);
8571
8572         btrfs_set_block_group_used(&cache->item, bytes_used);
8573         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8574         cache->flags = type;
8575         btrfs_set_block_group_flags(&cache->item, type);
8576
8577         cache->last_byte_to_unpin = (u64)-1;
8578         cache->cached = BTRFS_CACHE_FINISHED;
8579         ret = exclude_super_stripes(root, cache);
8580         if (ret) {
8581                 /*
8582                  * We may have excluded something, so call this just in
8583                  * case.
8584                  */
8585                 free_excluded_extents(root, cache);
8586                 kfree(cache->free_space_ctl);
8587                 kfree(cache);
8588                 return ret;
8589         }
8590
8591         add_new_free_space(cache, root->fs_info, chunk_offset,
8592                            chunk_offset + size);
8593
8594         free_excluded_extents(root, cache);
8595
8596         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8597         if (ret) {
8598                 btrfs_remove_free_space_cache(cache);
8599                 btrfs_put_block_group(cache);
8600                 return ret;
8601         }
8602
8603         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8604                                 &cache->space_info);
8605         if (ret) {
8606                 btrfs_remove_free_space_cache(cache);
8607                 spin_lock(&root->fs_info->block_group_cache_lock);
8608                 rb_erase(&cache->cache_node,
8609                          &root->fs_info->block_group_cache_tree);
8610                 spin_unlock(&root->fs_info->block_group_cache_lock);
8611                 btrfs_put_block_group(cache);
8612                 return ret;
8613         }
8614         update_global_block_rsv(root->fs_info);
8615
8616         spin_lock(&cache->space_info->lock);
8617         cache->space_info->bytes_readonly += cache->bytes_super;
8618         spin_unlock(&cache->space_info->lock);
8619
8620         __link_block_group(cache->space_info, cache);
8621
8622         list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8623
8624         set_avail_alloc_bits(extent_root->fs_info, type);
8625
8626         return 0;
8627 }
8628
8629 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8630 {
8631         u64 extra_flags = chunk_to_extended(flags) &
8632                                 BTRFS_EXTENDED_PROFILE_MASK;
8633
8634         write_seqlock(&fs_info->profiles_lock);
8635         if (flags & BTRFS_BLOCK_GROUP_DATA)
8636                 fs_info->avail_data_alloc_bits &= ~extra_flags;
8637         if (flags & BTRFS_BLOCK_GROUP_METADATA)
8638                 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8639         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8640                 fs_info->avail_system_alloc_bits &= ~extra_flags;
8641         write_sequnlock(&fs_info->profiles_lock);
8642 }
8643
8644 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8645                              struct btrfs_root *root, u64 group_start)
8646 {
8647         struct btrfs_path *path;
8648         struct btrfs_block_group_cache *block_group;
8649         struct btrfs_free_cluster *cluster;
8650         struct btrfs_root *tree_root = root->fs_info->tree_root;
8651         struct btrfs_key key;
8652         struct inode *inode;
8653         int ret;
8654         int index;
8655         int factor;
8656
8657         root = root->fs_info->extent_root;
8658
8659         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8660         BUG_ON(!block_group);
8661         BUG_ON(!block_group->ro);
8662
8663         /*
8664          * Free the reserved super bytes from this block group before
8665          * remove it.
8666          */
8667         free_excluded_extents(root, block_group);
8668
8669         memcpy(&key, &block_group->key, sizeof(key));
8670         index = get_block_group_index(block_group);
8671         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8672                                   BTRFS_BLOCK_GROUP_RAID1 |
8673                                   BTRFS_BLOCK_GROUP_RAID10))
8674                 factor = 2;
8675         else
8676                 factor = 1;
8677
8678         /* make sure this block group isn't part of an allocation cluster */
8679         cluster = &root->fs_info->data_alloc_cluster;
8680         spin_lock(&cluster->refill_lock);
8681         btrfs_return_cluster_to_free_space(block_group, cluster);
8682         spin_unlock(&cluster->refill_lock);
8683
8684         /*
8685          * make sure this block group isn't part of a metadata
8686          * allocation cluster
8687          */
8688         cluster = &root->fs_info->meta_alloc_cluster;
8689         spin_lock(&cluster->refill_lock);
8690         btrfs_return_cluster_to_free_space(block_group, cluster);
8691         spin_unlock(&cluster->refill_lock);
8692
8693         path = btrfs_alloc_path();
8694         if (!path) {
8695                 ret = -ENOMEM;
8696                 goto out;
8697         }
8698
8699         inode = lookup_free_space_inode(tree_root, block_group, path);
8700         if (!IS_ERR(inode)) {
8701                 ret = btrfs_orphan_add(trans, inode);
8702                 if (ret) {
8703                         btrfs_add_delayed_iput(inode);
8704                         goto out;
8705                 }
8706                 clear_nlink(inode);
8707                 /* One for the block groups ref */
8708                 spin_lock(&block_group->lock);
8709                 if (block_group->iref) {
8710                         block_group->iref = 0;
8711                         block_group->inode = NULL;
8712                         spin_unlock(&block_group->lock);
8713                         iput(inode);
8714                 } else {
8715                         spin_unlock(&block_group->lock);
8716                 }
8717                 /* One for our lookup ref */
8718                 btrfs_add_delayed_iput(inode);
8719         }
8720
8721         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8722         key.offset = block_group->key.objectid;
8723         key.type = 0;
8724
8725         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8726         if (ret < 0)
8727                 goto out;
8728         if (ret > 0)
8729                 btrfs_release_path(path);
8730         if (ret == 0) {
8731                 ret = btrfs_del_item(trans, tree_root, path);
8732                 if (ret)
8733                         goto out;
8734                 btrfs_release_path(path);
8735         }
8736
8737         spin_lock(&root->fs_info->block_group_cache_lock);
8738         rb_erase(&block_group->cache_node,
8739                  &root->fs_info->block_group_cache_tree);
8740
8741         if (root->fs_info->first_logical_byte == block_group->key.objectid)
8742                 root->fs_info->first_logical_byte = (u64)-1;
8743         spin_unlock(&root->fs_info->block_group_cache_lock);
8744
8745         down_write(&block_group->space_info->groups_sem);
8746         /*
8747          * we must use list_del_init so people can check to see if they
8748          * are still on the list after taking the semaphore
8749          */
8750         list_del_init(&block_group->list);
8751         if (list_empty(&block_group->space_info->block_groups[index]))
8752                 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8753         up_write(&block_group->space_info->groups_sem);
8754
8755         if (block_group->cached == BTRFS_CACHE_STARTED)
8756                 wait_block_group_cache_done(block_group);
8757
8758         btrfs_remove_free_space_cache(block_group);
8759
8760         spin_lock(&block_group->space_info->lock);
8761         block_group->space_info->total_bytes -= block_group->key.offset;
8762         block_group->space_info->bytes_readonly -= block_group->key.offset;
8763         block_group->space_info->disk_total -= block_group->key.offset * factor;
8764         spin_unlock(&block_group->space_info->lock);
8765
8766         memcpy(&key, &block_group->key, sizeof(key));
8767
8768         btrfs_clear_space_info_full(root->fs_info);
8769
8770         btrfs_put_block_group(block_group);
8771         btrfs_put_block_group(block_group);
8772
8773         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8774         if (ret > 0)
8775                 ret = -EIO;
8776         if (ret < 0)
8777                 goto out;
8778
8779         ret = btrfs_del_item(trans, root, path);
8780 out:
8781         btrfs_free_path(path);
8782         return ret;
8783 }
8784
8785 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8786 {
8787         struct btrfs_space_info *space_info;
8788         struct btrfs_super_block *disk_super;
8789         u64 features;
8790         u64 flags;
8791         int mixed = 0;
8792         int ret;
8793
8794         disk_super = fs_info->super_copy;
8795         if (!btrfs_super_root(disk_super))
8796                 return 1;
8797
8798         features = btrfs_super_incompat_flags(disk_super);
8799         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8800                 mixed = 1;
8801
8802         flags = BTRFS_BLOCK_GROUP_SYSTEM;
8803         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8804         if (ret)
8805                 goto out;
8806
8807         if (mixed) {
8808                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8809                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8810         } else {
8811                 flags = BTRFS_BLOCK_GROUP_METADATA;
8812                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8813                 if (ret)
8814                         goto out;
8815
8816                 flags = BTRFS_BLOCK_GROUP_DATA;
8817                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8818         }
8819 out:
8820         return ret;
8821 }
8822
8823 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8824 {
8825         return unpin_extent_range(root, start, end);
8826 }
8827
8828 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8829                                u64 num_bytes, u64 *actual_bytes)
8830 {
8831         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8832 }
8833
8834 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8835 {
8836         struct btrfs_fs_info *fs_info = root->fs_info;
8837         struct btrfs_block_group_cache *cache = NULL;
8838         u64 group_trimmed;
8839         u64 start;
8840         u64 end;
8841         u64 trimmed = 0;
8842         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8843         int ret = 0;
8844
8845         /*
8846          * try to trim all FS space, our block group may start from non-zero.
8847          */
8848         if (range->len == total_bytes)
8849                 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8850         else
8851                 cache = btrfs_lookup_block_group(fs_info, range->start);
8852
8853         while (cache) {
8854                 if (cache->key.objectid >= (range->start + range->len)) {
8855                         btrfs_put_block_group(cache);
8856                         break;
8857                 }
8858
8859                 start = max(range->start, cache->key.objectid);
8860                 end = min(range->start + range->len,
8861                                 cache->key.objectid + cache->key.offset);
8862
8863                 if (end - start >= range->minlen) {
8864                         if (!block_group_cache_done(cache)) {
8865                                 ret = cache_block_group(cache, 0);
8866                                 if (ret) {
8867                                         btrfs_put_block_group(cache);
8868                                         break;
8869                                 }
8870                                 ret = wait_block_group_cache_done(cache);
8871                                 if (ret) {
8872                                         btrfs_put_block_group(cache);
8873                                         break;
8874                                 }
8875                         }
8876                         ret = btrfs_trim_block_group(cache,
8877                                                      &group_trimmed,
8878                                                      start,
8879                                                      end,
8880                                                      range->minlen);
8881
8882                         trimmed += group_trimmed;
8883                         if (ret) {
8884                                 btrfs_put_block_group(cache);
8885                                 break;
8886                         }
8887                 }
8888
8889                 cache = next_block_group(fs_info->tree_root, cache);
8890         }
8891
8892         range->len = trimmed;
8893         return ret;
8894 }