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 #include "sysfs.h"
  39
  40 #undef SCRAMBLE_DELAYED_REFS
  41
  42 /*
  43  * control flags for do_chunk_alloc's force field
  44  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  45  * if we really need one.
  46  *
  47  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  48  * if we have very few chunks already allocated.  This is
  49  * used as part of the clustering code to help make sure
  50  * we have a good pool of storage to cluster in, without
  51  * filling the FS with empty chunks
  52  *
  53  * CHUNK_ALLOC_FORCE means it must try to allocate one
  54  *
  55  */
  56 enum {
  57         CHUNK_ALLOC_NO_FORCE = 0,
  58         CHUNK_ALLOC_LIMITED = 1,
  59         CHUNK_ALLOC_FORCE = 2,
  60 };
  61
  62 /*
  63  * Control how reservations are dealt with.
  64  *
  65  * RESERVE_FREE - freeing a reservation.
  66  * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
  67  *   ENOSPC accounting
  68  * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
  69  *   bytes_may_use as the ENOSPC accounting is done elsewhere
  70  */
  71 enum {
  72         RESERVE_FREE = 0,
  73         RESERVE_ALLOC = 1,
  74         RESERVE_ALLOC_NO_ACCOUNT = 2,
  75 };
  76
  77 static int update_block_group(struct btrfs_root *root,
  78                               u64 bytenr, u64 num_bytes, int alloc);
  79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  80                                 struct btrfs_root *root,
  81                                 u64 bytenr, u64 num_bytes, u64 parent,
  82                                 u64 root_objectid, u64 owner_objectid,
  83                                 u64 owner_offset, int refs_to_drop,
  84                                 struct btrfs_delayed_extent_op *extra_op);
  85 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  86                                     struct extent_buffer *leaf,
  87                                     struct btrfs_extent_item *ei);
  88 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  89                                       struct btrfs_root *root,
  90                                       u64 parent, u64 root_objectid,
  91                                       u64 flags, u64 owner, u64 offset,
  92                                       struct btrfs_key *ins, int ref_mod);
  93 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  94                                      struct btrfs_root *root,
  95                                      u64 parent, u64 root_objectid,
  96                                      u64 flags, struct btrfs_disk_key *key,
  97                                      int level, struct btrfs_key *ins);
  98 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  99                           struct btrfs_root *extent_root, u64 flags,
 100                           int force);
 101 static int find_next_key(struct btrfs_path *path, int level,
 102                          struct btrfs_key *key);
 103 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
 104                             int dump_block_groups);
 105 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
 106                                        u64 num_bytes, int reserve);
 107 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
 108                                u64 num_bytes);
 109 int btrfs_pin_extent(struct btrfs_root *root,
 110                      u64 bytenr, u64 num_bytes, int reserved);
 111
 112 static noinline int
 113 block_group_cache_done(struct btrfs_block_group_cache *cache)
 114 {
 115         smp_mb();
 116         return cache->cached == BTRFS_CACHE_FINISHED ||
 117                 cache->cached == BTRFS_CACHE_ERROR;
 118 }
 119
 120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 121 {
 122         return (cache->flags & bits) == bits;
 123 }
 124
 125 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 126 {
 127         atomic_inc(&cache->count);
 128 }
 129
 130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 131 {
 132         if (atomic_dec_and_test(&cache->count)) {
 133                 WARN_ON(cache->pinned > 0);
 134                 WARN_ON(cache->reserved > 0);
 135                 kfree(cache->free_space_ctl);
 136                 kfree(cache);
 137         }
 138 }
 139
 140 /*
 141  * this adds the block group to the fs_info rb tree for the block group
 142  * cache
 143  */
 144 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 145                                 struct btrfs_block_group_cache *block_group)
 146 {
 147         struct rb_node **p;
 148         struct rb_node *parent = NULL;
 149         struct btrfs_block_group_cache *cache;
 150
 151         spin_lock(&info->block_group_cache_lock);
 152         p = &info->block_group_cache_tree.rb_node;
 153
 154         while (*p) {
 155                 parent = *p;
 156                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 157                                  cache_node);
 158                 if (block_group->key.objectid < cache->key.objectid) {
 159                         p = &(*p)->rb_left;
 160                 } else if (block_group->key.objectid > cache->key.objectid) {
 161                         p = &(*p)->rb_right;
 162                 } else {
 163                         spin_unlock(&info->block_group_cache_lock);
 164                         return -EEXIST;
 165                 }
 166         }
 167
 168         rb_link_node(&block_group->cache_node, parent, p);
 169         rb_insert_color(&block_group->cache_node,
 170                         &info->block_group_cache_tree);
 171
 172         if (info->first_logical_byte > block_group->key.objectid)
 173                 info->first_logical_byte = block_group->key.objectid;
 174
 175         spin_unlock(&info->block_group_cache_lock);
 176
 177         return 0;
 178 }
 179
 180 /*
 181  * This will return the block group at or after bytenr if contains is 0, else
 182  * it will return the block group that contains the bytenr
 183  */
 184 static struct btrfs_block_group_cache *
 185 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 186                               int contains)
 187 {
 188         struct btrfs_block_group_cache *cache, *ret = NULL;
 189         struct rb_node *n;
 190         u64 end, start;
 191
 192         spin_lock(&info->block_group_cache_lock);
 193         n = info->block_group_cache_tree.rb_node;
 194
 195         while (n) {
 196                 cache = rb_entry(n, struct btrfs_block_group_cache,
 197                                  cache_node);
 198                 end = cache->key.objectid + cache->key.offset - 1;
 199                 start = cache->key.objectid;
 200
 201                 if (bytenr < start) {
 202                         if (!contains && (!ret || start < ret->key.objectid))
 203                                 ret = cache;
 204                         n = n->rb_left;
 205                 } else if (bytenr > start) {
 206                         if (contains && bytenr <= end) {
 207                                 ret = cache;
 208                                 break;
 209                         }
 210                         n = n->rb_right;
 211                 } else {
 212                         ret = cache;
 213                         break;
 214                 }
 215         }
 216         if (ret) {
 217                 btrfs_get_block_group(ret);
 218                 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
 219                         info->first_logical_byte = ret->key.objectid;
 220         }
 221         spin_unlock(&info->block_group_cache_lock);
 222
 223         return ret;
 224 }
 225
 226 static int add_excluded_extent(struct btrfs_root *root,
 227                                u64 start, u64 num_bytes)
 228 {
 229         u64 end = start + num_bytes - 1;
 230         set_extent_bits(&root->fs_info->freed_extents[0],
 231                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 232         set_extent_bits(&root->fs_info->freed_extents[1],
 233                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 234         return 0;
 235 }
 236
 237 static void free_excluded_extents(struct btrfs_root *root,
 238                                   struct btrfs_block_group_cache *cache)
 239 {
 240         u64 start, end;
 241
 242         start = cache->key.objectid;
 243         end = start + cache->key.offset - 1;
 244
 245         clear_extent_bits(&root->fs_info->freed_extents[0],
 246                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 247         clear_extent_bits(&root->fs_info->freed_extents[1],
 248                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 249 }
 250
 251 static int exclude_super_stripes(struct btrfs_root *root,
 252                                  struct btrfs_block_group_cache *cache)
 253 {
 254         u64 bytenr;
 255         u64 *logical;
 256         int stripe_len;
 257         int i, nr, ret;
 258
 259         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 260                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 261                 cache->bytes_super += stripe_len;
 262                 ret = add_excluded_extent(root, cache->key.objectid,
 263                                           stripe_len);
 264                 if (ret)
 265                         return ret;
 266         }
 267
 268         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 269                 bytenr = btrfs_sb_offset(i);
 270                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 271                                        cache->key.objectid, bytenr,
 272                                        0, &logical, &nr, &stripe_len);
 273                 if (ret)
 274                         return ret;
 275
 276                 while (nr--) {
 277                         u64 start, len;
 278
 279                         if (logical[nr] > cache->key.objectid +
 280                             cache->key.offset)
 281                                 continue;
 282
 283                         if (logical[nr] + stripe_len <= cache->key.objectid)
 284                                 continue;
 285
 286                         start = logical[nr];
 287                         if (start < cache->key.objectid) {
 288                                 start = cache->key.objectid;
 289                                 len = (logical[nr] + stripe_len) - start;
 290                         } else {
 291                                 len = min_t(u64, stripe_len,
 292                                             cache->key.objectid +
 293                                             cache->key.offset - start);
 294                         }
 295
 296                         cache->bytes_super += len;
 297                         ret = add_excluded_extent(root, start, len);
 298                         if (ret) {
 299                                 kfree(logical);
 300                                 return ret;
 301                         }
 302                 }
 303
 304                 kfree(logical);
 305         }
 306         return 0;
 307 }
 308
 309 static struct btrfs_caching_control *
 310 get_caching_control(struct btrfs_block_group_cache *cache)
 311 {
 312         struct btrfs_caching_control *ctl;
 313
 314         spin_lock(&cache->lock);
 315         if (cache->cached != BTRFS_CACHE_STARTED) {
 316                 spin_unlock(&cache->lock);
 317                 return NULL;
 318         }
 319
 320         /* We're loading it the fast way, so we don't have a caching_ctl. */
 321         if (!cache->caching_ctl) {
 322                 spin_unlock(&cache->lock);
 323                 return NULL;
 324         }
 325
 326         ctl = cache->caching_ctl;
 327         atomic_inc(&ctl->count);
 328         spin_unlock(&cache->lock);
 329         return ctl;
 330 }
 331
 332 static void put_caching_control(struct btrfs_caching_control *ctl)
 333 {
 334         if (atomic_dec_and_test(&ctl->count))
 335                 kfree(ctl);
 336 }
 337
 338 /*
 339  * this is only called by cache_block_group, since we could have freed extents
 340  * we need to check the pinned_extents for any extents that can't be used yet
 341  * since their free space will be released as soon as the transaction commits.
 342  */
 343 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 344                               struct btrfs_fs_info *info, u64 start, u64 end)
 345 {
 346         u64 extent_start, extent_end, size, total_added = 0;
 347         int ret;
 348
 349         while (start < end) {
 350                 ret = find_first_extent_bit(info->pinned_extents, start,
 351                                             &extent_start, &extent_end,
 352                                             EXTENT_DIRTY | EXTENT_UPTODATE,
 353                                             NULL);
 354                 if (ret)
 355                         break;
 356
 357                 if (extent_start <= start) {
 358                         start = extent_end + 1;
 359                 } else if (extent_start > start && extent_start < end) {
 360                         size = extent_start - start;
 361                         total_added += size;
 362                         ret = btrfs_add_free_space(block_group, start,
 363                                                    size);
 364                         BUG_ON(ret); /* -ENOMEM or logic error */
 365                         start = extent_end + 1;
 366                 } else {
 367                         break;
 368                 }
 369         }
 370
 371         if (start < end) {
 372                 size = end - start;
 373                 total_added += size;
 374                 ret = btrfs_add_free_space(block_group, start, size);
 375                 BUG_ON(ret); /* -ENOMEM or logic error */
 376         }
 377
 378         return total_added;
 379 }
 380
 381 static noinline void caching_thread(struct btrfs_work *work)
 382 {
 383         struct btrfs_block_group_cache *block_group;
 384         struct btrfs_fs_info *fs_info;
 385         struct btrfs_caching_control *caching_ctl;
 386         struct btrfs_root *extent_root;
 387         struct btrfs_path *path;
 388         struct extent_buffer *leaf;
 389         struct btrfs_key key;
 390         u64 total_found = 0;
 391         u64 last = 0;
 392         u32 nritems;
 393         int ret = -ENOMEM;
 394
 395         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 396         block_group = caching_ctl->block_group;
 397         fs_info = block_group->fs_info;
 398         extent_root = fs_info->extent_root;
 399
 400         path = btrfs_alloc_path();
 401         if (!path)
 402                 goto out;
 403
 404         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 405
 406         /*
 407          * We don't want to deadlock with somebody trying to allocate a new
 408          * extent for the extent root while also trying to search the extent
 409          * root to add free space.  So we skip locking and search the commit
 410          * root, since its read-only
 411          */
 412         path->skip_locking = 1;
 413         path->search_commit_root = 1;
 414         path->reada = 1;
 415
 416         key.objectid = last;
 417         key.offset = 0;
 418         key.type = BTRFS_EXTENT_ITEM_KEY;
 419 again:
 420         mutex_lock(&caching_ctl->mutex);
 421         /* need to make sure the commit_root doesn't disappear */
 422         down_read(&fs_info->extent_commit_sem);
 423
 424 next:
 425         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 426         if (ret < 0)
 427                 goto err;
 428
 429         leaf = path->nodes[0];
 430         nritems = btrfs_header_nritems(leaf);
 431
 432         while (1) {
 433                 if (btrfs_fs_closing(fs_info) > 1) {
 434                         last = (u64)-1;
 435                         break;
 436                 }
 437
 438                 if (path->slots[0] < nritems) {
 439                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 440                 } else {
 441                         ret = find_next_key(path, 0, &key);
 442                         if (ret)
 443                                 break;
 444
 445                         if (need_resched() ||
 446                             rwsem_is_contended(&fs_info->extent_commit_sem)) {
 447                                 caching_ctl->progress = last;
 448                                 btrfs_release_path(path);
 449                                 up_read(&fs_info->extent_commit_sem);
 450                                 mutex_unlock(&caching_ctl->mutex);
 451                                 cond_resched();
 452                                 goto again;
 453                         }
 454
 455                         ret = btrfs_next_leaf(extent_root, path);
 456                         if (ret < 0)
 457                                 goto err;
 458                         if (ret)
 459                                 break;
 460                         leaf = path->nodes[0];
 461                         nritems = btrfs_header_nritems(leaf);
 462                         continue;
 463                 }
 464
 465                 if (key.objectid < last) {
 466                         key.objectid = last;
 467                         key.offset = 0;
 468                         key.type = BTRFS_EXTENT_ITEM_KEY;
 469
 470                         caching_ctl->progress = last;
 471                         btrfs_release_path(path);
 472                         goto next;
 473                 }
 474
 475                 if (key.objectid < block_group->key.objectid) {
 476                         path->slots[0]++;
 477                         continue;
 478                 }
 479
 480                 if (key.objectid >= block_group->key.objectid +
 481                     block_group->key.offset)
 482                         break;
 483
 484                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
 485                     key.type == BTRFS_METADATA_ITEM_KEY) {
 486                         total_found += add_new_free_space(block_group,
 487                                                           fs_info, last,
 488                                                           key.objectid);
 489                         if (key.type == BTRFS_METADATA_ITEM_KEY)
 490                                 last = key.objectid +
 491                                         fs_info->tree_root->leafsize;
 492                         else
 493                                 last = key.objectid + key.offset;
 494
 495                         if (total_found > (1024 * 1024 * 2)) {
 496                                 total_found = 0;
 497                                 wake_up(&caching_ctl->wait);
 498                         }
 499                 }
 500                 path->slots[0]++;
 501         }
 502         ret = 0;
 503
 504         total_found += add_new_free_space(block_group, fs_info, last,
 505                                           block_group->key.objectid +
 506                                           block_group->key.offset);
 507         caching_ctl->progress = (u64)-1;
 508
 509         spin_lock(&block_group->lock);
 510         block_group->caching_ctl = NULL;
 511         block_group->cached = BTRFS_CACHE_FINISHED;
 512         spin_unlock(&block_group->lock);
 513
 514 err:
 515         btrfs_free_path(path);
 516         up_read(&fs_info->extent_commit_sem);
 517
 518         free_excluded_extents(extent_root, block_group);
 519
 520         mutex_unlock(&caching_ctl->mutex);
 521 out:
 522         if (ret) {
 523                 spin_lock(&block_group->lock);
 524                 block_group->caching_ctl = NULL;
 525                 block_group->cached = BTRFS_CACHE_ERROR;
 526                 spin_unlock(&block_group->lock);
 527         }
 528         wake_up(&caching_ctl->wait);
 529
 530         put_caching_control(caching_ctl);
 531         btrfs_put_block_group(block_group);
 532 }
 533
 534 static int cache_block_group(struct btrfs_block_group_cache *cache,
 535                              int load_cache_only)
 536 {
 537         DEFINE_WAIT(wait);
 538         struct btrfs_fs_info *fs_info = cache->fs_info;
 539         struct btrfs_caching_control *caching_ctl;
 540         int ret = 0;
 541
 542         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 543         if (!caching_ctl)
 544                 return -ENOMEM;
 545
 546         INIT_LIST_HEAD(&caching_ctl->list);
 547         mutex_init(&caching_ctl->mutex);
 548         init_waitqueue_head(&caching_ctl->wait);
 549         caching_ctl->block_group = cache;
 550         caching_ctl->progress = cache->key.objectid;
 551         atomic_set(&caching_ctl->count, 1);
 552         caching_ctl->work.func = caching_thread;
 553
 554         spin_lock(&cache->lock);
 555         /*
 556          * This should be a rare occasion, but this could happen I think in the
 557          * case where one thread starts to load the space cache info, and then
 558          * some other thread starts a transaction commit which tries to do an
 559          * allocation while the other thread is still loading the space cache
 560          * info.  The previous loop should have kept us from choosing this block
 561          * group, but if we've moved to the state where we will wait on caching
 562          * block groups we need to first check if we're doing a fast load here,
 563          * so we can wait for it to finish, otherwise we could end up allocating
 564          * from a block group who's cache gets evicted for one reason or
 565          * another.
 566          */
 567         while (cache->cached == BTRFS_CACHE_FAST) {
 568                 struct btrfs_caching_control *ctl;
 569
 570                 ctl = cache->caching_ctl;
 571                 atomic_inc(&ctl->count);
 572                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
 573                 spin_unlock(&cache->lock);
 574
 575                 schedule();
 576
 577                 finish_wait(&ctl->wait, &wait);
 578                 put_caching_control(ctl);
 579                 spin_lock(&cache->lock);
 580         }
 581
 582         if (cache->cached != BTRFS_CACHE_NO) {
 583                 spin_unlock(&cache->lock);
 584                 kfree(caching_ctl);
 585                 return 0;
 586         }
 587         WARN_ON(cache->caching_ctl);
 588         cache->caching_ctl = caching_ctl;
 589         cache->cached = BTRFS_CACHE_FAST;
 590         spin_unlock(&cache->lock);
 591
 592         if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
 593                 ret = load_free_space_cache(fs_info, cache);
 594
 595                 spin_lock(&cache->lock);
 596                 if (ret == 1) {
 597                         cache->caching_ctl = NULL;
 598                         cache->cached = BTRFS_CACHE_FINISHED;
 599                         cache->last_byte_to_unpin = (u64)-1;
 600                 } else {
 601                         if (load_cache_only) {
 602                                 cache->caching_ctl = NULL;
 603                                 cache->cached = BTRFS_CACHE_NO;
 604                         } else {
 605                                 cache->cached = BTRFS_CACHE_STARTED;
 606                         }
 607                 }
 608                 spin_unlock(&cache->lock);
 609                 wake_up(&caching_ctl->wait);
 610                 if (ret == 1) {
 611                         put_caching_control(caching_ctl);
 612                         free_excluded_extents(fs_info->extent_root, cache);
 613                         return 0;
 614                 }
 615         } else {
 616                 /*
 617                  * We are not going to do the fast caching, set cached to the
 618                  * appropriate value and wakeup any waiters.
 619                  */
 620                 spin_lock(&cache->lock);
 621                 if (load_cache_only) {
 622                         cache->caching_ctl = NULL;
 623                         cache->cached = BTRFS_CACHE_NO;
 624                 } else {
 625                         cache->cached = BTRFS_CACHE_STARTED;
 626                 }
 627                 spin_unlock(&cache->lock);
 628                 wake_up(&caching_ctl->wait);
 629         }
 630
 631         if (load_cache_only) {
 632                 put_caching_control(caching_ctl);
 633                 return 0;
 634         }
 635
 636         down_write(&fs_info->extent_commit_sem);
 637         atomic_inc(&caching_ctl->count);
 638         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 639         up_write(&fs_info->extent_commit_sem);
 640
 641         btrfs_get_block_group(cache);
 642
 643         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 644
 645         return ret;
 646 }
 647
 648 /*
 649  * return the block group that starts at or after bytenr
 650  */
 651 static struct btrfs_block_group_cache *
 652 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 653 {
 654         struct btrfs_block_group_cache *cache;
 655
 656         cache = block_group_cache_tree_search(info, bytenr, 0);
 657
 658         return cache;
 659 }
 660
 661 /*
 662  * return the block group that contains the given bytenr
 663  */
 664 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 665                                                  struct btrfs_fs_info *info,
 666                                                  u64 bytenr)
 667 {
 668         struct btrfs_block_group_cache *cache;
 669
 670         cache = block_group_cache_tree_search(info, bytenr, 1);
 671
 672         return cache;
 673 }
 674
 675 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 676                                                   u64 flags)
 677 {
 678         struct list_head *head = &info->space_info;
 679         struct btrfs_space_info *found;
 680
 681         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 682
 683         rcu_read_lock();
 684         list_for_each_entry_rcu(found, head, list) {
 685                 if (found->flags & flags) {
 686                         rcu_read_unlock();
 687                         return found;
 688                 }
 689         }
 690         rcu_read_unlock();
 691         return NULL;
 692 }
 693
 694 /*
 695  * after adding space to the filesystem, we need to clear the full flags
 696  * on all the space infos.
 697  */
 698 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 699 {
 700         struct list_head *head = &info->space_info;
 701         struct btrfs_space_info *found;
 702
 703         rcu_read_lock();
 704         list_for_each_entry_rcu(found, head, list)
 705                 found->full = 0;
 706         rcu_read_unlock();
 707 }
 708
 709 /* simple helper to search for an existing extent at a given offset */
 710 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 711 {
 712         int ret;
 713         struct btrfs_key key;
 714         struct btrfs_path *path;
 715
 716         path = btrfs_alloc_path();
 717         if (!path)
 718                 return -ENOMEM;
 719
 720         key.objectid = start;
 721         key.offset = len;
 722         key.type = BTRFS_EXTENT_ITEM_KEY;
 723         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 724                                 0, 0);
 725         if (ret > 0) {
 726                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 727                 if (key.objectid == start &&
 728                     key.type == BTRFS_METADATA_ITEM_KEY)
 729                         ret = 0;
 730         }
 731         btrfs_free_path(path);
 732         return ret;
 733 }
 734
 735 /*
 736  * helper function to lookup reference count and flags of a tree block.
 737  *
 738  * the head node for delayed ref is used to store the sum of all the
 739  * reference count modifications queued up in the rbtree. the head
 740  * node may also store the extent flags to set. This way you can check
 741  * to see what the reference count and extent flags would be if all of
 742  * the delayed refs are not processed.
 743  */
 744 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 745                              struct btrfs_root *root, u64 bytenr,
 746                              u64 offset, int metadata, u64 *refs, u64 *flags)
 747 {
 748         struct btrfs_delayed_ref_head *head;
 749         struct btrfs_delayed_ref_root *delayed_refs;
 750         struct btrfs_path *path;
 751         struct btrfs_extent_item *ei;
 752         struct extent_buffer *leaf;
 753         struct btrfs_key key;
 754         u32 item_size;
 755         u64 num_refs;
 756         u64 extent_flags;
 757         int ret;
 758
 759         /*
 760          * If we don't have skinny metadata, don't bother doing anything
 761          * different
 762          */
 763         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
 764                 offset = root->leafsize;
 765                 metadata = 0;
 766         }
 767
 768         path = btrfs_alloc_path();
 769         if (!path)
 770                 return -ENOMEM;
 771
 772         if (!trans) {
 773                 path->skip_locking = 1;
 774                 path->search_commit_root = 1;
 775         }
 776
 777 search_again:
 778         key.objectid = bytenr;
 779         key.offset = offset;
 780         if (metadata)
 781                 key.type = BTRFS_METADATA_ITEM_KEY;
 782         else
 783                 key.type = BTRFS_EXTENT_ITEM_KEY;
 784
 785 again:
 786         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 787                                 &key, path, 0, 0);
 788         if (ret < 0)
 789                 goto out_free;
 790
 791         if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
 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 search_again;
 859                 }
 860                 spin_lock(&head->lock);
 861                 if (head->extent_op && head->extent_op->update_flags)
 862                         extent_flags |= head->extent_op->flags_to_set;
 863                 else
 864                         BUG_ON(num_refs == 0);
 865
 866                 num_refs += head->node.ref_mod;
 867                 spin_unlock(&head->lock);
 868                 mutex_unlock(&head->mutex);
 869         }
 870         spin_unlock(&delayed_refs->lock);
 871 out:
 872         WARN_ON(num_refs == 0);
 873         if (refs)
 874                 *refs = num_refs;
 875         if (flags)
 876                 *flags = extent_flags;
 877 out_free:
 878         btrfs_free_path(path);
 879         return ret;
 880 }
 881
 882 /*
 883  * Back reference rules.  Back refs have three main goals:
 884  *
 885  * 1) differentiate between all holders of references to an extent so that
 886  *    when a reference is dropped we can make sure it was a valid reference
 887  *    before freeing the extent.
 888  *
 889  * 2) Provide enough information to quickly find the holders of an extent
 890  *    if we notice a given block is corrupted or bad.
 891  *
 892  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 893  *    maintenance.  This is actually the same as #2, but with a slightly
 894  *    different use case.
 895  *
 896  * There are two kinds of back refs. The implicit back refs is optimized
 897  * for pointers in non-shared tree blocks. For a given pointer in a block,
 898  * back refs of this kind provide information about the block's owner tree
 899  * and the pointer's key. These information allow us to find the block by
 900  * b-tree searching. The full back refs is for pointers in tree blocks not
 901  * referenced by their owner trees. The location of tree block is recorded
 902  * in the back refs. Actually the full back refs is generic, and can be
 903  * used in all cases the implicit back refs is used. The major shortcoming
 904  * of the full back refs is its overhead. Every time a tree block gets
 905  * COWed, we have to update back refs entry for all pointers in it.
 906  *
 907  * For a newly allocated tree block, we use implicit back refs for
 908  * pointers in it. This means most tree related operations only involve
 909  * implicit back refs. For a tree block created in old transaction, the
 910  * only way to drop a reference to it is COW it. So we can detect the
 911  * event that tree block loses its owner tree's reference and do the
 912  * back refs conversion.
 913  *
 914  * When a tree block is COW'd through a tree, there are four cases:
 915  *
 916  * The reference count of the block is one and the tree is the block's
 917  * owner tree. Nothing to do in this case.
 918  *
 919  * The reference count of the block is one and the tree is not the
 920  * block's owner tree. In this case, full back refs is used for pointers
 921  * in the block. Remove these full back refs, add implicit back refs for
 922  * every pointers in the new block.
 923  *
 924  * The reference count of the block is greater than one and the tree is
 925  * the block's owner tree. In this case, implicit back refs is used for
 926  * pointers in the block. Add full back refs for every pointers in the
 927  * block, increase lower level extents' reference counts. The original
 928  * implicit back refs are entailed to the new block.
 929  *
 930  * The reference count of the block is greater than one and the tree is
 931  * not the block's owner tree. Add implicit back refs for every pointer in
 932  * the new block, increase lower level extents' reference count.
 933  *
 934  * Back Reference Key composing:
 935  *
 936  * The key objectid corresponds to the first byte in the extent,
 937  * The key type is used to differentiate between types of back refs.
 938  * There are different meanings of the key offset for different types
 939  * of back refs.
 940  *
 941  * File extents can be referenced by:
 942  *
 943  * - multiple snapshots, subvolumes, or different generations in one subvol
 944  * - different files inside a single subvolume
 945  * - different offsets inside a file (bookend extents in file.c)
 946  *
 947  * The extent ref structure for the implicit back refs has fields for:
 948  *
 949  * - Objectid of the subvolume root
 950  * - objectid of the file holding the reference
 951  * - original offset in the file
 952  * - how many bookend extents
 953  *
 954  * The key offset for the implicit back refs is hash of the first
 955  * three fields.
 956  *
 957  * The extent ref structure for the full back refs has field for:
 958  *
 959  * - number of pointers in the tree leaf
 960  *
 961  * The key offset for the implicit back refs is the first byte of
 962  * the tree leaf
 963  *
 964  * When a file extent is allocated, The implicit back refs is used.
 965  * the fields are filled in:
 966  *
 967  *     (root_key.objectid, inode objectid, offset in file, 1)
 968  *
 969  * When a file extent is removed file truncation, we find the
 970  * corresponding implicit back refs and check the following fields:
 971  *
 972  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 973  *
 974  * Btree extents can be referenced by:
 975  *
 976  * - Different subvolumes
 977  *
 978  * Both the implicit back refs and the full back refs for tree blocks
 979  * only consist of key. The key offset for the implicit back refs is
 980  * objectid of block's owner tree. The key offset for the full back refs
 981  * is the first byte of parent block.
 982  *
 983  * When implicit back refs is used, information about the lowest key and
 984  * level of the tree block are required. These information are stored in
 985  * tree block info structure.
 986  */
 987
 988 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 989 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 990                                   struct btrfs_root *root,
 991                                   struct btrfs_path *path,
 992                                   u64 owner, u32 extra_size)
 993 {
 994         struct btrfs_extent_item *item;
 995         struct btrfs_extent_item_v0 *ei0;
 996         struct btrfs_extent_ref_v0 *ref0;
 997         struct btrfs_tree_block_info *bi;
 998         struct extent_buffer *leaf;
 999         struct btrfs_key key;
1000         struct btrfs_key found_key;
1001         u32 new_size = sizeof(*item);
1002         u64 refs;
1003         int ret;
1004
1005         leaf = path->nodes[0];
1006         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1007
1008         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1009         ei0 = btrfs_item_ptr(leaf, path->slots[0],
1010                              struct btrfs_extent_item_v0);
1011         refs = btrfs_extent_refs_v0(leaf, ei0);
1012
1013         if (owner == (u64)-1) {
1014                 while (1) {
1015                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1016                                 ret = btrfs_next_leaf(root, path);
1017                                 if (ret < 0)
1018                                         return ret;
1019                                 BUG_ON(ret > 0); /* Corruption */
1020                                 leaf = path->nodes[0];
1021                         }
1022                         btrfs_item_key_to_cpu(leaf, &found_key,
1023                                               path->slots[0]);
1024                         BUG_ON(key.objectid != found_key.objectid);
1025                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1026                                 path->slots[0]++;
1027                                 continue;
1028                         }
1029                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1030                                               struct btrfs_extent_ref_v0);
1031                         owner = btrfs_ref_objectid_v0(leaf, ref0);
1032                         break;
1033                 }
1034         }
1035         btrfs_release_path(path);
1036
1037         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1038                 new_size += sizeof(*bi);
1039
1040         new_size -= sizeof(*ei0);
1041         ret = btrfs_search_slot(trans, root, &key, path,
1042                                 new_size + extra_size, 1);
1043         if (ret < 0)
1044                 return ret;
1045         BUG_ON(ret); /* Corruption */
1046
1047         btrfs_extend_item(root, path, new_size);
1048
1049         leaf = path->nodes[0];
1050         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1051         btrfs_set_extent_refs(leaf, item, refs);
1052         /* FIXME: get real generation */
1053         btrfs_set_extent_generation(leaf, item, 0);
1054         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1055                 btrfs_set_extent_flags(leaf, item,
1056                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1057                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1058                 bi = (struct btrfs_tree_block_info *)(item + 1);
1059                 /* FIXME: get first key of the block */
1060                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1061                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1062         } else {
1063                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1064         }
1065         btrfs_mark_buffer_dirty(leaf);
1066         return 0;
1067 }
1068 #endif
1069
1070 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1071 {
1072         u32 high_crc = ~(u32)0;
1073         u32 low_crc = ~(u32)0;
1074         __le64 lenum;
1075
1076         lenum = cpu_to_le64(root_objectid);
1077         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1078         lenum = cpu_to_le64(owner);
1079         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1080         lenum = cpu_to_le64(offset);
1081         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1082
1083         return ((u64)high_crc << 31) ^ (u64)low_crc;
1084 }
1085
1086 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1087                                      struct btrfs_extent_data_ref *ref)
1088 {
1089         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1090                                     btrfs_extent_data_ref_objectid(leaf, ref),
1091                                     btrfs_extent_data_ref_offset(leaf, ref));
1092 }
1093
1094 static int match_extent_data_ref(struct extent_buffer *leaf,
1095                                  struct btrfs_extent_data_ref *ref,
1096                                  u64 root_objectid, u64 owner, u64 offset)
1097 {
1098         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1099             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1100             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1101                 return 0;
1102         return 1;
1103 }
1104
1105 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1106                                            struct btrfs_root *root,
1107                                            struct btrfs_path *path,
1108                                            u64 bytenr, u64 parent,
1109                                            u64 root_objectid,
1110                                            u64 owner, u64 offset)
1111 {
1112         struct btrfs_key key;
1113         struct btrfs_extent_data_ref *ref;
1114         struct extent_buffer *leaf;
1115         u32 nritems;
1116         int ret;
1117         int recow;
1118         int err = -ENOENT;
1119
1120         key.objectid = bytenr;
1121         if (parent) {
1122                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1123                 key.offset = parent;
1124         } else {
1125                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1126                 key.offset = hash_extent_data_ref(root_objectid,
1127                                                   owner, offset);
1128         }
1129 again:
1130         recow = 0;
1131         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1132         if (ret < 0) {
1133                 err = ret;
1134                 goto fail;
1135         }
1136
1137         if (parent) {
1138                 if (!ret)
1139                         return 0;
1140 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1141                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1142                 btrfs_release_path(path);
1143                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1144                 if (ret < 0) {
1145                         err = ret;
1146                         goto fail;
1147                 }
1148                 if (!ret)
1149                         return 0;
1150 #endif
1151                 goto fail;
1152         }
1153
1154         leaf = path->nodes[0];
1155         nritems = btrfs_header_nritems(leaf);
1156         while (1) {
1157                 if (path->slots[0] >= nritems) {
1158                         ret = btrfs_next_leaf(root, path);
1159                         if (ret < 0)
1160                                 err = ret;
1161                         if (ret)
1162                                 goto fail;
1163
1164                         leaf = path->nodes[0];
1165                         nritems = btrfs_header_nritems(leaf);
1166                         recow = 1;
1167                 }
1168
1169                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1170                 if (key.objectid != bytenr ||
1171                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1172                         goto fail;
1173
1174                 ref = btrfs_item_ptr(leaf, path->slots[0],
1175                                      struct btrfs_extent_data_ref);
1176
1177                 if (match_extent_data_ref(leaf, ref, root_objectid,
1178                                           owner, offset)) {
1179                         if (recow) {
1180                                 btrfs_release_path(path);
1181                                 goto again;
1182                         }
1183                         err = 0;
1184                         break;
1185                 }
1186                 path->slots[0]++;
1187         }
1188 fail:
1189         return err;
1190 }
1191
1192 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1193                                            struct btrfs_root *root,
1194                                            struct btrfs_path *path,
1195                                            u64 bytenr, u64 parent,
1196                                            u64 root_objectid, u64 owner,
1197                                            u64 offset, int refs_to_add)
1198 {
1199         struct btrfs_key key;
1200         struct extent_buffer *leaf;
1201         u32 size;
1202         u32 num_refs;
1203         int ret;
1204
1205         key.objectid = bytenr;
1206         if (parent) {
1207                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1208                 key.offset = parent;
1209                 size = sizeof(struct btrfs_shared_data_ref);
1210         } else {
1211                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1212                 key.offset = hash_extent_data_ref(root_objectid,
1213                                                   owner, offset);
1214                 size = sizeof(struct btrfs_extent_data_ref);
1215         }
1216
1217         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1218         if (ret && ret != -EEXIST)
1219                 goto fail;
1220
1221         leaf = path->nodes[0];
1222         if (parent) {
1223                 struct btrfs_shared_data_ref *ref;
1224                 ref = btrfs_item_ptr(leaf, path->slots[0],
1225                                      struct btrfs_shared_data_ref);
1226                 if (ret == 0) {
1227                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1228                 } else {
1229                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1230                         num_refs += refs_to_add;
1231                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1232                 }
1233         } else {
1234                 struct btrfs_extent_data_ref *ref;
1235                 while (ret == -EEXIST) {
1236                         ref = btrfs_item_ptr(leaf, path->slots[0],
1237                                              struct btrfs_extent_data_ref);
1238                         if (match_extent_data_ref(leaf, ref, root_objectid,
1239                                                   owner, offset))
1240                                 break;
1241                         btrfs_release_path(path);
1242                         key.offset++;
1243                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1244                                                       size);
1245                         if (ret && ret != -EEXIST)
1246                                 goto fail;
1247
1248                         leaf = path->nodes[0];
1249                 }
1250                 ref = btrfs_item_ptr(leaf, path->slots[0],
1251                                      struct btrfs_extent_data_ref);
1252                 if (ret == 0) {
1253                         btrfs_set_extent_data_ref_root(leaf, ref,
1254                                                        root_objectid);
1255                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1256                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1257                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1258                 } else {
1259                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1260                         num_refs += refs_to_add;
1261                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1262                 }
1263         }
1264         btrfs_mark_buffer_dirty(leaf);
1265         ret = 0;
1266 fail:
1267         btrfs_release_path(path);
1268         return ret;
1269 }
1270
1271 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1272                                            struct btrfs_root *root,
1273                                            struct btrfs_path *path,
1274                                            int refs_to_drop)
1275 {
1276         struct btrfs_key key;
1277         struct btrfs_extent_data_ref *ref1 = NULL;
1278         struct btrfs_shared_data_ref *ref2 = NULL;
1279         struct extent_buffer *leaf;
1280         u32 num_refs = 0;
1281         int ret = 0;
1282
1283         leaf = path->nodes[0];
1284         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1285
1286         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1287                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1288                                       struct btrfs_extent_data_ref);
1289                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1290         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1291                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1292                                       struct btrfs_shared_data_ref);
1293                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1296                 struct btrfs_extent_ref_v0 *ref0;
1297                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1298                                       struct btrfs_extent_ref_v0);
1299                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1300 #endif
1301         } else {
1302                 BUG();
1303         }
1304
1305         BUG_ON(num_refs < refs_to_drop);
1306         num_refs -= refs_to_drop;
1307
1308         if (num_refs == 0) {
1309                 ret = btrfs_del_item(trans, root, path);
1310         } else {
1311                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1312                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1313                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1314                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1316                 else {
1317                         struct btrfs_extent_ref_v0 *ref0;
1318                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1319                                         struct btrfs_extent_ref_v0);
1320                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1321                 }
1322 #endif
1323                 btrfs_mark_buffer_dirty(leaf);
1324         }
1325         return ret;
1326 }
1327
1328 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1329                                           struct btrfs_path *path,
1330                                           struct btrfs_extent_inline_ref *iref)
1331 {
1332         struct btrfs_key key;
1333         struct extent_buffer *leaf;
1334         struct btrfs_extent_data_ref *ref1;
1335         struct btrfs_shared_data_ref *ref2;
1336         u32 num_refs = 0;
1337
1338         leaf = path->nodes[0];
1339         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1340         if (iref) {
1341                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1342                     BTRFS_EXTENT_DATA_REF_KEY) {
1343                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1344                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1345                 } else {
1346                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1347                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348                 }
1349         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1350                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1351                                       struct btrfs_extent_data_ref);
1352                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1353         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1354                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1355                                       struct btrfs_shared_data_ref);
1356                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1359                 struct btrfs_extent_ref_v0 *ref0;
1360                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1361                                       struct btrfs_extent_ref_v0);
1362                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1363 #endif
1364         } else {
1365                 WARN_ON(1);
1366         }
1367         return num_refs;
1368 }
1369
1370 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1371                                           struct btrfs_root *root,
1372                                           struct btrfs_path *path,
1373                                           u64 bytenr, u64 parent,
1374                                           u64 root_objectid)
1375 {
1376         struct btrfs_key key;
1377         int ret;
1378
1379         key.objectid = bytenr;
1380         if (parent) {
1381                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382                 key.offset = parent;
1383         } else {
1384                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385                 key.offset = root_objectid;
1386         }
1387
1388         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1389         if (ret > 0)
1390                 ret = -ENOENT;
1391 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1392         if (ret == -ENOENT && parent) {
1393                 btrfs_release_path(path);
1394                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1395                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1396                 if (ret > 0)
1397                         ret = -ENOENT;
1398         }
1399 #endif
1400         return ret;
1401 }
1402
1403 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1404                                           struct btrfs_root *root,
1405                                           struct btrfs_path *path,
1406                                           u64 bytenr, u64 parent,
1407                                           u64 root_objectid)
1408 {
1409         struct btrfs_key key;
1410         int ret;
1411
1412         key.objectid = bytenr;
1413         if (parent) {
1414                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1415                 key.offset = parent;
1416         } else {
1417                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1418                 key.offset = root_objectid;
1419         }
1420
1421         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1422         btrfs_release_path(path);
1423         return ret;
1424 }
1425
1426 static inline int extent_ref_type(u64 parent, u64 owner)
1427 {
1428         int type;
1429         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1430                 if (parent > 0)
1431                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1432                 else
1433                         type = BTRFS_TREE_BLOCK_REF_KEY;
1434         } else {
1435                 if (parent > 0)
1436                         type = BTRFS_SHARED_DATA_REF_KEY;
1437                 else
1438                         type = BTRFS_EXTENT_DATA_REF_KEY;
1439         }
1440         return type;
1441 }
1442
1443 static int find_next_key(struct btrfs_path *path, int level,
1444                          struct btrfs_key *key)
1445
1446 {
1447         for (; level < BTRFS_MAX_LEVEL; level++) {
1448                 if (!path->nodes[level])
1449                         break;
1450                 if (path->slots[level] + 1 >=
1451                     btrfs_header_nritems(path->nodes[level]))
1452                         continue;
1453                 if (level == 0)
1454                         btrfs_item_key_to_cpu(path->nodes[level], key,
1455                                               path->slots[level] + 1);
1456                 else
1457                         btrfs_node_key_to_cpu(path->nodes[level], key,
1458                                               path->slots[level] + 1);
1459                 return 0;
1460         }
1461         return 1;
1462 }
1463
1464 /*
1465  * look for inline back ref. if back ref is found, *ref_ret is set
1466  * to the address of inline back ref, and 0 is returned.
1467  *
1468  * if back ref isn't found, *ref_ret is set to the address where it
1469  * should be inserted, and -ENOENT is returned.
1470  *
1471  * if insert is true and there are too many inline back refs, the path
1472  * points to the extent item, and -EAGAIN is returned.
1473  *
1474  * NOTE: inline back refs are ordered in the same way that back ref
1475  *       items in the tree are ordered.
1476  */
1477 static noinline_for_stack
1478 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1479                                  struct btrfs_root *root,
1480                                  struct btrfs_path *path,
1481                                  struct btrfs_extent_inline_ref **ref_ret,
1482                                  u64 bytenr, u64 num_bytes,
1483                                  u64 parent, u64 root_objectid,
1484                                  u64 owner, u64 offset, int insert)
1485 {
1486         struct btrfs_key key;
1487         struct extent_buffer *leaf;
1488         struct btrfs_extent_item *ei;
1489         struct btrfs_extent_inline_ref *iref;
1490         u64 flags;
1491         u64 item_size;
1492         unsigned long ptr;
1493         unsigned long end;
1494         int extra_size;
1495         int type;
1496         int want;
1497         int ret;
1498         int err = 0;
1499         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1500                                                  SKINNY_METADATA);
1501
1502         key.objectid = bytenr;
1503         key.type = BTRFS_EXTENT_ITEM_KEY;
1504         key.offset = num_bytes;
1505
1506         want = extent_ref_type(parent, owner);
1507         if (insert) {
1508                 extra_size = btrfs_extent_inline_ref_size(want);
1509                 path->keep_locks = 1;
1510         } else
1511                 extra_size = -1;
1512
1513         /*
1514          * Owner is our parent level, so we can just add one to get the level
1515          * for the block we are interested in.
1516          */
1517         if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1518                 key.type = BTRFS_METADATA_ITEM_KEY;
1519                 key.offset = owner;
1520         }
1521
1522 again:
1523         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1524         if (ret < 0) {
1525                 err = ret;
1526                 goto out;
1527         }
1528
1529         /*
1530          * We may be a newly converted file system which still has the old fat
1531          * extent entries for metadata, so try and see if we have one of those.
1532          */
1533         if (ret > 0 && skinny_metadata) {
1534                 skinny_metadata = false;
1535                 if (path->slots[0]) {
1536                         path->slots[0]--;
1537                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1538                                               path->slots[0]);
1539                         if (key.objectid == bytenr &&
1540                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1541                             key.offset == num_bytes)
1542                                 ret = 0;
1543                 }
1544                 if (ret) {
1545                         key.type = BTRFS_EXTENT_ITEM_KEY;
1546                         key.offset = num_bytes;
1547                         btrfs_release_path(path);
1548                         goto again;
1549                 }
1550         }
1551
1552         if (ret && !insert) {
1553                 err = -ENOENT;
1554                 goto out;
1555         } else if (WARN_ON(ret)) {
1556                 err = -EIO;
1557                 goto out;
1558         }
1559
1560         leaf = path->nodes[0];
1561         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1562 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1563         if (item_size < sizeof(*ei)) {
1564                 if (!insert) {
1565                         err = -ENOENT;
1566                         goto out;
1567                 }
1568                 ret = convert_extent_item_v0(trans, root, path, owner,
1569                                              extra_size);
1570                 if (ret < 0) {
1571                         err = ret;
1572                         goto out;
1573                 }
1574                 leaf = path->nodes[0];
1575                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1576         }
1577 #endif
1578         BUG_ON(item_size < sizeof(*ei));
1579
1580         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1581         flags = btrfs_extent_flags(leaf, ei);
1582
1583         ptr = (unsigned long)(ei + 1);
1584         end = (unsigned long)ei + item_size;
1585
1586         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1587                 ptr += sizeof(struct btrfs_tree_block_info);
1588                 BUG_ON(ptr > end);
1589         }
1590
1591         err = -ENOENT;
1592         while (1) {
1593                 if (ptr >= end) {
1594                         WARN_ON(ptr > end);
1595                         break;
1596                 }
1597                 iref = (struct btrfs_extent_inline_ref *)ptr;
1598                 type = btrfs_extent_inline_ref_type(leaf, iref);
1599                 if (want < type)
1600                         break;
1601                 if (want > type) {
1602                         ptr += btrfs_extent_inline_ref_size(type);
1603                         continue;
1604                 }
1605
1606                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1607                         struct btrfs_extent_data_ref *dref;
1608                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1609                         if (match_extent_data_ref(leaf, dref, root_objectid,
1610                                                   owner, offset)) {
1611                                 err = 0;
1612                                 break;
1613                         }
1614                         if (hash_extent_data_ref_item(leaf, dref) <
1615                             hash_extent_data_ref(root_objectid, owner, offset))
1616                                 break;
1617                 } else {
1618                         u64 ref_offset;
1619                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1620                         if (parent > 0) {
1621                                 if (parent == ref_offset) {
1622                                         err = 0;
1623                                         break;
1624                                 }
1625                                 if (ref_offset < parent)
1626                                         break;
1627                         } else {
1628                                 if (root_objectid == ref_offset) {
1629                                         err = 0;
1630                                         break;
1631                                 }
1632                                 if (ref_offset < root_objectid)
1633                                         break;
1634                         }
1635                 }
1636                 ptr += btrfs_extent_inline_ref_size(type);
1637         }
1638         if (err == -ENOENT && insert) {
1639                 if (item_size + extra_size >=
1640                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1641                         err = -EAGAIN;
1642                         goto out;
1643                 }
1644                 /*
1645                  * To add new inline back ref, we have to make sure
1646                  * there is no corresponding back ref item.
1647                  * For simplicity, we just do not add new inline back
1648                  * ref if there is any kind of item for this block
1649                  */
1650                 if (find_next_key(path, 0, &key) == 0 &&
1651                     key.objectid == bytenr &&
1652                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1653                         err = -EAGAIN;
1654                         goto out;
1655                 }
1656         }
1657         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1658 out:
1659         if (insert) {
1660                 path->keep_locks = 0;
1661                 btrfs_unlock_up_safe(path, 1);
1662         }
1663         return err;
1664 }
1665
1666 /*
1667  * helper to add new inline back ref
1668  */
1669 static noinline_for_stack
1670 void setup_inline_extent_backref(struct btrfs_root *root,
1671                                  struct btrfs_path *path,
1672                                  struct btrfs_extent_inline_ref *iref,
1673                                  u64 parent, u64 root_objectid,
1674                                  u64 owner, u64 offset, int refs_to_add,
1675                                  struct btrfs_delayed_extent_op *extent_op)
1676 {
1677         struct extent_buffer *leaf;
1678         struct btrfs_extent_item *ei;
1679         unsigned long ptr;
1680         unsigned long end;
1681         unsigned long item_offset;
1682         u64 refs;
1683         int size;
1684         int type;
1685
1686         leaf = path->nodes[0];
1687         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688         item_offset = (unsigned long)iref - (unsigned long)ei;
1689
1690         type = extent_ref_type(parent, owner);
1691         size = btrfs_extent_inline_ref_size(type);
1692
1693         btrfs_extend_item(root, path, size);
1694
1695         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1696         refs = btrfs_extent_refs(leaf, ei);
1697         refs += refs_to_add;
1698         btrfs_set_extent_refs(leaf, ei, refs);
1699         if (extent_op)
1700                 __run_delayed_extent_op(extent_op, leaf, ei);
1701
1702         ptr = (unsigned long)ei + item_offset;
1703         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1704         if (ptr < end - size)
1705                 memmove_extent_buffer(leaf, ptr + size, ptr,
1706                                       end - size - ptr);
1707
1708         iref = (struct btrfs_extent_inline_ref *)ptr;
1709         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1710         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1711                 struct btrfs_extent_data_ref *dref;
1712                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1713                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1714                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1715                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1716                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1717         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1718                 struct btrfs_shared_data_ref *sref;
1719                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1720                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1721                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1722         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1723                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1724         } else {
1725                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1726         }
1727         btrfs_mark_buffer_dirty(leaf);
1728 }
1729
1730 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1731                                  struct btrfs_root *root,
1732                                  struct btrfs_path *path,
1733                                  struct btrfs_extent_inline_ref **ref_ret,
1734                                  u64 bytenr, u64 num_bytes, u64 parent,
1735                                  u64 root_objectid, u64 owner, u64 offset)
1736 {
1737         int ret;
1738
1739         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1740                                            bytenr, num_bytes, parent,
1741                                            root_objectid, owner, offset, 0);
1742         if (ret != -ENOENT)
1743                 return ret;
1744
1745         btrfs_release_path(path);
1746         *ref_ret = NULL;
1747
1748         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1749                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1750                                             root_objectid);
1751         } else {
1752                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1753                                              root_objectid, owner, offset);
1754         }
1755         return ret;
1756 }
1757
1758 /*
1759  * helper to update/remove inline back ref
1760  */
1761 static noinline_for_stack
1762 void update_inline_extent_backref(struct btrfs_root *root,
1763                                   struct btrfs_path *path,
1764                                   struct btrfs_extent_inline_ref *iref,
1765                                   int refs_to_mod,
1766                                   struct btrfs_delayed_extent_op *extent_op)
1767 {
1768         struct extent_buffer *leaf;
1769         struct btrfs_extent_item *ei;
1770         struct btrfs_extent_data_ref *dref = NULL;
1771         struct btrfs_shared_data_ref *sref = NULL;
1772         unsigned long ptr;
1773         unsigned long end;
1774         u32 item_size;
1775         int size;
1776         int type;
1777         u64 refs;
1778
1779         leaf = path->nodes[0];
1780         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1781         refs = btrfs_extent_refs(leaf, ei);
1782         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1783         refs += refs_to_mod;
1784         btrfs_set_extent_refs(leaf, ei, refs);
1785         if (extent_op)
1786                 __run_delayed_extent_op(extent_op, leaf, ei);
1787
1788         type = btrfs_extent_inline_ref_type(leaf, iref);
1789
1790         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1791                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1792                 refs = btrfs_extent_data_ref_count(leaf, dref);
1793         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1794                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1795                 refs = btrfs_shared_data_ref_count(leaf, sref);
1796         } else {
1797                 refs = 1;
1798                 BUG_ON(refs_to_mod != -1);
1799         }
1800
1801         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1802         refs += refs_to_mod;
1803
1804         if (refs > 0) {
1805                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1806                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1807                 else
1808                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1809         } else {
1810                 size =  btrfs_extent_inline_ref_size(type);
1811                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1812                 ptr = (unsigned long)iref;
1813                 end = (unsigned long)ei + item_size;
1814                 if (ptr + size < end)
1815                         memmove_extent_buffer(leaf, ptr, ptr + size,
1816                                               end - ptr - size);
1817                 item_size -= size;
1818                 btrfs_truncate_item(root, path, item_size, 1);
1819         }
1820         btrfs_mark_buffer_dirty(leaf);
1821 }
1822
1823 static noinline_for_stack
1824 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1825                                  struct btrfs_root *root,
1826                                  struct btrfs_path *path,
1827                                  u64 bytenr, u64 num_bytes, u64 parent,
1828                                  u64 root_objectid, u64 owner,
1829                                  u64 offset, int refs_to_add,
1830                                  struct btrfs_delayed_extent_op *extent_op)
1831 {
1832         struct btrfs_extent_inline_ref *iref;
1833         int ret;
1834
1835         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1836                                            bytenr, num_bytes, parent,
1837                                            root_objectid, owner, offset, 1);
1838         if (ret == 0) {
1839                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1840                 update_inline_extent_backref(root, path, iref,
1841                                              refs_to_add, extent_op);
1842         } else if (ret == -ENOENT) {
1843                 setup_inline_extent_backref(root, path, iref, parent,
1844                                             root_objectid, owner, offset,
1845                                             refs_to_add, extent_op);
1846                 ret = 0;
1847         }
1848         return ret;
1849 }
1850
1851 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1852                                  struct btrfs_root *root,
1853                                  struct btrfs_path *path,
1854                                  u64 bytenr, u64 parent, u64 root_objectid,
1855                                  u64 owner, u64 offset, int refs_to_add)
1856 {
1857         int ret;
1858         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1859                 BUG_ON(refs_to_add != 1);
1860                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1861                                             parent, root_objectid);
1862         } else {
1863                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1864                                              parent, root_objectid,
1865                                              owner, offset, refs_to_add);
1866         }
1867         return ret;
1868 }
1869
1870 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1871                                  struct btrfs_root *root,
1872                                  struct btrfs_path *path,
1873                                  struct btrfs_extent_inline_ref *iref,
1874                                  int refs_to_drop, int is_data)
1875 {
1876         int ret = 0;
1877
1878         BUG_ON(!is_data && refs_to_drop != 1);
1879         if (iref) {
1880                 update_inline_extent_backref(root, path, iref,
1881                                              -refs_to_drop, NULL);
1882         } else if (is_data) {
1883                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1884         } else {
1885                 ret = btrfs_del_item(trans, root, path);
1886         }
1887         return ret;
1888 }
1889
1890 static int btrfs_issue_discard(struct block_device *bdev,
1891                                 u64 start, u64 len)
1892 {
1893         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1894 }
1895
1896 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1897                                 u64 num_bytes, u64 *actual_bytes)
1898 {
1899         int ret;
1900         u64 discarded_bytes = 0;
1901         struct btrfs_bio *bbio = NULL;
1902
1903
1904         /* Tell the block device(s) that the sectors can be discarded */
1905         ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1906                               bytenr, &num_bytes, &bbio, 0);
1907         /* Error condition is -ENOMEM */
1908         if (!ret) {
1909                 struct btrfs_bio_stripe *stripe = bbio->stripes;
1910                 int i;
1911
1912
1913                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1914                         if (!stripe->dev->can_discard)
1915                                 continue;
1916
1917                         ret = btrfs_issue_discard(stripe->dev->bdev,
1918                                                   stripe->physical,
1919                                                   stripe->length);
1920                         if (!ret)
1921                                 discarded_bytes += stripe->length;
1922                         else if (ret != -EOPNOTSUPP)
1923                                 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1924
1925                         /*
1926                          * Just in case we get back EOPNOTSUPP for some reason,
1927                          * just ignore the return value so we don't screw up
1928                          * people calling discard_extent.
1929                          */
1930                         ret = 0;
1931                 }
1932                 kfree(bbio);
1933         }
1934
1935         if (actual_bytes)
1936                 *actual_bytes = discarded_bytes;
1937
1938
1939         if (ret == -EOPNOTSUPP)
1940                 ret = 0;
1941         return ret;
1942 }
1943
1944 /* Can return -ENOMEM */
1945 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1946                          struct btrfs_root *root,
1947                          u64 bytenr, u64 num_bytes, u64 parent,
1948                          u64 root_objectid, u64 owner, u64 offset, int for_cow)
1949 {
1950         int ret;
1951         struct btrfs_fs_info *fs_info = root->fs_info;
1952
1953         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1954                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1955
1956         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1957                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1958                                         num_bytes,
1959                                         parent, root_objectid, (int)owner,
1960                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1961         } else {
1962                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1963                                         num_bytes,
1964                                         parent, root_objectid, owner, offset,
1965                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1966         }
1967         return ret;
1968 }
1969
1970 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1971                                   struct btrfs_root *root,
1972                                   u64 bytenr, u64 num_bytes,
1973                                   u64 parent, u64 root_objectid,
1974                                   u64 owner, u64 offset, int refs_to_add,
1975                                   struct btrfs_delayed_extent_op *extent_op)
1976 {
1977         struct btrfs_path *path;
1978         struct extent_buffer *leaf;
1979         struct btrfs_extent_item *item;
1980         u64 refs;
1981         int ret;
1982
1983         path = btrfs_alloc_path();
1984         if (!path)
1985                 return -ENOMEM;
1986
1987         path->reada = 1;
1988         path->leave_spinning = 1;
1989         /* this will setup the path even if it fails to insert the back ref */
1990         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1991                                            path, bytenr, num_bytes, parent,
1992                                            root_objectid, owner, offset,
1993                                            refs_to_add, extent_op);
1994         if (ret != -EAGAIN)
1995                 goto out;
1996
1997         leaf = path->nodes[0];
1998         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1999         refs = btrfs_extent_refs(leaf, item);
2000         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2001         if (extent_op)
2002                 __run_delayed_extent_op(extent_op, leaf, item);
2003
2004         btrfs_mark_buffer_dirty(leaf);
2005         btrfs_release_path(path);
2006
2007         path->reada = 1;
2008         path->leave_spinning = 1;
2009
2010         /* now insert the actual backref */
2011         ret = insert_extent_backref(trans, root->fs_info->extent_root,
2012                                     path, bytenr, parent, root_objectid,
2013                                     owner, offset, refs_to_add);
2014         if (ret)
2015                 btrfs_abort_transaction(trans, root, ret);
2016 out:
2017         btrfs_free_path(path);
2018         return ret;
2019 }
2020
2021 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2022                                 struct btrfs_root *root,
2023                                 struct btrfs_delayed_ref_node *node,
2024                                 struct btrfs_delayed_extent_op *extent_op,
2025                                 int insert_reserved)
2026 {
2027         int ret = 0;
2028         struct btrfs_delayed_data_ref *ref;
2029         struct btrfs_key ins;
2030         u64 parent = 0;
2031         u64 ref_root = 0;
2032         u64 flags = 0;
2033
2034         ins.objectid = node->bytenr;
2035         ins.offset = node->num_bytes;
2036         ins.type = BTRFS_EXTENT_ITEM_KEY;
2037
2038         ref = btrfs_delayed_node_to_data_ref(node);
2039         trace_run_delayed_data_ref(node, ref, node->action);
2040
2041         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2042                 parent = ref->parent;
2043         else
2044                 ref_root = ref->root;
2045
2046         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2047                 if (extent_op)
2048                         flags |= extent_op->flags_to_set;
2049                 ret = alloc_reserved_file_extent(trans, root,
2050                                                  parent, ref_root, flags,
2051                                                  ref->objectid, ref->offset,
2052                                                  &ins, node->ref_mod);
2053         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2054                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2055                                              node->num_bytes, parent,
2056                                              ref_root, ref->objectid,
2057                                              ref->offset, node->ref_mod,
2058                                              extent_op);
2059         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2060                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2061                                           node->num_bytes, parent,
2062                                           ref_root, ref->objectid,
2063                                           ref->offset, node->ref_mod,
2064                                           extent_op);
2065         } else {
2066                 BUG();
2067         }
2068         return ret;
2069 }
2070
2071 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2072                                     struct extent_buffer *leaf,
2073                                     struct btrfs_extent_item *ei)
2074 {
2075         u64 flags = btrfs_extent_flags(leaf, ei);
2076         if (extent_op->update_flags) {
2077                 flags |= extent_op->flags_to_set;
2078                 btrfs_set_extent_flags(leaf, ei, flags);
2079         }
2080
2081         if (extent_op->update_key) {
2082                 struct btrfs_tree_block_info *bi;
2083                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2084                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2085                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2086         }
2087 }
2088
2089 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2090                                  struct btrfs_root *root,
2091                                  struct btrfs_delayed_ref_node *node,
2092                                  struct btrfs_delayed_extent_op *extent_op)
2093 {
2094         struct btrfs_key key;
2095         struct btrfs_path *path;
2096         struct btrfs_extent_item *ei;
2097         struct extent_buffer *leaf;
2098         u32 item_size;
2099         int ret;
2100         int err = 0;
2101         int metadata = !extent_op->is_data;
2102
2103         if (trans->aborted)
2104                 return 0;
2105
2106         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2107                 metadata = 0;
2108
2109         path = btrfs_alloc_path();
2110         if (!path)
2111                 return -ENOMEM;
2112
2113         key.objectid = node->bytenr;
2114
2115         if (metadata) {
2116                 key.type = BTRFS_METADATA_ITEM_KEY;
2117                 key.offset = extent_op->level;
2118         } else {
2119                 key.type = BTRFS_EXTENT_ITEM_KEY;
2120                 key.offset = node->num_bytes;
2121         }
2122
2123 again:
2124         path->reada = 1;
2125         path->leave_spinning = 1;
2126         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2127                                 path, 0, 1);
2128         if (ret < 0) {
2129                 err = ret;
2130                 goto out;
2131         }
2132         if (ret > 0) {
2133                 if (metadata) {
2134                         if (path->slots[0] > 0) {
2135                                 path->slots[0]--;
2136                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
2137                                                       path->slots[0]);
2138                                 if (key.objectid == node->bytenr &&
2139                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
2140                                     key.offset == node->num_bytes)
2141                                         ret = 0;
2142                         }
2143                         if (ret > 0) {
2144                                 btrfs_release_path(path);
2145                                 metadata = 0;
2146
2147                                 key.objectid = node->bytenr;
2148                                 key.offset = node->num_bytes;
2149                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2150                                 goto again;
2151                         }
2152                 } else {
2153                         err = -EIO;
2154                         goto out;
2155                 }
2156         }
2157
2158         leaf = path->nodes[0];
2159         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2160 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2161         if (item_size < sizeof(*ei)) {
2162                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2163                                              path, (u64)-1, 0);
2164                 if (ret < 0) {
2165                         err = ret;
2166                         goto out;
2167                 }
2168                 leaf = path->nodes[0];
2169                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2170         }
2171 #endif
2172         BUG_ON(item_size < sizeof(*ei));
2173         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2174         __run_delayed_extent_op(extent_op, leaf, ei);
2175
2176         btrfs_mark_buffer_dirty(leaf);
2177 out:
2178         btrfs_free_path(path);
2179         return err;
2180 }
2181
2182 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2183                                 struct btrfs_root *root,
2184                                 struct btrfs_delayed_ref_node *node,
2185                                 struct btrfs_delayed_extent_op *extent_op,
2186                                 int insert_reserved)
2187 {
2188         int ret = 0;
2189         struct btrfs_delayed_tree_ref *ref;
2190         struct btrfs_key ins;
2191         u64 parent = 0;
2192         u64 ref_root = 0;
2193         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2194                                                  SKINNY_METADATA);
2195
2196         ref = btrfs_delayed_node_to_tree_ref(node);
2197         trace_run_delayed_tree_ref(node, ref, node->action);
2198
2199         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2200                 parent = ref->parent;
2201         else
2202                 ref_root = ref->root;
2203
2204         ins.objectid = node->bytenr;
2205         if (skinny_metadata) {
2206                 ins.offset = ref->level;
2207                 ins.type = BTRFS_METADATA_ITEM_KEY;
2208         } else {
2209                 ins.offset = node->num_bytes;
2210                 ins.type = BTRFS_EXTENT_ITEM_KEY;
2211         }
2212
2213         BUG_ON(node->ref_mod != 1);
2214         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2215                 BUG_ON(!extent_op || !extent_op->update_flags);
2216                 ret = alloc_reserved_tree_block(trans, root,
2217                                                 parent, ref_root,
2218                                                 extent_op->flags_to_set,
2219                                                 &extent_op->key,
2220                                                 ref->level, &ins);
2221         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2222                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2223                                              node->num_bytes, parent, ref_root,
2224                                              ref->level, 0, 1, extent_op);
2225         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2226                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2227                                           node->num_bytes, parent, ref_root,
2228                                           ref->level, 0, 1, extent_op);
2229         } else {
2230                 BUG();
2231         }
2232         return ret;
2233 }
2234
2235 /* helper function to actually process a single delayed ref entry */
2236 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2237                                struct btrfs_root *root,
2238                                struct btrfs_delayed_ref_node *node,
2239                                struct btrfs_delayed_extent_op *extent_op,
2240                                int insert_reserved)
2241 {
2242         int ret = 0;
2243
2244         if (trans->aborted) {
2245                 if (insert_reserved)
2246                         btrfs_pin_extent(root, node->bytenr,
2247                                          node->num_bytes, 1);
2248                 return 0;
2249         }
2250
2251         if (btrfs_delayed_ref_is_head(node)) {
2252                 struct btrfs_delayed_ref_head *head;
2253                 /*
2254                  * we've hit the end of the chain and we were supposed
2255                  * to insert this extent into the tree.  But, it got
2256                  * deleted before we ever needed to insert it, so all
2257                  * we have to do is clean up the accounting
2258                  */
2259                 BUG_ON(extent_op);
2260                 head = btrfs_delayed_node_to_head(node);
2261                 trace_run_delayed_ref_head(node, head, node->action);
2262
2263                 if (insert_reserved) {
2264                         btrfs_pin_extent(root, node->bytenr,
2265                                          node->num_bytes, 1);
2266                         if (head->is_data) {
2267                                 ret = btrfs_del_csums(trans, root,
2268                                                       node->bytenr,
2269                                                       node->num_bytes);
2270                         }
2271                 }
2272                 return ret;
2273         }
2274
2275         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2276             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2277                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2278                                            insert_reserved);
2279         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2280                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2281                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2282                                            insert_reserved);
2283         else
2284                 BUG();
2285         return ret;
2286 }
2287
2288 static noinline struct btrfs_delayed_ref_node *
2289 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2290 {
2291         struct rb_node *node;
2292         struct btrfs_delayed_ref_node *ref, *last = NULL;;
2293
2294         /*
2295          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2296          * this prevents ref count from going down to zero when
2297          * there still are pending delayed ref.
2298          */
2299         node = rb_first(&head->ref_root);
2300         while (node) {
2301                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2302                                 rb_node);
2303                 if (ref->action == BTRFS_ADD_DELAYED_REF)
2304                         return ref;
2305                 else if (last == NULL)
2306                         last = ref;
2307                 node = rb_next(node);
2308         }
2309         return last;
2310 }
2311
2312 /*
2313  * Returns 0 on success or if called with an already aborted transaction.
2314  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2315  */
2316 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2317                                              struct btrfs_root *root,
2318                                              unsigned long nr)
2319 {
2320         struct btrfs_delayed_ref_root *delayed_refs;
2321         struct btrfs_delayed_ref_node *ref;
2322         struct btrfs_delayed_ref_head *locked_ref = NULL;
2323         struct btrfs_delayed_extent_op *extent_op;
2324         struct btrfs_fs_info *fs_info = root->fs_info;
2325         ktime_t start = ktime_get();
2326         int ret;
2327         unsigned long count = 0;
2328         unsigned long actual_count = 0;
2329         int must_insert_reserved = 0;
2330
2331         delayed_refs = &trans->transaction->delayed_refs;
2332         while (1) {
2333                 if (!locked_ref) {
2334                         if (count >= nr)
2335                                 break;
2336
2337                         spin_lock(&delayed_refs->lock);
2338                         locked_ref = btrfs_select_ref_head(trans);
2339                         if (!locked_ref) {
2340                                 spin_unlock(&delayed_refs->lock);
2341                                 break;
2342                         }
2343
2344                         /* grab the lock that says we are going to process
2345                          * all the refs for this head */
2346                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2347                         spin_unlock(&delayed_refs->lock);
2348                         /*
2349                          * we may have dropped the spin lock to get the head
2350                          * mutex lock, and that might have given someone else
2351                          * time to free the head.  If that's true, it has been
2352                          * removed from our list and we can move on.
2353                          */
2354                         if (ret == -EAGAIN) {
2355                                 locked_ref = NULL;
2356                                 count++;
2357                                 continue;
2358                         }
2359                 }
2360
2361                 /*
2362                  * We need to try and merge add/drops of the same ref since we
2363                  * can run into issues with relocate dropping the implicit ref
2364                  * and then it being added back again before the drop can
2365                  * finish.  If we merged anything we need to re-loop so we can
2366                  * get a good ref.
2367                  */
2368                 spin_lock(&locked_ref->lock);
2369                 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2370                                          locked_ref);
2371
2372                 /*
2373                  * locked_ref is the head node, so we have to go one
2374                  * node back for any delayed ref updates
2375                  */
2376                 ref = select_delayed_ref(locked_ref);
2377
2378                 if (ref && ref->seq &&
2379                     btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2380                         spin_unlock(&locked_ref->lock);
2381                         btrfs_delayed_ref_unlock(locked_ref);
2382                         spin_lock(&delayed_refs->lock);
2383                         locked_ref->processing = 0;
2384                         delayed_refs->num_heads_ready++;
2385                         spin_unlock(&delayed_refs->lock);
2386                         locked_ref = NULL;
2387                         cond_resched();
2388                         continue;
2389                 }
2390
2391                 /*
2392                  * record the must insert reserved flag before we
2393                  * drop the spin lock.
2394                  */
2395                 must_insert_reserved = locked_ref->must_insert_reserved;
2396                 locked_ref->must_insert_reserved = 0;
2397
2398                 extent_op = locked_ref->extent_op;
2399                 locked_ref->extent_op = NULL;
2400
2401                 if (!ref) {
2402
2403
2404                         /* All delayed refs have been processed, Go ahead
2405                          * and send the head node to run_one_delayed_ref,
2406                          * so that any accounting fixes can happen
2407                          */
2408                         ref = &locked_ref->node;
2409
2410                         if (extent_op && must_insert_reserved) {
2411                                 btrfs_free_delayed_extent_op(extent_op);
2412                                 extent_op = NULL;
2413                         }
2414
2415                         if (extent_op) {
2416                                 spin_unlock(&locked_ref->lock);
2417                                 ret = run_delayed_extent_op(trans, root,
2418                                                             ref, extent_op);
2419                                 btrfs_free_delayed_extent_op(extent_op);
2420
2421                                 if (ret) {
2422                                         /*
2423                                          * Need to reset must_insert_reserved if
2424                                          * there was an error so the abort stuff
2425                                          * can cleanup the reserved space
2426                                          * properly.
2427                                          */
2428                                         if (must_insert_reserved)
2429                                                 locked_ref->must_insert_reserved = 1;
2430                                         locked_ref->processing = 0;
2431                                         btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2432                                         btrfs_delayed_ref_unlock(locked_ref);
2433                                         return ret;
2434                                 }
2435                                 continue;
2436                         }
2437
2438                         /*
2439                          * Need to drop our head ref lock and re-aqcuire the
2440                          * delayed ref lock and then re-check to make sure
2441                          * nobody got added.
2442                          */
2443                         spin_unlock(&locked_ref->lock);
2444                         spin_lock(&delayed_refs->lock);
2445                         spin_lock(&locked_ref->lock);
2446                         if (rb_first(&locked_ref->ref_root)) {
2447                                 spin_unlock(&locked_ref->lock);
2448                                 spin_unlock(&delayed_refs->lock);
2449                                 continue;
2450                         }
2451                         ref->in_tree = 0;
2452                         delayed_refs->num_heads--;
2453                         rb_erase(&locked_ref->href_node,
2454                                  &delayed_refs->href_root);
2455                         spin_unlock(&delayed_refs->lock);
2456                 } else {
2457                         actual_count++;
2458                         ref->in_tree = 0;
2459                         rb_erase(&ref->rb_node, &locked_ref->ref_root);
2460                 }
2461                 atomic_dec(&delayed_refs->num_entries);
2462
2463                 if (!btrfs_delayed_ref_is_head(ref)) {
2464                         /*
2465                          * when we play the delayed ref, also correct the
2466                          * ref_mod on head
2467                          */
2468                         switch (ref->action) {
2469                         case BTRFS_ADD_DELAYED_REF:
2470                         case BTRFS_ADD_DELAYED_EXTENT:
2471                                 locked_ref->node.ref_mod -= ref->ref_mod;
2472                                 break;
2473                         case BTRFS_DROP_DELAYED_REF:
2474                                 locked_ref->node.ref_mod += ref->ref_mod;
2475                                 break;
2476                         default:
2477                                 WARN_ON(1);
2478                         }
2479                 }
2480                 spin_unlock(&locked_ref->lock);
2481
2482                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2483                                           must_insert_reserved);
2484
2485                 btrfs_free_delayed_extent_op(extent_op);
2486                 if (ret) {
2487                         locked_ref->processing = 0;
2488                         btrfs_delayed_ref_unlock(locked_ref);
2489                         btrfs_put_delayed_ref(ref);
2490                         btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2491                         return ret;
2492                 }
2493
2494                 /*
2495                  * If this node is a head, that means all the refs in this head
2496                  * have been dealt with, and we will pick the next head to deal
2497                  * with, so we must unlock the head and drop it from the cluster
2498                  * list before we release it.
2499                  */
2500                 if (btrfs_delayed_ref_is_head(ref)) {
2501                         btrfs_delayed_ref_unlock(locked_ref);
2502                         locked_ref = NULL;
2503                 }
2504                 btrfs_put_delayed_ref(ref);
2505                 count++;
2506                 cond_resched();
2507         }
2508
2509         /*
2510          * We don't want to include ref heads since we can have empty ref heads
2511          * and those will drastically skew our runtime down since we just do
2512          * accounting, no actual extent tree updates.
2513          */
2514         if (actual_count > 0) {
2515                 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2516                 u64 avg;
2517
2518                 /*
2519                  * We weigh the current average higher than our current runtime
2520                  * to avoid large swings in the average.
2521                  */
2522                 spin_lock(&delayed_refs->lock);
2523                 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2524                 avg = div64_u64(avg, 4);
2525                 fs_info->avg_delayed_ref_runtime = avg;
2526                 spin_unlock(&delayed_refs->lock);
2527         }
2528         return 0;
2529 }
2530
2531 #ifdef SCRAMBLE_DELAYED_REFS
2532 /*
2533  * Normally delayed refs get processed in ascending bytenr order. This
2534  * correlates in most cases to the order added. To expose dependencies on this
2535  * order, we start to process the tree in the middle instead of the beginning
2536  */
2537 static u64 find_middle(struct rb_root *root)
2538 {
2539         struct rb_node *n = root->rb_node;
2540         struct btrfs_delayed_ref_node *entry;
2541         int alt = 1;
2542         u64 middle;
2543         u64 first = 0, last = 0;
2544
2545         n = rb_first(root);
2546         if (n) {
2547                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2548                 first = entry->bytenr;
2549         }
2550         n = rb_last(root);
2551         if (n) {
2552                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2553                 last = entry->bytenr;
2554         }
2555         n = root->rb_node;
2556
2557         while (n) {
2558                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2559                 WARN_ON(!entry->in_tree);
2560
2561                 middle = entry->bytenr;
2562
2563                 if (alt)
2564                         n = n->rb_left;
2565                 else
2566                         n = n->rb_right;
2567
2568                 alt = 1 - alt;
2569         }
2570         return middle;
2571 }
2572 #endif
2573
2574 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2575                                          struct btrfs_fs_info *fs_info)
2576 {
2577         struct qgroup_update *qgroup_update;
2578         int ret = 0;
2579
2580         if (list_empty(&trans->qgroup_ref_list) !=
2581             !trans->delayed_ref_elem.seq) {
2582                 /* list without seq or seq without list */
2583                 btrfs_err(fs_info,
2584                         "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2585                         list_empty(&trans->qgroup_ref_list) ? "" : " not",
2586                         (u32)(trans->delayed_ref_elem.seq >> 32),
2587                         (u32)trans->delayed_ref_elem.seq);
2588                 BUG();
2589         }
2590
2591         if (!trans->delayed_ref_elem.seq)
2592                 return 0;
2593
2594         while (!list_empty(&trans->qgroup_ref_list)) {
2595                 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2596                                                  struct qgroup_update, list);
2597                 list_del(&qgroup_update->list);
2598                 if (!ret)
2599                         ret = btrfs_qgroup_account_ref(
2600                                         trans, fs_info, qgroup_update->node,
2601                                         qgroup_update->extent_op);
2602                 kfree(qgroup_update);
2603         }
2604
2605         btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2606
2607         return ret;
2608 }
2609
2610 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2611 {
2612         u64 num_bytes;
2613
2614         num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2615                              sizeof(struct btrfs_extent_inline_ref));
2616         if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2617                 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2618
2619         /*
2620          * We don't ever fill up leaves all the way so multiply by 2 just to be
2621          * closer to what we're really going to want to ouse.
2622          */
2623         return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2624 }
2625
2626 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2627                                        struct btrfs_root *root)
2628 {
2629         struct btrfs_block_rsv *global_rsv;
2630         u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2631         u64 num_bytes;
2632         int ret = 0;
2633
2634         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2635         num_heads = heads_to_leaves(root, num_heads);
2636         if (num_heads > 1)
2637                 num_bytes += (num_heads - 1) * root->leafsize;
2638         num_bytes <<= 1;
2639         global_rsv = &root->fs_info->global_block_rsv;
2640
2641         /*
2642          * If we can't allocate any more chunks lets make sure we have _lots_ of
2643          * wiggle room since running delayed refs can create more delayed refs.
2644          */
2645         if (global_rsv->space_info->full)
2646                 num_bytes <<= 1;
2647
2648         spin_lock(&global_rsv->lock);
2649         if (global_rsv->reserved <= num_bytes)
2650                 ret = 1;
2651         spin_unlock(&global_rsv->lock);
2652         return ret;
2653 }
2654
2655 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2656                                        struct btrfs_root *root)
2657 {
2658         struct btrfs_fs_info *fs_info = root->fs_info;
2659         u64 num_entries =
2660                 atomic_read(&trans->transaction->delayed_refs.num_entries);
2661         u64 avg_runtime;
2662
2663         smp_mb();
2664         avg_runtime = fs_info->avg_delayed_ref_runtime;
2665         if (num_entries * avg_runtime >= NSEC_PER_SEC)
2666                 return 1;
2667
2668         return btrfs_check_space_for_delayed_refs(trans, root);
2669 }
2670
2671 /*
2672  * this starts processing the delayed reference count updates and
2673  * extent insertions we have queued up so far.  count can be
2674  * 0, which means to process everything in the tree at the start
2675  * of the run (but not newly added entries), or it can be some target
2676  * number you'd like to process.
2677  *
2678  * Returns 0 on success or if called with an aborted transaction
2679  * Returns <0 on error and aborts the transaction
2680  */
2681 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2682                            struct btrfs_root *root, unsigned long count)
2683 {
2684         struct rb_node *node;
2685         struct btrfs_delayed_ref_root *delayed_refs;
2686         struct btrfs_delayed_ref_head *head;
2687         int ret;
2688         int run_all = count == (unsigned long)-1;
2689         int run_most = 0;
2690
2691         /* We'll clean this up in btrfs_cleanup_transaction */
2692         if (trans->aborted)
2693                 return 0;
2694
2695         if (root == root->fs_info->extent_root)
2696                 root = root->fs_info->tree_root;
2697
2698         btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2699
2700         delayed_refs = &trans->transaction->delayed_refs;
2701         if (count == 0) {
2702                 count = atomic_read(&delayed_refs->num_entries) * 2;
2703                 run_most = 1;
2704         }
2705
2706 again:
2707 #ifdef SCRAMBLE_DELAYED_REFS
2708         delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2709 #endif
2710         ret = __btrfs_run_delayed_refs(trans, root, count);
2711         if (ret < 0) {
2712                 btrfs_abort_transaction(trans, root, ret);
2713                 return ret;
2714         }
2715
2716         if (run_all) {
2717                 if (!list_empty(&trans->new_bgs))
2718                         btrfs_create_pending_block_groups(trans, root);
2719
2720                 spin_lock(&delayed_refs->lock);
2721                 node = rb_first(&delayed_refs->href_root);
2722                 if (!node) {
2723                         spin_unlock(&delayed_refs->lock);
2724                         goto out;
2725                 }
2726                 count = (unsigned long)-1;
2727
2728                 while (node) {
2729                         head = rb_entry(node, struct btrfs_delayed_ref_head,
2730                                         href_node);
2731                         if (btrfs_delayed_ref_is_head(&head->node)) {
2732                                 struct btrfs_delayed_ref_node *ref;
2733
2734                                 ref = &head->node;
2735                                 atomic_inc(&ref->refs);
2736
2737                                 spin_unlock(&delayed_refs->lock);
2738                                 /*
2739                                  * Mutex was contended, block until it's
2740                                  * released and try again
2741                                  */
2742                                 mutex_lock(&head->mutex);
2743                                 mutex_unlock(&head->mutex);
2744
2745                                 btrfs_put_delayed_ref(ref);
2746                                 cond_resched();
2747                                 goto again;
2748                         } else {
2749                                 WARN_ON(1);
2750                         }
2751                         node = rb_next(node);
2752                 }
2753                 spin_unlock(&delayed_refs->lock);
2754                 cond_resched();
2755                 goto again;
2756         }
2757 out:
2758         assert_qgroups_uptodate(trans);
2759         return 0;
2760 }
2761
2762 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2763                                 struct btrfs_root *root,
2764                                 u64 bytenr, u64 num_bytes, u64 flags,
2765                                 int level, int is_data)
2766 {
2767         struct btrfs_delayed_extent_op *extent_op;
2768         int ret;
2769
2770         extent_op = btrfs_alloc_delayed_extent_op();
2771         if (!extent_op)
2772                 return -ENOMEM;
2773
2774         extent_op->flags_to_set = flags;
2775         extent_op->update_flags = 1;
2776         extent_op->update_key = 0;
2777         extent_op->is_data = is_data ? 1 : 0;
2778         extent_op->level = level;
2779
2780         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2781                                           num_bytes, extent_op);
2782         if (ret)
2783                 btrfs_free_delayed_extent_op(extent_op);
2784         return ret;
2785 }
2786
2787 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2788                                       struct btrfs_root *root,
2789                                       struct btrfs_path *path,
2790                                       u64 objectid, u64 offset, u64 bytenr)
2791 {
2792         struct btrfs_delayed_ref_head *head;
2793         struct btrfs_delayed_ref_node *ref;
2794         struct btrfs_delayed_data_ref *data_ref;
2795         struct btrfs_delayed_ref_root *delayed_refs;
2796         struct rb_node *node;
2797         int ret = 0;
2798
2799         delayed_refs = &trans->transaction->delayed_refs;
2800         spin_lock(&delayed_refs->lock);
2801         head = btrfs_find_delayed_ref_head(trans, bytenr);
2802         if (!head) {
2803                 spin_unlock(&delayed_refs->lock);
2804                 return 0;
2805         }
2806
2807         if (!mutex_trylock(&head->mutex)) {
2808                 atomic_inc(&head->node.refs);
2809                 spin_unlock(&delayed_refs->lock);
2810
2811                 btrfs_release_path(path);
2812
2813                 /*
2814                  * Mutex was contended, block until it's released and let
2815                  * caller try again
2816                  */
2817                 mutex_lock(&head->mutex);
2818                 mutex_unlock(&head->mutex);
2819                 btrfs_put_delayed_ref(&head->node);
2820                 return -EAGAIN;
2821         }
2822         spin_unlock(&delayed_refs->lock);
2823
2824         spin_lock(&head->lock);
2825         node = rb_first(&head->ref_root);
2826         while (node) {
2827                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2828                 node = rb_next(node);
2829
2830                 /* If it's a shared ref we know a cross reference exists */
2831                 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2832                         ret = 1;
2833                         break;
2834                 }
2835
2836                 data_ref = btrfs_delayed_node_to_data_ref(ref);
2837
2838                 /*
2839                  * If our ref doesn't match the one we're currently looking at
2840                  * then we have a cross reference.
2841                  */
2842                 if (data_ref->root != root->root_key.objectid ||
2843                     data_ref->objectid != objectid ||
2844                     data_ref->offset != offset) {
2845                         ret = 1;
2846                         break;
2847                 }
2848         }
2849         spin_unlock(&head->lock);
2850         mutex_unlock(&head->mutex);
2851         return ret;
2852 }
2853
2854 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2855                                         struct btrfs_root *root,
2856                                         struct btrfs_path *path,
2857                                         u64 objectid, u64 offset, u64 bytenr)
2858 {
2859         struct btrfs_root *extent_root = root->fs_info->extent_root;
2860         struct extent_buffer *leaf;
2861         struct btrfs_extent_data_ref *ref;
2862         struct btrfs_extent_inline_ref *iref;
2863         struct btrfs_extent_item *ei;
2864         struct btrfs_key key;
2865         u32 item_size;
2866         int ret;
2867
2868         key.objectid = bytenr;
2869         key.offset = (u64)-1;
2870         key.type = BTRFS_EXTENT_ITEM_KEY;
2871
2872         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2873         if (ret < 0)
2874                 goto out;
2875         BUG_ON(ret == 0); /* Corruption */
2876
2877         ret = -ENOENT;
2878         if (path->slots[0] == 0)
2879                 goto out;
2880
2881         path->slots[0]--;
2882         leaf = path->nodes[0];
2883         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2884
2885         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2886                 goto out;
2887
2888         ret = 1;
2889         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2890 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2891         if (item_size < sizeof(*ei)) {
2892                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2893                 goto out;
2894         }
2895 #endif
2896         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2897
2898         if (item_size != sizeof(*ei) +
2899             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2900                 goto out;
2901
2902         if (btrfs_extent_generation(leaf, ei) <=
2903             btrfs_root_last_snapshot(&root->root_item))
2904                 goto out;
2905
2906         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2907         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2908             BTRFS_EXTENT_DATA_REF_KEY)
2909                 goto out;
2910
2911         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2912         if (btrfs_extent_refs(leaf, ei) !=
2913             btrfs_extent_data_ref_count(leaf, ref) ||
2914             btrfs_extent_data_ref_root(leaf, ref) !=
2915             root->root_key.objectid ||
2916             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2917             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2918                 goto out;
2919
2920         ret = 0;
2921 out:
2922         return ret;
2923 }
2924
2925 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2926                           struct btrfs_root *root,
2927                           u64 objectid, u64 offset, u64 bytenr)
2928 {
2929         struct btrfs_path *path;
2930         int ret;
2931         int ret2;
2932
2933         path = btrfs_alloc_path();
2934         if (!path)
2935                 return -ENOENT;
2936
2937         do {
2938                 ret = check_committed_ref(trans, root, path, objectid,
2939                                           offset, bytenr);
2940                 if (ret && ret != -ENOENT)
2941                         goto out;
2942
2943                 ret2 = check_delayed_ref(trans, root, path, objectid,
2944                                          offset, bytenr);
2945         } while (ret2 == -EAGAIN);
2946
2947         if (ret2 && ret2 != -ENOENT) {
2948                 ret = ret2;
2949                 goto out;
2950         }
2951
2952         if (ret != -ENOENT || ret2 != -ENOENT)
2953                 ret = 0;
2954 out:
2955         btrfs_free_path(path);
2956         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2957                 WARN_ON(ret > 0);
2958         return ret;
2959 }
2960
2961 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2962                            struct btrfs_root *root,
2963                            struct extent_buffer *buf,
2964                            int full_backref, int inc, int for_cow)
2965 {
2966         u64 bytenr;
2967         u64 num_bytes;
2968         u64 parent;
2969         u64 ref_root;
2970         u32 nritems;
2971         struct btrfs_key key;
2972         struct btrfs_file_extent_item *fi;
2973         int i;
2974         int level;
2975         int ret = 0;
2976         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2977                             u64, u64, u64, u64, u64, u64, int);
2978
2979         ref_root = btrfs_header_owner(buf);
2980         nritems = btrfs_header_nritems(buf);
2981         level = btrfs_header_level(buf);
2982
2983         if (!root->ref_cows && level == 0)
2984                 return 0;
2985
2986         if (inc)
2987                 process_func = btrfs_inc_extent_ref;
2988         else
2989                 process_func = btrfs_free_extent;
2990
2991         if (full_backref)
2992                 parent = buf->start;
2993         else
2994                 parent = 0;
2995
2996         for (i = 0; i < nritems; i++) {
2997                 if (level == 0) {
2998                         btrfs_item_key_to_cpu(buf, &key, i);
2999                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3000                                 continue;
3001                         fi = btrfs_item_ptr(buf, i,
3002                                             struct btrfs_file_extent_item);
3003                         if (btrfs_file_extent_type(buf, fi) ==
3004                             BTRFS_FILE_EXTENT_INLINE)
3005                                 continue;
3006                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3007                         if (bytenr == 0)
3008                                 continue;
3009
3010                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3011                         key.offset -= btrfs_file_extent_offset(buf, fi);
3012                         ret = process_func(trans, root, bytenr, num_bytes,
3013                                            parent, ref_root, key.objectid,
3014                                            key.offset, for_cow);
3015                         if (ret)
3016                                 goto fail;
3017                 } else {
3018                         bytenr = btrfs_node_blockptr(buf, i);
3019                         num_bytes = btrfs_level_size(root, level - 1);
3020                         ret = process_func(trans, root, bytenr, num_bytes,
3021                                            parent, ref_root, level - 1, 0,
3022                                            for_cow);
3023                         if (ret)
3024                                 goto fail;
3025                 }
3026         }
3027         return 0;
3028 fail:
3029         return ret;
3030 }
3031
3032 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3033                   struct extent_buffer *buf, int full_backref, int for_cow)
3034 {
3035         return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3036 }
3037
3038 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3039                   struct extent_buffer *buf, int full_backref, int for_cow)
3040 {
3041         return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3042 }
3043
3044 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3045                                  struct btrfs_root *root,
3046                                  struct btrfs_path *path,
3047                                  struct btrfs_block_group_cache *cache)
3048 {
3049         int ret;
3050         struct btrfs_root *extent_root = root->fs_info->extent_root;
3051         unsigned long bi;
3052         struct extent_buffer *leaf;
3053
3054         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3055         if (ret < 0)
3056                 goto fail;
3057         BUG_ON(ret); /* Corruption */
3058
3059         leaf = path->nodes[0];
3060         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3061         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3062         btrfs_mark_buffer_dirty(leaf);
3063         btrfs_release_path(path);
3064 fail:
3065         if (ret) {
3066                 btrfs_abort_transaction(trans, root, ret);
3067                 return ret;
3068         }
3069         return 0;
3070
3071 }
3072
3073 static struct btrfs_block_group_cache *
3074 next_block_group(struct btrfs_root *root,
3075                  struct btrfs_block_group_cache *cache)
3076 {
3077         struct rb_node *node;
3078         spin_lock(&root->fs_info->block_group_cache_lock);
3079         node = rb_next(&cache->cache_node);
3080         btrfs_put_block_group(cache);
3081         if (node) {
3082                 cache = rb_entry(node, struct btrfs_block_group_cache,
3083                                  cache_node);
3084                 btrfs_get_block_group(cache);
3085         } else
3086                 cache = NULL;
3087         spin_unlock(&root->fs_info->block_group_cache_lock);
3088         return cache;
3089 }
3090
3091 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3092                             struct btrfs_trans_handle *trans,
3093                             struct btrfs_path *path)
3094 {
3095         struct btrfs_root *root = block_group->fs_info->tree_root;
3096         struct inode *inode = NULL;
3097         u64 alloc_hint = 0;
3098         int dcs = BTRFS_DC_ERROR;
3099         int num_pages = 0;
3100         int retries = 0;
3101         int ret = 0;
3102
3103         /*
3104          * If this block group is smaller than 100 megs don't bother caching the
3105          * block group.
3106          */
3107         if (block_group->key.offset < (100 * 1024 * 1024)) {
3108                 spin_lock(&block_group->lock);
3109                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3110                 spin_unlock(&block_group->lock);
3111                 return 0;
3112         }
3113
3114 again:
3115         inode = lookup_free_space_inode(root, block_group, path);
3116         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3117                 ret = PTR_ERR(inode);
3118                 btrfs_release_path(path);
3119                 goto out;
3120         }
3121
3122         if (IS_ERR(inode)) {
3123                 BUG_ON(retries);
3124                 retries++;
3125
3126                 if (block_group->ro)
3127                         goto out_free;
3128
3129                 ret = create_free_space_inode(root, trans, block_group, path);
3130                 if (ret)
3131                         goto out_free;
3132                 goto again;
3133         }
3134
3135         /* We've already setup this transaction, go ahead and exit */
3136         if (block_group->cache_generation == trans->transid &&
3137             i_size_read(inode)) {
3138                 dcs = BTRFS_DC_SETUP;
3139                 goto out_put;
3140         }
3141
3142         /*
3143          * We want to set the generation to 0, that way if anything goes wrong
3144          * from here on out we know not to trust this cache when we load up next
3145          * time.
3146          */
3147         BTRFS_I(inode)->generation = 0;
3148         ret = btrfs_update_inode(trans, root, inode);
3149         WARN_ON(ret);
3150
3151         if (i_size_read(inode) > 0) {
3152                 ret = btrfs_check_trunc_cache_free_space(root,
3153                                         &root->fs_info->global_block_rsv);
3154                 if (ret)
3155                         goto out_put;
3156
3157                 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3158                 if (ret)
3159                         goto out_put;
3160         }
3161
3162         spin_lock(&block_group->lock);
3163         if (block_group->cached != BTRFS_CACHE_FINISHED ||
3164             !btrfs_test_opt(root, SPACE_CACHE)) {
3165                 /*
3166                  * don't bother trying to write stuff out _if_
3167                  * a) we're not cached,
3168                  * b) we're with nospace_cache mount option.
3169                  */
3170                 dcs = BTRFS_DC_WRITTEN;
3171                 spin_unlock(&block_group->lock);
3172                 goto out_put;
3173         }
3174         spin_unlock(&block_group->lock);
3175
3176         /*
3177          * Try to preallocate enough space based on how big the block group is.
3178          * Keep in mind this has to include any pinned space which could end up
3179          * taking up quite a bit since it's not folded into the other space
3180          * cache.
3181          */
3182         num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3183         if (!num_pages)
3184                 num_pages = 1;
3185
3186         num_pages *= 16;
3187         num_pages *= PAGE_CACHE_SIZE;
3188
3189         ret = btrfs_check_data_free_space(inode, num_pages);
3190         if (ret)
3191                 goto out_put;
3192
3193         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3194                                               num_pages, num_pages,
3195                                               &alloc_hint);
3196         if (!ret)
3197                 dcs = BTRFS_DC_SETUP;
3198         btrfs_free_reserved_data_space(inode, num_pages);
3199
3200 out_put:
3201         iput(inode);
3202 out_free:
3203         btrfs_release_path(path);
3204 out:
3205         spin_lock(&block_group->lock);
3206         if (!ret && dcs == BTRFS_DC_SETUP)
3207                 block_group->cache_generation = trans->transid;
3208         block_group->disk_cache_state = dcs;
3209         spin_unlock(&block_group->lock);
3210
3211         return ret;
3212 }
3213
3214 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3215                                    struct btrfs_root *root)
3216 {
3217         struct btrfs_block_group_cache *cache;
3218         int err = 0;
3219         struct btrfs_path *path;
3220         u64 last = 0;
3221
3222         path = btrfs_alloc_path();
3223         if (!path)
3224                 return -ENOMEM;
3225
3226 again:
3227         while (1) {
3228                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3229                 while (cache) {
3230                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3231                                 break;
3232                         cache = next_block_group(root, cache);
3233                 }
3234                 if (!cache) {
3235                         if (last == 0)
3236                                 break;
3237                         last = 0;
3238                         continue;
3239                 }
3240                 err = cache_save_setup(cache, trans, path);
3241                 last = cache->key.objectid + cache->key.offset;
3242                 btrfs_put_block_group(cache);
3243         }
3244
3245         while (1) {
3246                 if (last == 0) {
3247                         err = btrfs_run_delayed_refs(trans, root,
3248                                                      (unsigned long)-1);
3249                         if (err) /* File system offline */
3250                                 goto out;
3251                 }
3252
3253                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3254                 while (cache) {
3255                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3256                                 btrfs_put_block_group(cache);
3257                                 goto again;
3258                         }
3259
3260                         if (cache->dirty)
3261                                 break;
3262                         cache = next_block_group(root, cache);
3263                 }
3264                 if (!cache) {
3265                         if (last == 0)
3266                                 break;
3267                         last = 0;
3268                         continue;
3269                 }
3270
3271                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3272                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3273                 cache->dirty = 0;
3274                 last = cache->key.objectid + cache->key.offset;
3275
3276                 err = write_one_cache_group(trans, root, path, cache);
3277                 btrfs_put_block_group(cache);
3278                 if (err) /* File system offline */
3279                         goto out;
3280         }
3281
3282         while (1) {
3283                 /*
3284                  * I don't think this is needed since we're just marking our
3285                  * preallocated extent as written, but just in case it can't
3286                  * hurt.
3287                  */
3288                 if (last == 0) {
3289                         err = btrfs_run_delayed_refs(trans, root,
3290                                                      (unsigned long)-1);
3291                         if (err) /* File system offline */
3292                                 goto out;
3293                 }
3294
3295                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3296                 while (cache) {
3297                         /*
3298                          * Really this shouldn't happen, but it could if we
3299                          * couldn't write the entire preallocated extent and
3300                          * splitting the extent resulted in a new block.
3301                          */
3302                         if (cache->dirty) {
3303                                 btrfs_put_block_group(cache);
3304                                 goto again;
3305                         }
3306                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3307                                 break;
3308                         cache = next_block_group(root, cache);
3309                 }
3310                 if (!cache) {
3311                         if (last == 0)
3312                                 break;
3313                         last = 0;
3314                         continue;
3315                 }
3316
3317                 err = btrfs_write_out_cache(root, trans, cache, path);
3318
3319                 /*
3320                  * If we didn't have an error then the cache state is still
3321                  * NEED_WRITE, so we can set it to WRITTEN.
3322                  */
3323                 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3324                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
3325                 last = cache->key.objectid + cache->key.offset;
3326                 btrfs_put_block_group(cache);
3327         }
3328 out:
3329
3330         btrfs_free_path(path);
3331         return err;
3332 }
3333
3334 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3335 {
3336         struct btrfs_block_group_cache *block_group;
3337         int readonly = 0;
3338
3339         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3340         if (!block_group || block_group->ro)
3341                 readonly = 1;
3342         if (block_group)
3343                 btrfs_put_block_group(block_group);
3344         return readonly;
3345 }
3346
3347 static const char *alloc_name(u64 flags)
3348 {
3349         switch (flags) {
3350         case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3351                 return "mixed";
3352         case BTRFS_BLOCK_GROUP_METADATA:
3353                 return "metadata";
3354         case BTRFS_BLOCK_GROUP_DATA:
3355                 return "data";
3356         case BTRFS_BLOCK_GROUP_SYSTEM:
3357                 return "system";
3358         default:
3359                 WARN_ON(1);
3360                 return "invalid-combination";
3361         };
3362 }
3363
3364 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3365                              u64 total_bytes, u64 bytes_used,
3366                              struct btrfs_space_info **space_info)
3367 {
3368         struct btrfs_space_info *found;
3369         int i;
3370         int factor;
3371         int ret;
3372
3373         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3374                      BTRFS_BLOCK_GROUP_RAID10))
3375                 factor = 2;
3376         else
3377                 factor = 1;
3378
3379         found = __find_space_info(info, flags);
3380         if (found) {
3381                 spin_lock(&found->lock);
3382                 found->total_bytes += total_bytes;
3383                 found->disk_total += total_bytes * factor;
3384                 found->bytes_used += bytes_used;
3385                 found->disk_used += bytes_used * factor;
3386                 found->full = 0;
3387                 spin_unlock(&found->lock);
3388                 *space_info = found;
3389                 return 0;
3390         }
3391         found = kzalloc(sizeof(*found), GFP_NOFS);
3392         if (!found)
3393                 return -ENOMEM;
3394
3395         ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3396         if (ret) {
3397                 kfree(found);
3398                 return ret;
3399         }
3400
3401         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
3402                 INIT_LIST_HEAD(&found->block_groups[i]);
3403                 kobject_init(&found->block_group_kobjs[i], &btrfs_raid_ktype);
3404         }
3405         init_rwsem(&found->groups_sem);
3406         spin_lock_init(&found->lock);
3407         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3408         found->total_bytes = total_bytes;
3409         found->disk_total = total_bytes * factor;
3410         found->bytes_used = bytes_used;
3411         found->disk_used = bytes_used * factor;
3412         found->bytes_pinned = 0;
3413         found->bytes_reserved = 0;
3414         found->bytes_readonly = 0;
3415         found->bytes_may_use = 0;
3416         found->full = 0;
3417         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3418         found->chunk_alloc = 0;
3419         found->flush = 0;
3420         init_waitqueue_head(&found->wait);
3421
3422         ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3423                                     info->space_info_kobj, "%s",
3424                                     alloc_name(found->flags));
3425         if (ret) {
3426                 kfree(found);
3427                 return ret;
3428         }
3429
3430         *space_info = found;
3431         list_add_rcu(&found->list, &info->space_info);
3432         if (flags & BTRFS_BLOCK_GROUP_DATA)
3433                 info->data_sinfo = found;
3434
3435         return ret;
3436 }
3437
3438 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3439 {
3440         u64 extra_flags = chunk_to_extended(flags) &
3441                                 BTRFS_EXTENDED_PROFILE_MASK;
3442
3443         write_seqlock(&fs_info->profiles_lock);
3444         if (flags & BTRFS_BLOCK_GROUP_DATA)
3445                 fs_info->avail_data_alloc_bits |= extra_flags;
3446         if (flags & BTRFS_BLOCK_GROUP_METADATA)
3447                 fs_info->avail_metadata_alloc_bits |= extra_flags;
3448         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3449                 fs_info->avail_system_alloc_bits |= extra_flags;
3450         write_sequnlock(&fs_info->profiles_lock);
3451 }
3452
3453 /*
3454  * returns target flags in extended format or 0 if restripe for this
3455  * chunk_type is not in progress
3456  *
3457  * should be called with either volume_mutex or balance_lock held
3458  */
3459 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3460 {
3461         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3462         u64 target = 0;
3463
3464         if (!bctl)
3465                 return 0;
3466
3467         if (flags & BTRFS_BLOCK_GROUP_DATA &&
3468             bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3469                 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3470         } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3471                    bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3472                 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3473         } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3474                    bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3475                 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3476         }
3477
3478         return target;
3479 }
3480
3481 /*
3482  * @flags: available profiles in extended format (see ctree.h)
3483  *
3484  * Returns reduced profile in chunk format.  If profile changing is in
3485  * progress (either running or paused) picks the target profile (if it's
3486  * already available), otherwise falls back to plain reducing.
3487  */
3488 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3489 {
3490         /*
3491          * we add in the count of missing devices because we want
3492          * to make sure that any RAID levels on a degraded FS
3493          * continue to be honored.
3494          */
3495         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3496                 root->fs_info->fs_devices->missing_devices;
3497         u64 target;
3498         u64 tmp;
3499
3500         /*
3501          * see if restripe for this chunk_type is in progress, if so
3502          * try to reduce to the target profile
3503          */
3504         spin_lock(&root->fs_info->balance_lock);
3505         target = get_restripe_target(root->fs_info, flags);
3506         if (target) {
3507                 /* pick target profile only if it's already available */
3508                 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3509                         spin_unlock(&root->fs_info->balance_lock);
3510                         return extended_to_chunk(target);
3511                 }
3512         }
3513         spin_unlock(&root->fs_info->balance_lock);
3514
3515         /* First, mask out the RAID levels which aren't possible */
3516         if (num_devices == 1)
3517                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3518                            BTRFS_BLOCK_GROUP_RAID5);
3519         if (num_devices < 3)
3520                 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3521         if (num_devices < 4)
3522                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3523
3524         tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3525                        BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3526                        BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3527         flags &= ~tmp;
3528
3529         if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3530                 tmp = BTRFS_BLOCK_GROUP_RAID6;
3531         else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3532                 tmp = BTRFS_BLOCK_GROUP_RAID5;
3533         else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3534                 tmp = BTRFS_BLOCK_GROUP_RAID10;
3535         else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3536                 tmp = BTRFS_BLOCK_GROUP_RAID1;
3537         else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3538                 tmp = BTRFS_BLOCK_GROUP_RAID0;
3539
3540         return extended_to_chunk(flags | tmp);
3541 }
3542
3543 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3544 {
3545         unsigned seq;
3546
3547         do {
3548                 seq = read_seqbegin(&root->fs_info->profiles_lock);
3549
3550                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3551                         flags |= root->fs_info->avail_data_alloc_bits;
3552                 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3553                         flags |= root->fs_info->avail_system_alloc_bits;
3554                 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3555                         flags |= root->fs_info->avail_metadata_alloc_bits;
3556         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3557
3558         return btrfs_reduce_alloc_profile(root, flags);
3559 }
3560
3561 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3562 {
3563         u64 flags;
3564         u64 ret;
3565
3566         if (data)
3567                 flags = BTRFS_BLOCK_GROUP_DATA;
3568         else if (root == root->fs_info->chunk_root)
3569                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3570         else
3571                 flags = BTRFS_BLOCK_GROUP_METADATA;
3572
3573         ret = get_alloc_profile(root, flags);
3574         return ret;
3575 }
3576
3577 /*
3578  * This will check the space that the inode allocates from to make sure we have
3579  * enough space for bytes.
3580  */
3581 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3582 {
3583         struct btrfs_space_info *data_sinfo;
3584         struct btrfs_root *root = BTRFS_I(inode)->root;
3585         struct btrfs_fs_info *fs_info = root->fs_info;
3586         u64 used;
3587         int ret = 0, committed = 0, alloc_chunk = 1;
3588
3589         /* make sure bytes are sectorsize aligned */
3590         bytes = ALIGN(bytes, root->sectorsize);
3591
3592         if (btrfs_is_free_space_inode(inode)) {
3593                 committed = 1;
3594                 ASSERT(current->journal_info);
3595         }
3596
3597         data_sinfo = fs_info->data_sinfo;
3598         if (!data_sinfo)
3599                 goto alloc;
3600
3601 again:
3602         /* make sure we have enough space to handle the data first */
3603         spin_lock(&data_sinfo->lock);
3604         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3605                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3606                 data_sinfo->bytes_may_use;
3607
3608         if (used + bytes > data_sinfo->total_bytes) {
3609                 struct btrfs_trans_handle *trans;
3610
3611                 /*
3612                  * if we don't have enough free bytes in this space then we need
3613                  * to alloc a new chunk.
3614                  */
3615                 if (!data_sinfo->full && alloc_chunk) {
3616                         u64 alloc_target;
3617
3618                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3619                         spin_unlock(&data_sinfo->lock);
3620 alloc:
3621                         alloc_target = btrfs_get_alloc_profile(root, 1);
3622                         /*
3623                          * It is ugly that we don't call nolock join
3624                          * transaction for the free space inode case here.
3625                          * But it is safe because we only do the data space
3626                          * reservation for the free space cache in the
3627                          * transaction context, the common join transaction
3628                          * just increase the counter of the current transaction
3629                          * handler, doesn't try to acquire the trans_lock of
3630                          * the fs.
3631                          */
3632                         trans = btrfs_join_transaction(root);
3633                         if (IS_ERR(trans))
3634                                 return PTR_ERR(trans);
3635
3636                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3637                                              alloc_target,
3638                                              CHUNK_ALLOC_NO_FORCE);
3639                         btrfs_end_transaction(trans, root);
3640                         if (ret < 0) {
3641                                 if (ret != -ENOSPC)
3642                                         return ret;
3643                                 else
3644                                         goto commit_trans;
3645                         }
3646
3647                         if (!data_sinfo)
3648                                 data_sinfo = fs_info->data_sinfo;
3649
3650                         goto again;
3651                 }
3652
3653                 /*
3654                  * If we don't have enough pinned space to deal with this
3655                  * allocation don't bother committing the transaction.
3656                  */
3657                 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3658                                            bytes) < 0)
3659                         committed = 1;
3660                 spin_unlock(&data_sinfo->lock);
3661
3662                 /* commit the current transaction and try again */
3663 commit_trans:
3664                 if (!committed &&
3665                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3666                         committed = 1;
3667
3668                         trans = btrfs_join_transaction(root);
3669                         if (IS_ERR(trans))
3670                                 return PTR_ERR(trans);
3671                         ret = btrfs_commit_transaction(trans, root);
3672                         if (ret)
3673                                 return ret;
3674                         goto again;
3675                 }
3676
3677                 trace_btrfs_space_reservation(root->fs_info,
3678                                               "space_info:enospc",
3679                                               data_sinfo->flags, bytes, 1);
3680                 return -ENOSPC;
3681         }
3682         data_sinfo->bytes_may_use += bytes;
3683         trace_btrfs_space_reservation(root->fs_info, "space_info",
3684                                       data_sinfo->flags, bytes, 1);
3685         spin_unlock(&data_sinfo->lock);
3686
3687         return 0;
3688 }
3689
3690 /*
3691  * Called if we need to clear a data reservation for this inode.
3692  */
3693 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3694 {
3695         struct btrfs_root *root = BTRFS_I(inode)->root;
3696         struct btrfs_space_info *data_sinfo;
3697
3698         /* make sure bytes are sectorsize aligned */
3699         bytes = ALIGN(bytes, root->sectorsize);
3700
3701         data_sinfo = root->fs_info->data_sinfo;
3702         spin_lock(&data_sinfo->lock);
3703         WARN_ON(data_sinfo->bytes_may_use < bytes);
3704         data_sinfo->bytes_may_use -= bytes;
3705         trace_btrfs_space_reservation(root->fs_info, "space_info",
3706                                       data_sinfo->flags, bytes, 0);
3707         spin_unlock(&data_sinfo->lock);
3708 }
3709
3710 static void force_metadata_allocation(struct btrfs_fs_info *info)
3711 {
3712         struct list_head *head = &info->space_info;
3713         struct btrfs_space_info *found;
3714
3715         rcu_read_lock();
3716         list_for_each_entry_rcu(found, head, list) {
3717                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3718                         found->force_alloc = CHUNK_ALLOC_FORCE;
3719         }
3720         rcu_read_unlock();
3721 }
3722
3723 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3724 {
3725         return (global->size << 1);
3726 }
3727
3728 static int should_alloc_chunk(struct btrfs_root *root,
3729                               struct btrfs_space_info *sinfo, int force)
3730 {
3731         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3732         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3733         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3734         u64 thresh;
3735
3736         if (force == CHUNK_ALLOC_FORCE)
3737                 return 1;
3738
3739         /*
3740          * We need to take into account the global rsv because for all intents
3741          * and purposes it's used space.  Don't worry about locking the
3742          * global_rsv, it doesn't change except when the transaction commits.
3743          */
3744         if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3745                 num_allocated += calc_global_rsv_need_space(global_rsv);
3746
3747         /*
3748          * in limited mode, we want to have some free space up to
3749          * about 1% of the FS size.
3750          */
3751         if (force == CHUNK_ALLOC_LIMITED) {
3752                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3753                 thresh = max_t(u64, 64 * 1024 * 1024,
3754                                div_factor_fine(thresh, 1));
3755
3756                 if (num_bytes - num_allocated < thresh)
3757                         return 1;
3758         }
3759
3760         if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3761                 return 0;
3762         return 1;
3763 }
3764
3765 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3766 {
3767         u64 num_dev;
3768
3769         if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3770                     BTRFS_BLOCK_GROUP_RAID0 |
3771                     BTRFS_BLOCK_GROUP_RAID5 |
3772                     BTRFS_BLOCK_GROUP_RAID6))
3773                 num_dev = root->fs_info->fs_devices->rw_devices;
3774         else if (type & BTRFS_BLOCK_GROUP_RAID1)
3775                 num_dev = 2;
3776         else
3777                 num_dev = 1;    /* DUP or single */
3778
3779         /* metadata for updaing devices and chunk tree */
3780         return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3781 }
3782
3783 static void check_system_chunk(struct btrfs_trans_handle *trans,
3784                                struct btrfs_root *root, u64 type)
3785 {
3786         struct btrfs_space_info *info;
3787         u64 left;
3788         u64 thresh;
3789
3790         info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3791         spin_lock(&info->lock);
3792         left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3793                 info->bytes_reserved - info->bytes_readonly;
3794         spin_unlock(&info->lock);
3795
3796         thresh = get_system_chunk_thresh(root, type);
3797         if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3798                 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3799                         left, thresh, type);
3800                 dump_space_info(info, 0, 0);
3801         }
3802
3803         if (left < thresh) {
3804                 u64 flags;
3805
3806                 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3807                 btrfs_alloc_chunk(trans, root, flags);
3808         }
3809 }
3810
3811 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3812                           struct btrfs_root *extent_root, u64 flags, int force)
3813 {
3814         struct btrfs_space_info *space_info;
3815         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3816         int wait_for_alloc = 0;
3817         int ret = 0;
3818
3819         /* Don't re-enter if we're already allocating a chunk */
3820         if (trans->allocating_chunk)
3821                 return -ENOSPC;
3822
3823         space_info = __find_space_info(extent_root->fs_info, flags);
3824         if (!space_info) {
3825                 ret = update_space_info(extent_root->fs_info, flags,
3826                                         0, 0, &space_info);
3827                 BUG_ON(ret); /* -ENOMEM */
3828         }
3829         BUG_ON(!space_info); /* Logic error */
3830
3831 again:
3832         spin_lock(&space_info->lock);
3833         if (force < space_info->force_alloc)
3834                 force = space_info->force_alloc;
3835         if (space_info->full) {
3836                 if (should_alloc_chunk(extent_root, space_info, force))
3837                         ret = -ENOSPC;
3838                 else
3839                         ret = 0;
3840                 spin_unlock(&space_info->lock);
3841                 return ret;
3842         }
3843
3844         if (!should_alloc_chunk(extent_root, space_info, force)) {
3845                 spin_unlock(&space_info->lock);
3846                 return 0;
3847         } else if (space_info->chunk_alloc) {
3848                 wait_for_alloc = 1;
3849         } else {
3850                 space_info->chunk_alloc = 1;
3851         }
3852
3853         spin_unlock(&space_info->lock);
3854
3855         mutex_lock(&fs_info->chunk_mutex);
3856
3857         /*
3858          * The chunk_mutex is held throughout the entirety of a chunk
3859          * allocation, so once we've acquired the chunk_mutex we know that the
3860          * other guy is done and we need to recheck and see if we should
3861          * allocate.
3862          */
3863         if (wait_for_alloc) {
3864                 mutex_unlock(&fs_info->chunk_mutex);
3865                 wait_for_alloc = 0;
3866                 goto again;
3867         }
3868
3869         trans->allocating_chunk = true;
3870
3871         /*
3872          * If we have mixed data/metadata chunks we want to make sure we keep
3873          * allocating mixed chunks instead of individual chunks.
3874          */
3875         if (btrfs_mixed_space_info(space_info))
3876                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3877
3878         /*
3879          * if we're doing a data chunk, go ahead and make sure that
3880          * we keep a reasonable number of metadata chunks allocated in the
3881          * FS as well.
3882          */
3883         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3884                 fs_info->data_chunk_allocations++;
3885                 if (!(fs_info->data_chunk_allocations %
3886                       fs_info->metadata_ratio))
3887                         force_metadata_allocation(fs_info);
3888         }
3889
3890         /*
3891          * Check if we have enough space in SYSTEM chunk because we may need
3892          * to update devices.
3893          */
3894         check_system_chunk(trans, extent_root, flags);
3895
3896         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3897         trans->allocating_chunk = false;
3898
3899         spin_lock(&space_info->lock);
3900         if (ret < 0 && ret != -ENOSPC)
3901                 goto out;
3902         if (ret)
3903                 space_info->full = 1;
3904         else
3905                 ret = 1;
3906
3907         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3908 out:
3909         space_info->chunk_alloc = 0;
3910         spin_unlock(&space_info->lock);
3911         mutex_unlock(&fs_info->chunk_mutex);
3912         return ret;
3913 }
3914
3915 static int can_overcommit(struct btrfs_root *root,
3916                           struct btrfs_space_info *space_info, u64 bytes,
3917                           enum btrfs_reserve_flush_enum flush)
3918 {
3919         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3920         u64 profile = btrfs_get_alloc_profile(root, 0);
3921         u64 space_size;
3922         u64 avail;
3923         u64 used;
3924
3925         used = space_info->bytes_used + space_info->bytes_reserved +
3926                 space_info->bytes_pinned + space_info->bytes_readonly;
3927
3928         /*
3929          * We only want to allow over committing if we have lots of actual space
3930          * free, but if we don't have enough space to handle the global reserve
3931          * space then we could end up having a real enospc problem when trying
3932          * to allocate a chunk or some other such important allocation.
3933          */
3934         spin_lock(&global_rsv->lock);
3935         space_size = calc_global_rsv_need_space(global_rsv);
3936         spin_unlock(&global_rsv->lock);
3937         if (used + space_size >= space_info->total_bytes)
3938                 return 0;
3939
3940         used += space_info->bytes_may_use;
3941
3942         spin_lock(&root->fs_info->free_chunk_lock);
3943         avail = root->fs_info->free_chunk_space;
3944         spin_unlock(&root->fs_info->free_chunk_lock);
3945
3946         /*
3947          * If we have dup, raid1 or raid10 then only half of the free
3948          * space is actually useable.  For raid56, the space info used
3949          * doesn't include the parity drive, so we don't have to
3950          * change the math
3951          */
3952         if (profile & (BTRFS_BLOCK_GROUP_DUP |
3953                        BTRFS_BLOCK_GROUP_RAID1 |
3954                        BTRFS_BLOCK_GROUP_RAID10))
3955                 avail >>= 1;
3956
3957         /*
3958          * If we aren't flushing all things, let us overcommit up to
3959          * 1/2th of the space. If we can flush, don't let us overcommit
3960          * too much, let it overcommit up to 1/8 of the space.
3961          */
3962         if (flush == BTRFS_RESERVE_FLUSH_ALL)
3963                 avail >>= 3;
3964         else
3965                 avail >>= 1;
3966
3967         if (used + bytes < space_info->total_bytes + avail)
3968                 return 1;
3969         return 0;
3970 }
3971
3972 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3973                                          unsigned long nr_pages)
3974 {
3975         struct super_block *sb = root->fs_info->sb;
3976
3977         if (down_read_trylock(&sb->s_umount)) {
3978                 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3979                 up_read(&sb->s_umount);
3980         } else {
3981                 /*
3982                  * We needn't worry the filesystem going from r/w to r/o though
3983                  * we don't acquire ->s_umount mutex, because the filesystem
3984                  * should guarantee the delalloc inodes list be empty after
3985                  * the filesystem is readonly(all dirty pages are written to
3986                  * the disk).
3987                  */
3988                 btrfs_start_delalloc_roots(root->fs_info, 0);
3989                 if (!current->journal_info)
3990                         btrfs_wait_ordered_roots(root->fs_info, -1);
3991         }
3992 }
3993
3994 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
3995 {
3996         u64 bytes;
3997         int nr;
3998
3999         bytes = btrfs_calc_trans_metadata_size(root, 1);
4000         nr = (int)div64_u64(to_reclaim, bytes);
4001         if (!nr)
4002                 nr = 1;
4003         return nr;
4004 }
4005
4006 #define EXTENT_SIZE_PER_ITEM    (256 * 1024)
4007
4008 /*
4009  * shrink metadata reservation for delalloc
4010  */
4011 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4012                             bool wait_ordered)
4013 {
4014         struct btrfs_block_rsv *block_rsv;
4015         struct btrfs_space_info *space_info;
4016         struct btrfs_trans_handle *trans;
4017         u64 delalloc_bytes;
4018         u64 max_reclaim;
4019         long time_left;
4020         unsigned long nr_pages;
4021         int loops;
4022         int items;
4023         enum btrfs_reserve_flush_enum flush;
4024
4025         /* Calc the number of the pages we need flush for space reservation */
4026         items = calc_reclaim_items_nr(root, to_reclaim);
4027         to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4028
4029         trans = (struct btrfs_trans_handle *)current->journal_info;
4030         block_rsv = &root->fs_info->delalloc_block_rsv;
4031         space_info = block_rsv->space_info;
4032
4033         delalloc_bytes = percpu_counter_sum_positive(
4034                                                 &root->fs_info->delalloc_bytes);
4035         if (delalloc_bytes == 0) {
4036                 if (trans)
4037                         return;
4038                 if (wait_ordered)
4039                         btrfs_wait_ordered_roots(root->fs_info, items);
4040                 return;
4041         }
4042
4043         loops = 0;
4044         while (delalloc_bytes && loops < 3) {
4045                 max_reclaim = min(delalloc_bytes, to_reclaim);
4046                 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4047                 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4048                 /*
4049                  * We need to wait for the async pages to actually start before
4050                  * we do anything.
4051                  */
4052                 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4053                 if (!max_reclaim)
4054                         goto skip_async;
4055
4056                 if (max_reclaim <= nr_pages)
4057                         max_reclaim = 0;
4058                 else
4059                         max_reclaim -= nr_pages;
4060
4061                 wait_event(root->fs_info->async_submit_wait,
4062                            atomic_read(&root->fs_info->async_delalloc_pages) <=
4063                            (int)max_reclaim);
4064 skip_async:
4065                 if (!trans)
4066                         flush = BTRFS_RESERVE_FLUSH_ALL;
4067                 else
4068                         flush = BTRFS_RESERVE_NO_FLUSH;
4069                 spin_lock(&space_info->lock);
4070                 if (can_overcommit(root, space_info, orig, flush)) {
4071                         spin_unlock(&space_info->lock);
4072                         break;
4073                 }
4074                 spin_unlock(&space_info->lock);
4075
4076                 loops++;
4077                 if (wait_ordered && !trans) {
4078                         btrfs_wait_ordered_roots(root->fs_info, items);
4079                 } else {
4080                         time_left = schedule_timeout_killable(1);
4081                         if (time_left)
4082                                 break;
4083                 }
4084                 delalloc_bytes = percpu_counter_sum_positive(
4085                                                 &root->fs_info->delalloc_bytes);
4086         }
4087 }
4088
4089 /**
4090  * maybe_commit_transaction - possibly commit the transaction if its ok to
4091  * @root - the root we're allocating for
4092  * @bytes - the number of bytes we want to reserve
4093  * @force - force the commit
4094  *
4095  * This will check to make sure that committing the transaction will actually
4096  * get us somewhere and then commit the transaction if it does.  Otherwise it
4097  * will return -ENOSPC.
4098  */
4099 static int may_commit_transaction(struct btrfs_root *root,
4100                                   struct btrfs_space_info *space_info,
4101                                   u64 bytes, int force)
4102 {
4103         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4104         struct btrfs_trans_handle *trans;
4105
4106         trans = (struct btrfs_trans_handle *)current->journal_info;
4107         if (trans)
4108                 return -EAGAIN;
4109
4110         if (force)
4111                 goto commit;
4112
4113         /* See if there is enough pinned space to make this reservation */
4114         spin_lock(&space_info->lock);
4115         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4116                                    bytes) >= 0) {
4117                 spin_unlock(&space_info->lock);
4118                 goto commit;
4119         }
4120         spin_unlock(&space_info->lock);
4121
4122         /*
4123          * See if there is some space in the delayed insertion reservation for
4124          * this reservation.
4125          */
4126         if (space_info != delayed_rsv->space_info)
4127                 return -ENOSPC;
4128
4129         spin_lock(&space_info->lock);
4130         spin_lock(&delayed_rsv->lock);
4131         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4132                                    bytes - delayed_rsv->size) >= 0) {
4133                 spin_unlock(&delayed_rsv->lock);
4134                 spin_unlock(&space_info->lock);
4135                 return -ENOSPC;
4136         }
4137         spin_unlock(&delayed_rsv->lock);
4138         spin_unlock(&space_info->lock);
4139
4140 commit:
4141         trans = btrfs_join_transaction(root);
4142         if (IS_ERR(trans))
4143                 return -ENOSPC;
4144
4145         return btrfs_commit_transaction(trans, root);
4146 }
4147
4148 enum flush_state {
4149         FLUSH_DELAYED_ITEMS_NR  =       1,
4150         FLUSH_DELAYED_ITEMS     =       2,
4151         FLUSH_DELALLOC          =       3,
4152         FLUSH_DELALLOC_WAIT     =       4,
4153         ALLOC_CHUNK             =       5,
4154         COMMIT_TRANS            =       6,
4155 };
4156
4157 static int flush_space(struct btrfs_root *root,
4158                        struct btrfs_space_info *space_info, u64 num_bytes,
4159                        u64 orig_bytes, int state)
4160 {
4161         struct btrfs_trans_handle *trans;
4162         int nr;
4163         int ret = 0;
4164
4165         switch (state) {
4166         case FLUSH_DELAYED_ITEMS_NR:
4167         case FLUSH_DELAYED_ITEMS:
4168                 if (state == FLUSH_DELAYED_ITEMS_NR)
4169                         nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4170                 else
4171                         nr = -1;
4172
4173                 trans = btrfs_join_transaction(root);
4174                 if (IS_ERR(trans)) {
4175                         ret = PTR_ERR(trans);
4176                         break;
4177                 }
4178                 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4179                 btrfs_end_transaction(trans, root);
4180                 break;
4181         case FLUSH_DELALLOC:
4182         case FLUSH_DELALLOC_WAIT:
4183                 shrink_delalloc(root, num_bytes, orig_bytes,
4184                                 state == FLUSH_DELALLOC_WAIT);
4185                 break;
4186         case ALLOC_CHUNK:
4187                 trans = btrfs_join_transaction(root);
4188                 if (IS_ERR(trans)) {
4189                         ret = PTR_ERR(trans);
4190                         break;
4191                 }
4192                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4193                                      btrfs_get_alloc_profile(root, 0),
4194                                      CHUNK_ALLOC_NO_FORCE);
4195                 btrfs_end_transaction(trans, root);
4196                 if (ret == -ENOSPC)
4197                         ret = 0;
4198                 break;
4199         case COMMIT_TRANS:
4200                 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4201                 break;
4202         default:
4203                 ret = -ENOSPC;
4204                 break;
4205         }
4206
4207         return ret;
4208 }
4209 /**
4210  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4211  * @root - the root we're allocating for
4212  * @block_rsv - the block_rsv we're allocating for
4213  * @orig_bytes - the number of bytes we want
4214  * @flush - whether or not we can flush to make our reservation
4215  *
4216  * This will reserve orgi_bytes number of bytes from the space info associated
4217  * with the block_rsv.  If there is not enough space it will make an attempt to
4218  * flush out space to make room.  It will do this by flushing delalloc if
4219  * possible or committing the transaction.  If flush is 0 then no attempts to
4220  * regain reservations will be made and this will fail if there is not enough
4221  * space already.
4222  */
4223 static int reserve_metadata_bytes(struct btrfs_root *root,
4224                                   struct btrfs_block_rsv *block_rsv,
4225                                   u64 orig_bytes,
4226                                   enum btrfs_reserve_flush_enum flush)
4227 {
4228         struct btrfs_space_info *space_info = block_rsv->space_info;
4229         u64 used;
4230         u64 num_bytes = orig_bytes;
4231         int flush_state = FLUSH_DELAYED_ITEMS_NR;
4232         int ret = 0;
4233         bool flushing = false;
4234
4235 again:
4236         ret = 0;
4237         spin_lock(&space_info->lock);
4238         /*
4239          * We only want to wait if somebody other than us is flushing and we
4240          * are actually allowed to flush all things.
4241          */
4242         while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4243                space_info->flush) {
4244                 spin_unlock(&space_info->lock);
4245                 /*
4246                  * If we have a trans handle we can't wait because the flusher
4247                  * may have to commit the transaction, which would mean we would
4248                  * deadlock since we are waiting for the flusher to finish, but
4249                  * hold the current transaction open.
4250                  */
4251                 if (current->journal_info)
4252                         return -EAGAIN;
4253                 ret = wait_event_killable(space_info->wait, !space_info->flush);
4254                 /* Must have been killed, return */
4255                 if (ret)
4256                         return -EINTR;
4257
4258                 spin_lock(&space_info->lock);
4259         }
4260
4261         ret = -ENOSPC;
4262         used = space_info->bytes_used + space_info->bytes_reserved +
4263                 space_info->bytes_pinned + space_info->bytes_readonly +
4264                 space_info->bytes_may_use;
4265
4266         /*
4267          * The idea here is that we've not already over-reserved the block group
4268          * then we can go ahead and save our reservation first and then start
4269          * flushing if we need to.  Otherwise if we've already overcommitted
4270          * lets start flushing stuff first and then come back and try to make
4271          * our reservation.
4272          */
4273         if (used <= space_info->total_bytes) {
4274                 if (used + orig_bytes <= space_info->total_bytes) {
4275                         space_info->bytes_may_use += orig_bytes;
4276                         trace_btrfs_space_reservation(root->fs_info,
4277                                 "space_info", space_info->flags, orig_bytes, 1);
4278                         ret = 0;
4279                 } else {
4280                         /*
4281                          * Ok set num_bytes to orig_bytes since we aren't
4282                          * overocmmitted, this way we only try and reclaim what
4283                          * we need.
4284                          */
4285                         num_bytes = orig_bytes;
4286                 }
4287         } else {
4288                 /*
4289                  * Ok we're over committed, set num_bytes to the overcommitted
4290                  * amount plus the amount of bytes that we need for this
4291                  * reservation.
4292                  */
4293                 num_bytes = used - space_info->total_bytes +
4294                         (orig_bytes * 2);
4295         }
4296
4297         if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4298                 space_info->bytes_may_use += orig_bytes;
4299                 trace_btrfs_space_reservation(root->fs_info, "space_info",
4300                                               space_info->flags, orig_bytes,
4301                                               1);
4302                 ret = 0;
4303         }
4304
4305         /*
4306          * Couldn't make our reservation, save our place so while we're trying
4307          * to reclaim space we can actually use it instead of somebody else
4308          * stealing it from us.
4309          *
4310          * We make the other tasks wait for the flush only when we can flush
4311          * all things.
4312          */
4313         if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4314                 flushing = true;
4315                 space_info->flush = 1;
4316         }
4317
4318         spin_unlock(&space_info->lock);
4319
4320         if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4321                 goto out;
4322
4323         ret = flush_space(root, space_info, num_bytes, orig_bytes,
4324                           flush_state);
4325         flush_state++;
4326
4327         /*
4328          * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4329          * would happen. So skip delalloc flush.
4330          */
4331         if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4332             (flush_state == FLUSH_DELALLOC ||
4333              flush_state == FLUSH_DELALLOC_WAIT))
4334                 flush_state = ALLOC_CHUNK;
4335
4336         if (!ret)
4337                 goto again;
4338         else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4339                  flush_state < COMMIT_TRANS)
4340                 goto again;
4341         else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4342                  flush_state <= COMMIT_TRANS)
4343                 goto again;
4344
4345 out:
4346         if (ret == -ENOSPC &&
4347             unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4348                 struct btrfs_block_rsv *global_rsv =
4349                         &root->fs_info->global_block_rsv;
4350
4351                 if (block_rsv != global_rsv &&
4352                     !block_rsv_use_bytes(global_rsv, orig_bytes))
4353                         ret = 0;
4354         }
4355         if (ret == -ENOSPC)
4356                 trace_btrfs_space_reservation(root->fs_info,
4357                                               "space_info:enospc",
4358                                               space_info->flags, orig_bytes, 1);
4359         if (flushing) {
4360                 spin_lock(&space_info->lock);
4361                 space_info->flush = 0;
4362                 wake_up_all(&space_info->wait);
4363                 spin_unlock(&space_info->lock);
4364         }
4365         return ret;
4366 }
4367
4368 static struct btrfs_block_rsv *get_block_rsv(
4369                                         const struct btrfs_trans_handle *trans,
4370                                         const struct btrfs_root *root)
4371 {
4372         struct btrfs_block_rsv *block_rsv = NULL;
4373
4374         if (root->ref_cows)
4375                 block_rsv = trans->block_rsv;
4376
4377         if (root == root->fs_info->csum_root && trans->adding_csums)
4378                 block_rsv = trans->block_rsv;
4379
4380         if (root == root->fs_info->uuid_root)
4381                 block_rsv = trans->block_rsv;
4382
4383         if (!block_rsv)
4384                 block_rsv = root->block_rsv;
4385
4386         if (!block_rsv)
4387                 block_rsv = &root->fs_info->empty_block_rsv;
4388
4389         return block_rsv;
4390 }
4391
4392 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4393                                u64 num_bytes)
4394 {
4395         int ret = -ENOSPC;
4396         spin_lock(&block_rsv->lock);
4397         if (block_rsv->reserved >= num_bytes) {
4398                 block_rsv->reserved -= num_bytes;
4399                 if (block_rsv->reserved < block_rsv->size)
4400                         block_rsv->full = 0;
4401                 ret = 0;
4402         }
4403         spin_unlock(&block_rsv->lock);
4404         return ret;
4405 }
4406
4407 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4408                                 u64 num_bytes, int update_size)
4409 {
4410         spin_lock(&block_rsv->lock);
4411         block_rsv->reserved += num_bytes;
4412         if (update_size)
4413                 block_rsv->size += num_bytes;
4414         else if (block_rsv->reserved >= block_rsv->size)
4415                 block_rsv->full = 1;
4416         spin_unlock(&block_rsv->lock);
4417 }
4418
4419 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4420                              struct btrfs_block_rsv *dest, u64 num_bytes,
4421                              int min_factor)
4422 {
4423         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4424         u64 min_bytes;
4425
4426         if (global_rsv->space_info != dest->space_info)
4427                 return -ENOSPC;
4428
4429         spin_lock(&global_rsv->lock);
4430         min_bytes = div_factor(global_rsv->size, min_factor);
4431         if (global_rsv->reserved < min_bytes + num_bytes) {
4432                 spin_unlock(&global_rsv->lock);
4433                 return -ENOSPC;
4434         }
4435         global_rsv->reserved -= num_bytes;
4436         if (global_rsv->reserved < global_rsv->size)
4437                 global_rsv->full = 0;
4438         spin_unlock(&global_rsv->lock);
4439
4440         block_rsv_add_bytes(dest, num_bytes, 1);
4441         return 0;
4442 }
4443
4444 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4445                                     struct btrfs_block_rsv *block_rsv,
4446                                     struct btrfs_block_rsv *dest, u64 num_bytes)
4447 {
4448         struct btrfs_space_info *space_info = block_rsv->space_info;
4449
4450         spin_lock(&block_rsv->lock);
4451         if (num_bytes == (u64)-1)
4452                 num_bytes = block_rsv->size;
4453         block_rsv->size -= num_bytes;
4454         if (block_rsv->reserved >= block_rsv->size) {
4455                 num_bytes = block_rsv->reserved - block_rsv->size;
4456                 block_rsv->reserved = block_rsv->size;
4457                 block_rsv->full = 1;
4458         } else {
4459                 num_bytes = 0;
4460         }
4461         spin_unlock(&block_rsv->lock);
4462
4463         if (num_bytes > 0) {
4464                 if (dest) {
4465                         spin_lock(&dest->lock);
4466                         if (!dest->full) {
4467                                 u64 bytes_to_add;
4468
4469                                 bytes_to_add = dest->size - dest->reserved;
4470                                 bytes_to_add = min(num_bytes, bytes_to_add);
4471                                 dest->reserved += bytes_to_add;
4472                                 if (dest->reserved >= dest->size)
4473                                         dest->full = 1;
4474                                 num_bytes -= bytes_to_add;
4475                         }
4476                         spin_unlock(&dest->lock);
4477                 }
4478                 if (num_bytes) {
4479                         spin_lock(&space_info->lock);
4480                         space_info->bytes_may_use -= num_bytes;
4481                         trace_btrfs_space_reservation(fs_info, "space_info",
4482                                         space_info->flags, num_bytes, 0);
4483                         spin_unlock(&space_info->lock);
4484                 }
4485         }
4486 }
4487
4488 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4489                                    struct btrfs_block_rsv *dst, u64 num_bytes)
4490 {
4491         int ret;
4492
4493         ret = block_rsv_use_bytes(src, num_bytes);
4494         if (ret)
4495                 return ret;
4496
4497         block_rsv_add_bytes(dst, num_bytes, 1);
4498         return 0;
4499 }
4500
4501 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4502 {
4503         memset(rsv, 0, sizeof(*rsv));
4504         spin_lock_init(&rsv->lock);
4505         rsv->type = type;
4506 }
4507
4508 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4509                                               unsigned short type)
4510 {
4511         struct btrfs_block_rsv *block_rsv;
4512         struct btrfs_fs_info *fs_info = root->fs_info;
4513
4514         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4515         if (!block_rsv)
4516                 return NULL;
4517
4518         btrfs_init_block_rsv(block_rsv, type);
4519         block_rsv->space_info = __find_space_info(fs_info,
4520                                                   BTRFS_BLOCK_GROUP_METADATA);
4521         return block_rsv;
4522 }
4523
4524 void btrfs_free_block_rsv(struct btrfs_root *root,
4525                           struct btrfs_block_rsv *rsv)
4526 {
4527         if (!rsv)
4528                 return;
4529         btrfs_block_rsv_release(root, rsv, (u64)-1);
4530         kfree(rsv);
4531 }
4532
4533 int btrfs_block_rsv_add(struct btrfs_root *root,
4534                         struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4535                         enum btrfs_reserve_flush_enum flush)
4536 {
4537         int ret;
4538
4539         if (num_bytes == 0)
4540                 return 0;
4541
4542         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4543         if (!ret) {
4544                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4545                 return 0;
4546         }
4547
4548         return ret;
4549 }
4550
4551 int btrfs_block_rsv_check(struct btrfs_root *root,
4552                           struct btrfs_block_rsv *block_rsv, int min_factor)
4553 {
4554         u64 num_bytes = 0;
4555         int ret = -ENOSPC;
4556
4557         if (!block_rsv)
4558                 return 0;
4559
4560         spin_lock(&block_rsv->lock);
4561         num_bytes = div_factor(block_rsv->size, min_factor);
4562         if (block_rsv->reserved >= num_bytes)
4563                 ret = 0;
4564         spin_unlock(&block_rsv->lock);
4565
4566         return ret;
4567 }
4568
4569 int btrfs_block_rsv_refill(struct btrfs_root *root,
4570                            struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4571                            enum btrfs_reserve_flush_enum flush)
4572 {
4573         u64 num_bytes = 0;
4574         int ret = -ENOSPC;
4575
4576         if (!block_rsv)
4577                 return 0;
4578
4579         spin_lock(&block_rsv->lock);
4580         num_bytes = min_reserved;
4581         if (block_rsv->reserved >= num_bytes)
4582                 ret = 0;
4583         else
4584                 num_bytes -= block_rsv->reserved;
4585         spin_unlock(&block_rsv->lock);
4586
4587         if (!ret)
4588                 return 0;
4589
4590         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4591         if (!ret) {
4592                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4593                 return 0;
4594         }
4595
4596         return ret;
4597 }
4598
4599 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4600                             struct btrfs_block_rsv *dst_rsv,
4601                             u64 num_bytes)
4602 {
4603         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4604 }
4605
4606 void btrfs_block_rsv_release(struct btrfs_root *root,
4607                              struct btrfs_block_rsv *block_rsv,
4608                              u64 num_bytes)
4609 {
4610         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4611         if (global_rsv == block_rsv ||
4612             block_rsv->space_info != global_rsv->space_info)
4613                 global_rsv = NULL;
4614         block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4615                                 num_bytes);
4616 }
4617
4618 /*
4619  * helper to calculate size of global block reservation.
4620  * the desired value is sum of space used by extent tree,
4621  * checksum tree and root tree
4622  */
4623 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4624 {
4625         struct btrfs_space_info *sinfo;
4626         u64 num_bytes;
4627         u64 meta_used;
4628         u64 data_used;
4629         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4630
4631         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4632         spin_lock(&sinfo->lock);
4633         data_used = sinfo->bytes_used;
4634         spin_unlock(&sinfo->lock);
4635
4636         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4637         spin_lock(&sinfo->lock);
4638         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4639                 data_used = 0;
4640         meta_used = sinfo->bytes_used;
4641         spin_unlock(&sinfo->lock);
4642
4643         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4644                     csum_size * 2;
4645         num_bytes += div64_u64(data_used + meta_used, 50);
4646
4647         if (num_bytes * 3 > meta_used)
4648                 num_bytes = div64_u64(meta_used, 3);
4649
4650         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4651 }
4652
4653 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4654 {
4655         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4656         struct btrfs_space_info *sinfo = block_rsv->space_info;
4657         u64 num_bytes;
4658
4659         num_bytes = calc_global_metadata_size(fs_info);
4660
4661         spin_lock(&sinfo->lock);
4662         spin_lock(&block_rsv->lock);
4663
4664         block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4665
4666         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4667                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4668                     sinfo->bytes_may_use;
4669
4670         if (sinfo->total_bytes > num_bytes) {
4671                 num_bytes = sinfo->total_bytes - num_bytes;
4672                 block_rsv->reserved += num_bytes;
4673                 sinfo->bytes_may_use += num_bytes;
4674                 trace_btrfs_space_reservation(fs_info, "space_info",
4675                                       sinfo->flags, num_bytes, 1);
4676         }
4677
4678         if (block_rsv->reserved >= block_rsv->size) {
4679                 num_bytes = block_rsv->reserved - block_rsv->size;
4680                 sinfo->bytes_may_use -= num_bytes;
4681                 trace_btrfs_space_reservation(fs_info, "space_info",
4682                                       sinfo->flags, num_bytes, 0);
4683                 block_rsv->reserved = block_rsv->size;
4684                 block_rsv->full = 1;
4685         }
4686
4687         spin_unlock(&block_rsv->lock);
4688         spin_unlock(&sinfo->lock);
4689 }
4690
4691 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4692 {
4693         struct btrfs_space_info *space_info;
4694
4695         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4696         fs_info->chunk_block_rsv.space_info = space_info;
4697
4698         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4699         fs_info->global_block_rsv.space_info = space_info;
4700         fs_info->delalloc_block_rsv.space_info = space_info;
4701         fs_info->trans_block_rsv.space_info = space_info;
4702         fs_info->empty_block_rsv.space_info = space_info;
4703         fs_info->delayed_block_rsv.space_info = space_info;
4704
4705         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4706         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4707         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4708         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4709         if (fs_info->quota_root)
4710                 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4711         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4712
4713         update_global_block_rsv(fs_info);
4714 }
4715
4716 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4717 {
4718         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4719                                 (u64)-1);
4720         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4721         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4722         WARN_ON(fs_info->trans_block_rsv.size > 0);
4723         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4724         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4725         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4726         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4727         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4728 }
4729
4730 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4731                                   struct btrfs_root *root)
4732 {
4733         if (!trans->block_rsv)
4734                 return;
4735
4736         if (!trans->bytes_reserved)
4737                 return;
4738
4739         trace_btrfs_space_reservation(root->fs_info, "transaction",
4740                                       trans->transid, trans->bytes_reserved, 0);
4741         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4742         trans->bytes_reserved = 0;
4743 }
4744
4745 /* Can only return 0 or -ENOSPC */
4746 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4747                                   struct inode *inode)
4748 {
4749         struct btrfs_root *root = BTRFS_I(inode)->root;
4750         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4751         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4752
4753         /*
4754          * We need to hold space in order to delete our orphan item once we've
4755          * added it, so this takes the reservation so we can release it later
4756          * when we are truly done with the orphan item.
4757          */
4758         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4759         trace_btrfs_space_reservation(root->fs_info, "orphan",
4760                                       btrfs_ino(inode), num_bytes, 1);
4761         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4762 }
4763
4764 void btrfs_orphan_release_metadata(struct inode *inode)
4765 {
4766         struct btrfs_root *root = BTRFS_I(inode)->root;
4767         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4768         trace_btrfs_space_reservation(root->fs_info, "orphan",
4769                                       btrfs_ino(inode), num_bytes, 0);
4770         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4771 }
4772
4773 /*
4774  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4775  * root: the root of the parent directory
4776  * rsv: block reservation
4777  * items: the number of items that we need do reservation
4778  * qgroup_reserved: used to return the reserved size in qgroup
4779  *
4780  * This function is used to reserve the space for snapshot/subvolume
4781  * creation and deletion. Those operations are different with the
4782  * common file/directory operations, they change two fs/file trees
4783  * and root tree, the number of items that the qgroup reserves is
4784  * different with the free space reservation. So we can not use
4785  * the space reseravtion mechanism in start_transaction().
4786  */
4787 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4788                                      struct btrfs_block_rsv *rsv,
4789                                      int items,
4790                                      u64 *qgroup_reserved,
4791                                      bool use_global_rsv)
4792 {
4793         u64 num_bytes;
4794         int ret;
4795         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4796
4797         if (root->fs_info->quota_enabled) {
4798                 /* One for parent inode, two for dir entries */
4799                 num_bytes = 3 * root->leafsize;
4800                 ret = btrfs_qgroup_reserve(root, num_bytes);
4801                 if (ret)
4802                         return ret;
4803         } else {
4804                 num_bytes = 0;
4805         }
4806
4807         *qgroup_reserved = num_bytes;
4808
4809         num_bytes = btrfs_calc_trans_metadata_size(root, items);
4810         rsv->space_info = __find_space_info(root->fs_info,
4811                                             BTRFS_BLOCK_GROUP_METADATA);
4812         ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4813                                   BTRFS_RESERVE_FLUSH_ALL);
4814
4815         if (ret == -ENOSPC && use_global_rsv)
4816                 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4817
4818         if (ret) {
4819                 if (*qgroup_reserved)
4820                         btrfs_qgroup_free(root, *qgroup_reserved);
4821         }
4822
4823         return ret;
4824 }
4825
4826 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4827                                       struct btrfs_block_rsv *rsv,
4828                                       u64 qgroup_reserved)
4829 {
4830         btrfs_block_rsv_release(root, rsv, (u64)-1);
4831         if (qgroup_reserved)
4832                 btrfs_qgroup_free(root, qgroup_reserved);
4833 }
4834
4835 /**
4836  * drop_outstanding_extent - drop an outstanding extent
4837  * @inode: the inode we're dropping the extent for
4838  *
4839  * This is called when we are freeing up an outstanding extent, either called
4840  * after an error or after an extent is written.  This will return the number of
4841  * reserved extents that need to be freed.  This must be called with
4842  * BTRFS_I(inode)->lock held.
4843  */
4844 static unsigned drop_outstanding_extent(struct inode *inode)
4845 {
4846         unsigned drop_inode_space = 0;
4847         unsigned dropped_extents = 0;
4848
4849         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4850         BTRFS_I(inode)->outstanding_extents--;
4851
4852         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4853             test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4854                                &BTRFS_I(inode)->runtime_flags))
4855                 drop_inode_space = 1;
4856
4857         /*
4858          * If we have more or the same amount of outsanding extents than we have
4859          * reserved then we need to leave the reserved extents count alone.
4860          */
4861         if (BTRFS_I(inode)->outstanding_extents >=
4862             BTRFS_I(inode)->reserved_extents)
4863                 return drop_inode_space;
4864
4865         dropped_extents = BTRFS_I(inode)->reserved_extents -
4866                 BTRFS_I(inode)->outstanding_extents;
4867         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4868         return dropped_extents + drop_inode_space;
4869 }
4870
4871 /**
4872  * calc_csum_metadata_size - return the amount of metada space that must be
4873  *      reserved/free'd for the given bytes.
4874  * @inode: the inode we're manipulating
4875  * @num_bytes: the number of bytes in question
4876  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4877  *
4878  * This adjusts the number of csum_bytes in the inode and then returns the
4879  * correct amount of metadata that must either be reserved or freed.  We
4880  * calculate how many checksums we can fit into one leaf and then divide the
4881  * number of bytes that will need to be checksumed by this value to figure out
4882  * how many checksums will be required.  If we are adding bytes then the number
4883  * may go up and we will return the number of additional bytes that must be
4884  * reserved.  If it is going down we will return the number of bytes that must
4885  * be freed.
4886  *
4887  * This must be called with BTRFS_I(inode)->lock held.
4888  */
4889 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4890                                    int reserve)
4891 {
4892         struct btrfs_root *root = BTRFS_I(inode)->root;
4893         u64 csum_size;
4894         int num_csums_per_leaf;
4895         int num_csums;
4896         int old_csums;
4897
4898         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4899             BTRFS_I(inode)->csum_bytes == 0)
4900                 return 0;
4901
4902         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4903         if (reserve)
4904                 BTRFS_I(inode)->csum_bytes += num_bytes;
4905         else
4906                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4907         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4908         num_csums_per_leaf = (int)div64_u64(csum_size,
4909                                             sizeof(struct btrfs_csum_item) +
4910                                             sizeof(struct btrfs_disk_key));
4911         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4912         num_csums = num_csums + num_csums_per_leaf - 1;
4913         num_csums = num_csums / num_csums_per_leaf;
4914
4915         old_csums = old_csums + num_csums_per_leaf - 1;
4916         old_csums = old_csums / num_csums_per_leaf;
4917
4918         /* No change, no need to reserve more */
4919         if (old_csums == num_csums)
4920                 return 0;
4921
4922         if (reserve)
4923                 return btrfs_calc_trans_metadata_size(root,
4924                                                       num_csums - old_csums);
4925
4926         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4927 }
4928
4929 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4930 {
4931         struct btrfs_root *root = BTRFS_I(inode)->root;
4932         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4933         u64 to_reserve = 0;
4934         u64 csum_bytes;
4935         unsigned nr_extents = 0;
4936         int extra_reserve = 0;
4937         enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4938         int ret = 0;
4939         bool delalloc_lock = true;
4940         u64 to_free = 0;
4941         unsigned dropped;
4942
4943         /* If we are a free space inode we need to not flush since we will be in
4944          * the middle of a transaction commit.  We also don't need the delalloc
4945          * mutex since we won't race with anybody.  We need this mostly to make
4946          * lockdep shut its filthy mouth.
4947          */
4948         if (btrfs_is_free_space_inode(inode)) {
4949                 flush = BTRFS_RESERVE_NO_FLUSH;
4950                 delalloc_lock = false;
4951         }
4952
4953         if (flush != BTRFS_RESERVE_NO_FLUSH &&
4954             btrfs_transaction_in_commit(root->fs_info))
4955                 schedule_timeout(1);
4956
4957         if (delalloc_lock)
4958                 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4959
4960         num_bytes = ALIGN(num_bytes, root->sectorsize);
4961
4962         spin_lock(&BTRFS_I(inode)->lock);
4963         BTRFS_I(inode)->outstanding_extents++;
4964
4965         if (BTRFS_I(inode)->outstanding_extents >
4966             BTRFS_I(inode)->reserved_extents)
4967                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4968                         BTRFS_I(inode)->reserved_extents;
4969
4970         /*
4971          * Add an item to reserve for updating the inode when we complete the
4972          * delalloc io.
4973          */
4974         if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4975                       &BTRFS_I(inode)->runtime_flags)) {
4976                 nr_extents++;
4977                 extra_reserve = 1;
4978         }
4979
4980         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4981         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4982         csum_bytes = BTRFS_I(inode)->csum_bytes;
4983         spin_unlock(&BTRFS_I(inode)->lock);
4984
4985         if (root->fs_info->quota_enabled) {
4986                 ret = btrfs_qgroup_reserve(root, num_bytes +
4987                                            nr_extents * root->leafsize);
4988                 if (ret)
4989                         goto out_fail;
4990         }
4991
4992         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4993         if (unlikely(ret)) {
4994                 if (root->fs_info->quota_enabled)
4995                         btrfs_qgroup_free(root, num_bytes +
4996                                                 nr_extents * root->leafsize);
4997                 goto out_fail;
4998         }
4999
5000         spin_lock(&BTRFS_I(inode)->lock);
5001         if (extra_reserve) {
5002                 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5003                         &BTRFS_I(inode)->runtime_flags);
5004                 nr_extents--;
5005         }
5006         BTRFS_I(inode)->reserved_extents += nr_extents;
5007         spin_unlock(&BTRFS_I(inode)->lock);
5008
5009         if (delalloc_lock)
5010                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5011
5012         if (to_reserve)
5013                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5014                                               btrfs_ino(inode), to_reserve, 1);
5015         block_rsv_add_bytes(block_rsv, to_reserve, 1);
5016
5017         return 0;
5018
5019 out_fail:
5020         spin_lock(&BTRFS_I(inode)->lock);
5021         dropped = drop_outstanding_extent(inode);
5022         /*
5023          * If the inodes csum_bytes is the same as the original
5024          * csum_bytes then we know we haven't raced with any free()ers
5025          * so we can just reduce our inodes csum bytes and carry on.
5026          */
5027         if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5028                 calc_csum_metadata_size(inode, num_bytes, 0);
5029         } else {
5030                 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5031                 u64 bytes;
5032
5033                 /*
5034                  * This is tricky, but first we need to figure out how much we
5035                  * free'd from any free-ers that occured during this
5036                  * reservation, so we reset ->csum_bytes to the csum_bytes
5037                  * before we dropped our lock, and then call the free for the
5038                  * number of bytes that were freed while we were trying our
5039                  * reservation.
5040                  */
5041                 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5042                 BTRFS_I(inode)->csum_bytes = csum_bytes;
5043                 to_free = calc_csum_metadata_size(inode, bytes, 0);
5044
5045
5046                 /*
5047                  * Now we need to see how much we would have freed had we not
5048                  * been making this reservation and our ->csum_bytes were not
5049                  * artificially inflated.
5050                  */
5051                 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5052                 bytes = csum_bytes - orig_csum_bytes;
5053                 bytes = calc_csum_metadata_size(inode, bytes, 0);
5054
5055                 /*
5056                  * Now reset ->csum_bytes to what it should be.  If bytes is
5057                  * more than to_free then we would have free'd more space had we
5058                  * not had an artificially high ->csum_bytes, so we need to free
5059                  * the remainder.  If bytes is the same or less then we don't
5060                  * need to do anything, the other free-ers did the correct
5061                  * thing.
5062                  */
5063                 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5064                 if (bytes > to_free)
5065                         to_free = bytes - to_free;
5066                 else
5067                         to_free = 0;
5068         }
5069         spin_unlock(&BTRFS_I(inode)->lock);
5070         if (dropped)
5071                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5072
5073         if (to_free) {
5074                 btrfs_block_rsv_release(root, block_rsv, to_free);
5075                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5076                                               btrfs_ino(inode), to_free, 0);
5077         }
5078         if (delalloc_lock)
5079                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5080         return ret;
5081 }
5082
5083 /**
5084  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5085  * @inode: the inode to release the reservation for
5086  * @num_bytes: the number of bytes we're releasing
5087  *
5088  * This will release the metadata reservation for an inode.  This can be called
5089  * once we complete IO for a given set of bytes to release their metadata
5090  * reservations.
5091  */
5092 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5093 {
5094         struct btrfs_root *root = BTRFS_I(inode)->root;
5095         u64 to_free = 0;
5096         unsigned dropped;
5097
5098         num_bytes = ALIGN(num_bytes, root->sectorsize);
5099         spin_lock(&BTRFS_I(inode)->lock);
5100         dropped = drop_outstanding_extent(inode);
5101
5102         if (num_bytes)
5103                 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5104         spin_unlock(&BTRFS_I(inode)->lock);
5105         if (dropped > 0)
5106                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5107
5108         trace_btrfs_space_reservation(root->fs_info, "delalloc",
5109                                       btrfs_ino(inode), to_free, 0);
5110         if (root->fs_info->quota_enabled) {
5111                 btrfs_qgroup_free(root, num_bytes +
5112                                         dropped * root->leafsize);
5113         }
5114
5115         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5116                                 to_free);
5117 }
5118
5119 /**
5120  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5121  * @inode: inode we're writing to
5122  * @num_bytes: the number of bytes we want to allocate
5123  *
5124  * This will do the following things
5125  *
5126  * o reserve space in the data space info for num_bytes
5127  * o reserve space in the metadata space info based on number of outstanding
5128  *   extents and how much csums will be needed
5129  * o add to the inodes ->delalloc_bytes
5130  * o add it to the fs_info's delalloc inodes list.
5131  *
5132  * This will return 0 for success and -ENOSPC if there is no space left.
5133  */
5134 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5135 {
5136         int ret;
5137
5138         ret = btrfs_check_data_free_space(inode, num_bytes);
5139         if (ret)
5140                 return ret;
5141
5142         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5143         if (ret) {
5144                 btrfs_free_reserved_data_space(inode, num_bytes);
5145                 return ret;
5146         }
5147
5148         return 0;
5149 }
5150
5151 /**
5152  * btrfs_delalloc_release_space - release data and metadata space for delalloc
5153  * @inode: inode we're releasing space for
5154  * @num_bytes: the number of bytes we want to free up
5155  *
5156  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
5157  * called in the case that we don't need the metadata AND data reservations
5158  * anymore.  So if there is an error or we insert an inline extent.
5159  *
5160  * This function will release the metadata space that was not used and will
5161  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5162  * list if there are no delalloc bytes left.
5163  */
5164 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5165 {
5166         btrfs_delalloc_release_metadata(inode, num_bytes);
5167         btrfs_free_reserved_data_space(inode, num_bytes);
5168 }
5169
5170 static int update_block_group(struct btrfs_root *root,
5171                               u64 bytenr, u64 num_bytes, int alloc)
5172 {
5173         struct btrfs_block_group_cache *cache = NULL;
5174         struct btrfs_fs_info *info = root->fs_info;
5175         u64 total = num_bytes;
5176         u64 old_val;
5177         u64 byte_in_group;
5178         int factor;
5179
5180         /* block accounting for super block */
5181         spin_lock(&info->delalloc_root_lock);
5182         old_val = btrfs_super_bytes_used(info->super_copy);
5183         if (alloc)
5184                 old_val += num_bytes;
5185         else
5186                 old_val -= num_bytes;
5187         btrfs_set_super_bytes_used(info->super_copy, old_val);
5188         spin_unlock(&info->delalloc_root_lock);
5189
5190         while (total) {
5191                 cache = btrfs_lookup_block_group(info, bytenr);
5192                 if (!cache)
5193                         return -ENOENT;
5194                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5195                                     BTRFS_BLOCK_GROUP_RAID1 |
5196                                     BTRFS_BLOCK_GROUP_RAID10))
5197                         factor = 2;
5198                 else
5199                         factor = 1;
5200                 /*
5201                  * If this block group has free space cache written out, we
5202                  * need to make sure to load it if we are removing space.  This
5203                  * is because we need the unpinning stage to actually add the
5204                  * space back to the block group, otherwise we will leak space.
5205                  */
5206                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5207                         cache_block_group(cache, 1);
5208
5209                 byte_in_group = bytenr - cache->key.objectid;
5210                 WARN_ON(byte_in_group > cache->key.offset);
5211
5212                 spin_lock(&cache->space_info->lock);
5213                 spin_lock(&cache->lock);
5214
5215                 if (btrfs_test_opt(root, SPACE_CACHE) &&
5216                     cache->disk_cache_state < BTRFS_DC_CLEAR)
5217                         cache->disk_cache_state = BTRFS_DC_CLEAR;
5218
5219                 cache->dirty = 1;
5220                 old_val = btrfs_block_group_used(&cache->item);
5221                 num_bytes = min(total, cache->key.offset - byte_in_group);
5222                 if (alloc) {
5223                         old_val += num_bytes;
5224                         btrfs_set_block_group_used(&cache->item, old_val);
5225                         cache->reserved -= num_bytes;
5226                         cache->space_info->bytes_reserved -= num_bytes;
5227                         cache->space_info->bytes_used += num_bytes;
5228                         cache->space_info->disk_used += num_bytes * factor;
5229                         spin_unlock(&cache->lock);
5230                         spin_unlock(&cache->space_info->lock);
5231                 } else {
5232                         old_val -= num_bytes;
5233                         btrfs_set_block_group_used(&cache->item, old_val);
5234                         cache->pinned += num_bytes;
5235                         cache->space_info->bytes_pinned += num_bytes;
5236                         cache->space_info->bytes_used -= num_bytes;
5237                         cache->space_info->disk_used -= num_bytes * factor;
5238                         spin_unlock(&cache->lock);
5239                         spin_unlock(&cache->space_info->lock);
5240
5241                         set_extent_dirty(info->pinned_extents,
5242                                          bytenr, bytenr + num_bytes - 1,
5243                                          GFP_NOFS | __GFP_NOFAIL);
5244                 }
5245                 btrfs_put_block_group(cache);
5246                 total -= num_bytes;
5247                 bytenr += num_bytes;
5248         }
5249         return 0;
5250 }
5251
5252 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5253 {
5254         struct btrfs_block_group_cache *cache;
5255         u64 bytenr;
5256
5257         spin_lock(&root->fs_info->block_group_cache_lock);
5258         bytenr = root->fs_info->first_logical_byte;
5259         spin_unlock(&root->fs_info->block_group_cache_lock);
5260
5261         if (bytenr < (u64)-1)
5262                 return bytenr;
5263
5264         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5265         if (!cache)
5266                 return 0;
5267
5268         bytenr = cache->key.objectid;
5269         btrfs_put_block_group(cache);
5270
5271         return bytenr;
5272 }
5273
5274 static int pin_down_extent(struct btrfs_root *root,
5275                            struct btrfs_block_group_cache *cache,
5276                            u64 bytenr, u64 num_bytes, int reserved)
5277 {
5278         spin_lock(&cache->space_info->lock);
5279         spin_lock(&cache->lock);
5280         cache->pinned += num_bytes;
5281         cache->space_info->bytes_pinned += num_bytes;
5282         if (reserved) {
5283                 cache->reserved -= num_bytes;
5284                 cache->space_info->bytes_reserved -= num_bytes;
5285         }
5286         spin_unlock(&cache->lock);
5287         spin_unlock(&cache->space_info->lock);
5288
5289         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5290                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5291         if (reserved)
5292                 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5293         return 0;
5294 }
5295
5296 /*
5297  * this function must be called within transaction
5298  */
5299 int btrfs_pin_extent(struct btrfs_root *root,
5300                      u64 bytenr, u64 num_bytes, int reserved)
5301 {
5302         struct btrfs_block_group_cache *cache;
5303
5304         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5305         BUG_ON(!cache); /* Logic error */
5306
5307         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5308
5309         btrfs_put_block_group(cache);
5310         return 0;
5311 }
5312
5313 /*
5314  * this function must be called within transaction
5315  */
5316 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5317                                     u64 bytenr, u64 num_bytes)
5318 {
5319         struct btrfs_block_group_cache *cache;
5320         int ret;
5321
5322         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5323         if (!cache)
5324                 return -EINVAL;
5325
5326         /*
5327          * pull in the free space cache (if any) so that our pin
5328          * removes the free space from the cache.  We have load_only set
5329          * to one because the slow code to read in the free extents does check
5330          * the pinned extents.
5331          */
5332         cache_block_group(cache, 1);
5333
5334         pin_down_extent(root, cache, bytenr, num_bytes, 0);
5335
5336         /* remove us from the free space cache (if we're there at all) */
5337         ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5338         btrfs_put_block_group(cache);
5339         return ret;
5340 }
5341
5342 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5343 {
5344         int ret;
5345         struct btrfs_block_group_cache *block_group;
5346         struct btrfs_caching_control *caching_ctl;
5347
5348         block_group = btrfs_lookup_block_group(root->fs_info, start);
5349         if (!block_group)
5350                 return -EINVAL;
5351
5352         cache_block_group(block_group, 0);
5353         caching_ctl = get_caching_control(block_group);
5354
5355         if (!caching_ctl) {
5356                 /* Logic error */
5357                 BUG_ON(!block_group_cache_done(block_group));
5358                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5359         } else {
5360                 mutex_lock(&caching_ctl->mutex);
5361
5362                 if (start >= caching_ctl->progress) {
5363                         ret = add_excluded_extent(root, start, num_bytes);
5364                 } else if (start + num_bytes <= caching_ctl->progress) {
5365                         ret = btrfs_remove_free_space(block_group,
5366                                                       start, num_bytes);
5367                 } else {
5368                         num_bytes = caching_ctl->progress - start;
5369                         ret = btrfs_remove_free_space(block_group,
5370                                                       start, num_bytes);
5371                         if (ret)
5372                                 goto out_lock;
5373
5374                         num_bytes = (start + num_bytes) -
5375                                 caching_ctl->progress;
5376                         start = caching_ctl->progress;
5377                         ret = add_excluded_extent(root, start, num_bytes);
5378                 }
5379 out_lock:
5380                 mutex_unlock(&caching_ctl->mutex);
5381                 put_caching_control(caching_ctl);
5382         }
5383         btrfs_put_block_group(block_group);
5384         return ret;
5385 }
5386
5387 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5388                                  struct extent_buffer *eb)
5389 {
5390         struct btrfs_file_extent_item *item;
5391         struct btrfs_key key;
5392         int found_type;
5393         int i;
5394
5395         if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5396                 return 0;
5397
5398         for (i = 0; i < btrfs_header_nritems(eb); i++) {
5399                 btrfs_item_key_to_cpu(eb, &key, i);
5400                 if (key.type != BTRFS_EXTENT_DATA_KEY)
5401                         continue;
5402                 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5403                 found_type = btrfs_file_extent_type(eb, item);
5404                 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5405                         continue;
5406                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5407                         continue;
5408                 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5409                 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5410                 __exclude_logged_extent(log, key.objectid, key.offset);
5411         }
5412
5413         return 0;
5414 }
5415
5416 /**
5417  * btrfs_update_reserved_bytes - update the block_group and space info counters
5418  * @cache:      The cache we are manipulating
5419  * @num_bytes:  The number of bytes in question
5420  * @reserve:    One of the reservation enums
5421  *
5422  * This is called by the allocator when it reserves space, or by somebody who is
5423  * freeing space that was never actually used on disk.  For example if you
5424  * reserve some space for a new leaf in transaction A and before transaction A
5425  * commits you free that leaf, you call this with reserve set to 0 in order to
5426  * clear the reservation.
5427  *
5428  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5429  * ENOSPC accounting.  For data we handle the reservation through clearing the
5430  * delalloc bits in the io_tree.  We have to do this since we could end up
5431  * allocating less disk space for the amount of data we have reserved in the
5432  * case of compression.
5433  *
5434  * If this is a reservation and the block group has become read only we cannot
5435  * make the reservation and return -EAGAIN, otherwise this function always
5436  * succeeds.
5437  */
5438 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5439                                        u64 num_bytes, int reserve)
5440 {
5441         struct btrfs_space_info *space_info = cache->space_info;
5442         int ret = 0;
5443
5444         spin_lock(&space_info->lock);
5445         spin_lock(&cache->lock);
5446         if (reserve != RESERVE_FREE) {
5447                 if (cache->ro) {
5448                         ret = -EAGAIN;
5449                 } else {
5450                         cache->reserved += num_bytes;
5451                         space_info->bytes_reserved += num_bytes;
5452                         if (reserve == RESERVE_ALLOC) {
5453                                 trace_btrfs_space_reservation(cache->fs_info,
5454                                                 "space_info", space_info->flags,
5455                                                 num_bytes, 0);
5456                                 space_info->bytes_may_use -= num_bytes;
5457                         }
5458                 }
5459         } else {
5460                 if (cache->ro)
5461                         space_info->bytes_readonly += num_bytes;
5462                 cache->reserved -= num_bytes;
5463                 space_info->bytes_reserved -= num_bytes;
5464         }
5465         spin_unlock(&cache->lock);
5466         spin_unlock(&space_info->lock);
5467         return ret;
5468 }
5469
5470 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5471                                 struct btrfs_root *root)
5472 {
5473         struct btrfs_fs_info *fs_info = root->fs_info;
5474         struct btrfs_caching_control *next;
5475         struct btrfs_caching_control *caching_ctl;
5476         struct btrfs_block_group_cache *cache;
5477         struct btrfs_space_info *space_info;
5478
5479         down_write(&fs_info->extent_commit_sem);
5480
5481         list_for_each_entry_safe(caching_ctl, next,
5482                                  &fs_info->caching_block_groups, list) {
5483                 cache = caching_ctl->block_group;
5484                 if (block_group_cache_done(cache)) {
5485                         cache->last_byte_to_unpin = (u64)-1;
5486                         list_del_init(&caching_ctl->list);
5487                         put_caching_control(caching_ctl);
5488                 } else {
5489                         cache->last_byte_to_unpin = caching_ctl->progress;
5490                 }
5491         }
5492
5493         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5494                 fs_info->pinned_extents = &fs_info->freed_extents[1];
5495         else
5496                 fs_info->pinned_extents = &fs_info->freed_extents[0];
5497
5498         up_write(&fs_info->extent_commit_sem);
5499
5500         list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5501                 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5502
5503         update_global_block_rsv(fs_info);
5504 }
5505
5506 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5507 {
5508         struct btrfs_fs_info *fs_info = root->fs_info;
5509         struct btrfs_block_group_cache *cache = NULL;
5510         struct btrfs_space_info *space_info;
5511         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5512         u64 len;
5513         bool readonly;
5514
5515         while (start <= end) {
5516                 readonly = false;
5517                 if (!cache ||
5518                     start >= cache->key.objectid + cache->key.offset) {
5519                         if (cache)
5520                                 btrfs_put_block_group(cache);
5521                         cache = btrfs_lookup_block_group(fs_info, start);
5522                         BUG_ON(!cache); /* Logic error */
5523                 }
5524
5525                 len = cache->key.objectid + cache->key.offset - start;
5526                 len = min(len, end + 1 - start);
5527
5528                 if (start < cache->last_byte_to_unpin) {
5529                         len = min(len, cache->last_byte_to_unpin - start);
5530                         btrfs_add_free_space(cache, start, len);
5531                 }
5532
5533                 start += len;
5534                 space_info = cache->space_info;
5535
5536                 spin_lock(&space_info->lock);
5537                 spin_lock(&cache->lock);
5538                 cache->pinned -= len;
5539                 space_info->bytes_pinned -= len;
5540                 if (cache->ro) {
5541                         space_info->bytes_readonly += len;
5542                         readonly = true;
5543                 }
5544                 spin_unlock(&cache->lock);
5545                 if (!readonly && global_rsv->space_info == space_info) {
5546                         spin_lock(&global_rsv->lock);
5547                         if (!global_rsv->full) {
5548                                 len = min(len, global_rsv->size -
5549                                           global_rsv->reserved);
5550                                 global_rsv->reserved += len;
5551                                 space_info->bytes_may_use += len;
5552                                 if (global_rsv->reserved >= global_rsv->size)
5553                                         global_rsv->full = 1;
5554                         }
5555                         spin_unlock(&global_rsv->lock);
5556                 }
5557                 spin_unlock(&space_info->lock);
5558         }
5559
5560         if (cache)
5561                 btrfs_put_block_group(cache);
5562         return 0;
5563 }
5564
5565 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5566                                struct btrfs_root *root)
5567 {
5568         struct btrfs_fs_info *fs_info = root->fs_info;
5569         struct extent_io_tree *unpin;
5570         u64 start;
5571         u64 end;
5572         int ret;
5573
5574         if (trans->aborted)
5575                 return 0;
5576
5577         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5578                 unpin = &fs_info->freed_extents[1];
5579         else
5580                 unpin = &fs_info->freed_extents[0];
5581
5582         while (1) {
5583                 ret = find_first_extent_bit(unpin, 0, &start, &end,
5584                                             EXTENT_DIRTY, NULL);
5585                 if (ret)
5586                         break;
5587
5588                 if (btrfs_test_opt(root, DISCARD))
5589                         ret = btrfs_discard_extent(root, start,
5590                                                    end + 1 - start, NULL);
5591
5592                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5593                 unpin_extent_range(root, start, end);
5594                 cond_resched();
5595         }
5596
5597         return 0;
5598 }
5599
5600 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5601                              u64 owner, u64 root_objectid)
5602 {
5603         struct btrfs_space_info *space_info;
5604         u64 flags;
5605
5606         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5607                 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5608                         flags = BTRFS_BLOCK_GROUP_SYSTEM;
5609                 else
5610                         flags = BTRFS_BLOCK_GROUP_METADATA;
5611         } else {
5612                 flags = BTRFS_BLOCK_GROUP_DATA;
5613         }
5614
5615         space_info = __find_space_info(fs_info, flags);
5616         BUG_ON(!space_info); /* Logic bug */
5617         percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5618 }
5619
5620
5621 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5622                                 struct btrfs_root *root,
5623                                 u64 bytenr, u64 num_bytes, u64 parent,
5624                                 u64 root_objectid, u64 owner_objectid,
5625                                 u64 owner_offset, int refs_to_drop,
5626                                 struct btrfs_delayed_extent_op *extent_op)
5627 {
5628         struct btrfs_key key;
5629         struct btrfs_path *path;
5630         struct btrfs_fs_info *info = root->fs_info;
5631         struct btrfs_root *extent_root = info->extent_root;
5632         struct extent_buffer *leaf;
5633         struct btrfs_extent_item *ei;
5634         struct btrfs_extent_inline_ref *iref;
5635         int ret;
5636         int is_data;
5637         int extent_slot = 0;
5638         int found_extent = 0;
5639         int num_to_del = 1;
5640         u32 item_size;
5641         u64 refs;
5642         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5643                                                  SKINNY_METADATA);
5644
5645         path = btrfs_alloc_path();
5646         if (!path)
5647                 return -ENOMEM;
5648
5649         path->reada = 1;
5650         path->leave_spinning = 1;
5651
5652         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5653         BUG_ON(!is_data && refs_to_drop != 1);
5654
5655         if (is_data)
5656                 skinny_metadata = 0;
5657
5658         ret = lookup_extent_backref(trans, extent_root, path, &iref,
5659                                     bytenr, num_bytes, parent,
5660                                     root_objectid, owner_objectid,
5661                                     owner_offset);
5662         if (ret == 0) {
5663                 extent_slot = path->slots[0];
5664                 while (extent_slot >= 0) {
5665                         btrfs_item_key_to_cpu(path->nodes[0], &key,
5666                                               extent_slot);
5667                         if (key.objectid != bytenr)
5668                                 break;
5669                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5670                             key.offset == num_bytes) {
5671                                 found_extent = 1;
5672                                 break;
5673                         }
5674                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
5675                             key.offset == owner_objectid) {
5676                                 found_extent = 1;
5677                                 break;
5678                         }
5679                         if (path->slots[0] - extent_slot > 5)
5680                                 break;
5681                         extent_slot--;
5682                 }
5683 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5684                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5685                 if (found_extent && item_size < sizeof(*ei))
5686                         found_extent = 0;
5687 #endif
5688                 if (!found_extent) {
5689                         BUG_ON(iref);
5690                         ret = remove_extent_backref(trans, extent_root, path,
5691                                                     NULL, refs_to_drop,
5692                                                     is_data);
5693                         if (ret) {
5694                                 btrfs_abort_transaction(trans, extent_root, ret);
5695                                 goto out;
5696                         }
5697                         btrfs_release_path(path);
5698                         path->leave_spinning = 1;
5699
5700                         key.objectid = bytenr;
5701                         key.type = BTRFS_EXTENT_ITEM_KEY;
5702                         key.offset = num_bytes;
5703
5704                         if (!is_data && skinny_metadata) {
5705                                 key.type = BTRFS_METADATA_ITEM_KEY;
5706                                 key.offset = owner_objectid;
5707                         }
5708
5709                         ret = btrfs_search_slot(trans, extent_root,
5710                                                 &key, path, -1, 1);
5711                         if (ret > 0 && skinny_metadata && path->slots[0]) {
5712                                 /*
5713                                  * Couldn't find our skinny metadata item,
5714                                  * see if we have ye olde extent item.
5715                                  */
5716                                 path->slots[0]--;
5717                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
5718                                                       path->slots[0]);
5719                                 if (key.objectid == bytenr &&
5720                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
5721                                     key.offset == num_bytes)
5722                                         ret = 0;
5723                         }
5724
5725                         if (ret > 0 && skinny_metadata) {
5726                                 skinny_metadata = false;
5727                                 key.type = BTRFS_EXTENT_ITEM_KEY;
5728                                 key.offset = num_bytes;
5729                                 btrfs_release_path(path);
5730                                 ret = btrfs_search_slot(trans, extent_root,
5731                                                         &key, path, -1, 1);
5732                         }
5733
5734                         if (ret) {
5735                                 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5736                                         ret, bytenr);
5737                                 if (ret > 0)
5738                                         btrfs_print_leaf(extent_root,
5739                                                          path->nodes[0]);
5740                         }
5741                         if (ret < 0) {
5742                                 btrfs_abort_transaction(trans, extent_root, ret);
5743                                 goto out;
5744                         }
5745                         extent_slot = path->slots[0];
5746                 }
5747         } else if (WARN_ON(ret == -ENOENT)) {
5748                 btrfs_print_leaf(extent_root, path->nodes[0]);
5749                 btrfs_err(info,
5750                         "unable to find ref byte nr %llu parent %llu root %llu  owner %llu offset %llu",
5751                         bytenr, parent, root_objectid, owner_objectid,
5752                         owner_offset);
5753         } else {
5754                 btrfs_abort_transaction(trans, extent_root, ret);
5755                 goto out;
5756         }
5757
5758         leaf = path->nodes[0];
5759         item_size = btrfs_item_size_nr(leaf, extent_slot);
5760 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5761         if (item_size < sizeof(*ei)) {
5762                 BUG_ON(found_extent || extent_slot != path->slots[0]);
5763                 ret = convert_extent_item_v0(trans, extent_root, path,
5764                                              owner_objectid, 0);
5765                 if (ret < 0) {
5766                         btrfs_abort_transaction(trans, extent_root, ret);
5767                         goto out;
5768                 }
5769
5770                 btrfs_release_path(path);
5771                 path->leave_spinning = 1;
5772
5773                 key.objectid = bytenr;
5774                 key.type = BTRFS_EXTENT_ITEM_KEY;
5775                 key.offset = num_bytes;
5776
5777                 ret = btrfs_search_slot(trans, extent_root, &key, path,
5778                                         -1, 1);
5779                 if (ret) {
5780                         btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5781                                 ret, bytenr);
5782                         btrfs_print_leaf(extent_root, path->nodes[0]);
5783                 }
5784                 if (ret < 0) {
5785                         btrfs_abort_transaction(trans, extent_root, ret);
5786                         goto out;
5787                 }
5788
5789                 extent_slot = path->slots[0];
5790                 leaf = path->nodes[0];
5791                 item_size = btrfs_item_size_nr(leaf, extent_slot);
5792         }
5793 #endif
5794         BUG_ON(item_size < sizeof(*ei));
5795         ei = btrfs_item_ptr(leaf, extent_slot,
5796                             struct btrfs_extent_item);
5797         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5798             key.type == BTRFS_EXTENT_ITEM_KEY) {
5799                 struct btrfs_tree_block_info *bi;
5800                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5801                 bi = (struct btrfs_tree_block_info *)(ei + 1);
5802                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5803         }
5804
5805         refs = btrfs_extent_refs(leaf, ei);
5806         if (refs < refs_to_drop) {
5807                 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5808                           "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5809                 ret = -EINVAL;
5810                 btrfs_abort_transaction(trans, extent_root, ret);
5811                 goto out;
5812         }
5813         refs -= refs_to_drop;
5814
5815         if (refs > 0) {
5816                 if (extent_op)
5817                         __run_delayed_extent_op(extent_op, leaf, ei);
5818                 /*
5819                  * In the case of inline back ref, reference count will
5820                  * be updated by remove_extent_backref
5821                  */
5822                 if (iref) {
5823                         BUG_ON(!found_extent);
5824                 } else {
5825                         btrfs_set_extent_refs(leaf, ei, refs);
5826                         btrfs_mark_buffer_dirty(leaf);
5827                 }
5828                 if (found_extent) {
5829                         ret = remove_extent_backref(trans, extent_root, path,
5830                                                     iref, refs_to_drop,
5831                                                     is_data);
5832                         if (ret) {
5833                                 btrfs_abort_transaction(trans, extent_root, ret);
5834                                 goto out;
5835                         }
5836                 }
5837                 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5838                                  root_objectid);
5839         } else {
5840                 if (found_extent) {
5841                         BUG_ON(is_data && refs_to_drop !=
5842                                extent_data_ref_count(root, path, iref));
5843                         if (iref) {
5844                                 BUG_ON(path->slots[0] != extent_slot);
5845                         } else {
5846                                 BUG_ON(path->slots[0] != extent_slot + 1);
5847                                 path->slots[0] = extent_slot;
5848                                 num_to_del = 2;
5849                         }
5850                 }
5851
5852                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5853                                       num_to_del);
5854                 if (ret) {
5855                         btrfs_abort_transaction(trans, extent_root, ret);
5856                         goto out;
5857                 }
5858                 btrfs_release_path(path);
5859
5860                 if (is_data) {
5861                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5862                         if (ret) {
5863                                 btrfs_abort_transaction(trans, extent_root, ret);
5864                                 goto out;
5865                         }
5866                 }
5867
5868                 ret = update_block_group(root, bytenr, num_bytes, 0);
5869                 if (ret) {
5870                         btrfs_abort_transaction(trans, extent_root, ret);
5871                         goto out;
5872                 }
5873         }
5874 out:
5875         btrfs_free_path(path);
5876         return ret;
5877 }
5878
5879 /*
5880  * when we free an block, it is possible (and likely) that we free the last
5881  * delayed ref for that extent as well.  This searches the delayed ref tree for
5882  * a given extent, and if there are no other delayed refs to be processed, it
5883  * removes it from the tree.
5884  */
5885 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5886                                       struct btrfs_root *root, u64 bytenr)
5887 {
5888         struct btrfs_delayed_ref_head *head;
5889         struct btrfs_delayed_ref_root *delayed_refs;
5890         int ret = 0;
5891
5892         delayed_refs = &trans->transaction->delayed_refs;
5893         spin_lock(&delayed_refs->lock);
5894         head = btrfs_find_delayed_ref_head(trans, bytenr);
5895         if (!head)
5896                 goto out;
5897
5898         spin_lock(&head->lock);
5899         if (rb_first(&head->ref_root))
5900                 goto out;
5901
5902         if (head->extent_op) {
5903                 if (!head->must_insert_reserved)
5904                         goto out;
5905                 btrfs_free_delayed_extent_op(head->extent_op);
5906                 head->extent_op = NULL;
5907         }
5908
5909         /*
5910          * waiting for the lock here would deadlock.  If someone else has it
5911          * locked they are already in the process of dropping it anyway
5912          */
5913         if (!mutex_trylock(&head->mutex))
5914                 goto out;
5915
5916         /*
5917          * at this point we have a head with no other entries.  Go
5918          * ahead and process it.
5919          */
5920         head->node.in_tree = 0;
5921         rb_erase(&head->href_node, &delayed_refs->href_root);
5922
5923         atomic_dec(&delayed_refs->num_entries);
5924
5925         /*
5926          * we don't take a ref on the node because we're removing it from the
5927          * tree, so we just steal the ref the tree was holding.
5928          */
5929         delayed_refs->num_heads--;
5930         if (head->processing == 0)
5931                 delayed_refs->num_heads_ready--;
5932         head->processing = 0;
5933         spin_unlock(&head->lock);
5934         spin_unlock(&delayed_refs->lock);
5935
5936         BUG_ON(head->extent_op);
5937         if (head->must_insert_reserved)
5938                 ret = 1;
5939
5940         mutex_unlock(&head->mutex);
5941         btrfs_put_delayed_ref(&head->node);
5942         return ret;
5943 out:
5944         spin_unlock(&head->lock);
5945         spin_unlock(&delayed_refs->lock);
5946         return 0;
5947 }
5948
5949 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5950                            struct btrfs_root *root,
5951                            struct extent_buffer *buf,
5952                            u64 parent, int last_ref)
5953 {
5954         struct btrfs_block_group_cache *cache = NULL;
5955         int pin = 1;
5956         int ret;
5957
5958         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5959                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5960                                         buf->start, buf->len,
5961                                         parent, root->root_key.objectid,
5962                                         btrfs_header_level(buf),
5963                                         BTRFS_DROP_DELAYED_REF, NULL, 0);
5964                 BUG_ON(ret); /* -ENOMEM */
5965         }
5966
5967         if (!last_ref)
5968                 return;
5969
5970         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5971
5972         if (btrfs_header_generation(buf) == trans->transid) {
5973                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5974                         ret = check_ref_cleanup(trans, root, buf->start);
5975                         if (!ret)
5976                                 goto out;
5977                 }
5978
5979                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5980                         pin_down_extent(root, cache, buf->start, buf->len, 1);
5981                         goto out;
5982                 }
5983
5984                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5985
5986                 btrfs_add_free_space(cache, buf->start, buf->len);
5987                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5988                 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
5989                 pin = 0;
5990         }
5991 out:
5992         if (pin)
5993                 add_pinned_bytes(root->fs_info, buf->len,
5994                                  btrfs_header_level(buf),
5995                                  root->root_key.objectid);
5996
5997         /*
5998          * Deleting the buffer, clear the corrupt flag since it doesn't matter
5999          * anymore.
6000          */
6001         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6002         btrfs_put_block_group(cache);
6003 }
6004
6005 /* Can return -ENOMEM */
6006 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6007                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6008                       u64 owner, u64 offset, int for_cow)
6009 {
6010         int ret;
6011         struct btrfs_fs_info *fs_info = root->fs_info;
6012
6013         add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6014
6015         /*
6016          * tree log blocks never actually go into the extent allocation
6017          * tree, just update pinning info and exit early.
6018          */
6019         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6020                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6021                 /* unlocks the pinned mutex */
6022                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6023                 ret = 0;
6024         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6025                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6026                                         num_bytes,
6027                                         parent, root_objectid, (int)owner,
6028                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6029         } else {
6030                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6031                                                 num_bytes,
6032                                                 parent, root_objectid, owner,
6033                                                 offset, BTRFS_DROP_DELAYED_REF,
6034                                                 NULL, for_cow);
6035         }
6036         return ret;
6037 }
6038
6039 static u64 stripe_align(struct btrfs_root *root,
6040                         struct btrfs_block_group_cache *cache,
6041                         u64 val, u64 num_bytes)
6042 {
6043         u64 ret = ALIGN(val, root->stripesize);
6044         return ret;
6045 }
6046
6047 /*
6048  * when we wait for progress in the block group caching, its because
6049  * our allocation attempt failed at least once.  So, we must sleep
6050  * and let some progress happen before we try again.
6051  *
6052  * This function will sleep at least once waiting for new free space to
6053  * show up, and then it will check the block group free space numbers
6054  * for our min num_bytes.  Another option is to have it go ahead
6055  * and look in the rbtree for a free extent of a given size, but this
6056  * is a good start.
6057  *
6058  * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6059  * any of the information in this block group.
6060  */
6061 static noinline void
6062 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6063                                 u64 num_bytes)
6064 {
6065         struct btrfs_caching_control *caching_ctl;
6066
6067         caching_ctl = get_caching_control(cache);
6068         if (!caching_ctl)
6069                 return;
6070
6071         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6072                    (cache->free_space_ctl->free_space >= num_bytes));
6073
6074         put_caching_control(caching_ctl);
6075 }
6076
6077 static noinline int
6078 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6079 {
6080         struct btrfs_caching_control *caching_ctl;
6081         int ret = 0;
6082
6083         caching_ctl = get_caching_control(cache);
6084         if (!caching_ctl)
6085                 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6086
6087         wait_event(caching_ctl->wait, block_group_cache_done(cache));
6088         if (cache->cached == BTRFS_CACHE_ERROR)
6089                 ret = -EIO;
6090         put_caching_control(caching_ctl);
6091         return ret;
6092 }
6093
6094 int __get_raid_index(u64 flags)
6095 {
6096         if (flags & BTRFS_BLOCK_GROUP_RAID10)
6097                 return BTRFS_RAID_RAID10;
6098         else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6099                 return BTRFS_RAID_RAID1;
6100         else if (flags & BTRFS_BLOCK_GROUP_DUP)
6101                 return BTRFS_RAID_DUP;
6102         else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6103                 return BTRFS_RAID_RAID0;
6104         else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6105                 return BTRFS_RAID_RAID5;
6106         else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6107                 return BTRFS_RAID_RAID6;
6108
6109         return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6110 }
6111
6112 int get_block_group_index(struct btrfs_block_group_cache *cache)
6113 {
6114         return __get_raid_index(cache->flags);
6115 }
6116
6117 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6118         [BTRFS_RAID_RAID10]     = "raid10",
6119         [BTRFS_RAID_RAID1]      = "raid1",
6120         [BTRFS_RAID_DUP]        = "dup",
6121         [BTRFS_RAID_RAID0]      = "raid0",
6122         [BTRFS_RAID_SINGLE]     = "single",
6123         [BTRFS_RAID_RAID5]      = "raid5",
6124         [BTRFS_RAID_RAID6]      = "raid6",
6125 };
6126
6127 static const char *get_raid_name(enum btrfs_raid_types type)
6128 {
6129         if (type >= BTRFS_NR_RAID_TYPES)
6130                 return NULL;
6131
6132         return btrfs_raid_type_names[type];
6133 }
6134
6135 enum btrfs_loop_type {
6136         LOOP_CACHING_NOWAIT = 0,
6137         LOOP_CACHING_WAIT = 1,
6138         LOOP_ALLOC_CHUNK = 2,
6139         LOOP_NO_EMPTY_SIZE = 3,
6140 };
6141
6142 /*
6143  * walks the btree of allocated extents and find a hole of a given size.
6144  * The key ins is changed to record the hole:
6145  * ins->objectid == start position
6146  * ins->flags = BTRFS_EXTENT_ITEM_KEY
6147  * ins->offset == the size of the hole.
6148  * Any available blocks before search_start are skipped.
6149  *
6150  * If there is no suitable free space, we will record the max size of
6151  * the free space extent currently.
6152  */
6153 static noinline int find_free_extent(struct btrfs_root *orig_root,
6154                                      u64 num_bytes, u64 empty_size,
6155                                      u64 hint_byte, struct btrfs_key *ins,
6156                                      u64 flags)
6157 {
6158         int ret = 0;
6159         struct btrfs_root *root = orig_root->fs_info->extent_root;
6160         struct btrfs_free_cluster *last_ptr = NULL;
6161         struct btrfs_block_group_cache *block_group = NULL;
6162         u64 search_start = 0;
6163         u64 max_extent_size = 0;
6164         int empty_cluster = 2 * 1024 * 1024;
6165         struct btrfs_space_info *space_info;
6166         int loop = 0;
6167         int index = __get_raid_index(flags);
6168         int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6169                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6170         bool failed_cluster_refill = false;
6171         bool failed_alloc = false;
6172         bool use_cluster = true;
6173         bool have_caching_bg = false;
6174
6175         WARN_ON(num_bytes < root->sectorsize);
6176         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6177         ins->objectid = 0;
6178         ins->offset = 0;
6179
6180         trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6181
6182         space_info = __find_space_info(root->fs_info, flags);
6183         if (!space_info) {
6184                 btrfs_err(root->fs_info, "No space info for %llu", flags);
6185                 return -ENOSPC;
6186         }
6187
6188         /*
6189          * If the space info is for both data and metadata it means we have a
6190          * small filesystem and we can't use the clustering stuff.
6191          */
6192         if (btrfs_mixed_space_info(space_info))
6193                 use_cluster = false;
6194
6195         if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6196                 last_ptr = &root->fs_info->meta_alloc_cluster;
6197                 if (!btrfs_test_opt(root, SSD))
6198                         empty_cluster = 64 * 1024;
6199         }
6200
6201         if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6202             btrfs_test_opt(root, SSD)) {
6203                 last_ptr = &root->fs_info->data_alloc_cluster;
6204         }
6205
6206         if (last_ptr) {
6207                 spin_lock(&last_ptr->lock);
6208                 if (last_ptr->block_group)
6209                         hint_byte = last_ptr->window_start;
6210                 spin_unlock(&last_ptr->lock);
6211         }
6212
6213         search_start = max(search_start, first_logical_byte(root, 0));
6214         search_start = max(search_start, hint_byte);
6215
6216         if (!last_ptr)
6217                 empty_cluster = 0;
6218
6219         if (search_start == hint_byte) {
6220                 block_group = btrfs_lookup_block_group(root->fs_info,
6221                                                        search_start);
6222                 /*
6223                  * we don't want to use the block group if it doesn't match our
6224                  * allocation bits, or if its not cached.
6225                  *
6226                  * However if we are re-searching with an ideal block group
6227                  * picked out then we don't care that the block group is cached.
6228                  */
6229                 if (block_group && block_group_bits(block_group, flags) &&
6230                     block_group->cached != BTRFS_CACHE_NO) {
6231                         down_read(&space_info->groups_sem);
6232                         if (list_empty(&block_group->list) ||
6233                             block_group->ro) {
6234                                 /*
6235                                  * someone is removing this block group,
6236                                  * we can't jump into the have_block_group
6237                                  * target because our list pointers are not
6238                                  * valid
6239                                  */
6240                                 btrfs_put_block_group(block_group);
6241                                 up_read(&space_info->groups_sem);
6242                         } else {
6243                                 index = get_block_group_index(block_group);
6244                                 goto have_block_group;
6245                         }
6246                 } else if (block_group) {
6247                         btrfs_put_block_group(block_group);
6248                 }
6249         }
6250 search:
6251         have_caching_bg = false;
6252         down_read(&space_info->groups_sem);
6253         list_for_each_entry(block_group, &space_info->block_groups[index],
6254                             list) {
6255                 u64 offset;
6256                 int cached;
6257
6258                 btrfs_get_block_group(block_group);
6259                 search_start = block_group->key.objectid;
6260
6261                 /*
6262                  * this can happen if we end up cycling through all the
6263                  * raid types, but we want to make sure we only allocate
6264                  * for the proper type.
6265                  */
6266                 if (!block_group_bits(block_group, flags)) {
6267                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
6268                                 BTRFS_BLOCK_GROUP_RAID1 |
6269                                 BTRFS_BLOCK_GROUP_RAID5 |
6270                                 BTRFS_BLOCK_GROUP_RAID6 |
6271                                 BTRFS_BLOCK_GROUP_RAID10;
6272
6273                         /*
6274                          * if they asked for extra copies and this block group
6275                          * doesn't provide them, bail.  This does allow us to
6276                          * fill raid0 from raid1.
6277                          */
6278                         if ((flags & extra) && !(block_group->flags & extra))
6279                                 goto loop;
6280                 }
6281
6282 have_block_group:
6283                 cached = block_group_cache_done(block_group);
6284                 if (unlikely(!cached)) {
6285                         ret = cache_block_group(block_group, 0);
6286                         BUG_ON(ret < 0);
6287                         ret = 0;
6288                 }
6289
6290                 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6291                         goto loop;
6292                 if (unlikely(block_group->ro))
6293                         goto loop;
6294
6295                 /*
6296                  * Ok we want to try and use the cluster allocator, so
6297                  * lets look there
6298                  */
6299                 if (last_ptr) {
6300                         struct btrfs_block_group_cache *used_block_group;
6301                         unsigned long aligned_cluster;
6302                         /*
6303                          * the refill lock keeps out other
6304                          * people trying to start a new cluster
6305                          */
6306                         spin_lock(&last_ptr->refill_lock);
6307                         used_block_group = last_ptr->block_group;
6308                         if (used_block_group != block_group &&
6309                             (!used_block_group ||
6310                              used_block_group->ro ||
6311                              !block_group_bits(used_block_group, flags)))
6312                                 goto refill_cluster;
6313
6314                         if (used_block_group != block_group)
6315                                 btrfs_get_block_group(used_block_group);
6316
6317                         offset = btrfs_alloc_from_cluster(used_block_group,
6318                                                 last_ptr,
6319                                                 num_bytes,
6320                                                 used_block_group->key.objectid,
6321                                                 &max_extent_size);
6322                         if (offset) {
6323                                 /* we have a block, we're done */
6324                                 spin_unlock(&last_ptr->refill_lock);
6325                                 trace_btrfs_reserve_extent_cluster(root,
6326                                                 used_block_group,
6327                                                 search_start, num_bytes);
6328                                 if (used_block_group != block_group) {
6329                                         btrfs_put_block_group(block_group);
6330                                         block_group = used_block_group;
6331                                 }
6332                                 goto checks;
6333                         }
6334
6335                         WARN_ON(last_ptr->block_group != used_block_group);
6336                         if (used_block_group != block_group)
6337                                 btrfs_put_block_group(used_block_group);
6338 refill_cluster:
6339                         /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6340                          * set up a new clusters, so lets just skip it
6341                          * and let the allocator find whatever block
6342                          * it can find.  If we reach this point, we
6343                          * will have tried the cluster allocator
6344                          * plenty of times and not have found
6345                          * anything, so we are likely way too
6346                          * fragmented for the clustering stuff to find
6347                          * anything.
6348                          *
6349                          * However, if the cluster is taken from the
6350                          * current block group, release the cluster
6351                          * first, so that we stand a better chance of
6352                          * succeeding in the unclustered
6353                          * allocation.  */
6354                         if (loop >= LOOP_NO_EMPTY_SIZE &&
6355                             last_ptr->block_group != block_group) {
6356                                 spin_unlock(&last_ptr->refill_lock);
6357                                 goto unclustered_alloc;
6358                         }
6359
6360                         /*
6361                          * this cluster didn't work out, free it and
6362                          * start over
6363                          */
6364                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
6365
6366                         if (loop >= LOOP_NO_EMPTY_SIZE) {
6367                                 spin_unlock(&last_ptr->refill_lock);
6368                                 goto unclustered_alloc;
6369                         }
6370
6371                         aligned_cluster = max_t(unsigned long,
6372                                                 empty_cluster + empty_size,
6373                                               block_group->full_stripe_len);
6374
6375                         /* allocate a cluster in this block group */
6376                         ret = btrfs_find_space_cluster(root, block_group,
6377                                                        last_ptr, search_start,
6378                                                        num_bytes,
6379                                                        aligned_cluster);
6380                         if (ret == 0) {
6381                                 /*
6382                                  * now pull our allocation out of this
6383                                  * cluster
6384                                  */
6385                                 offset = btrfs_alloc_from_cluster(block_group,
6386                                                         last_ptr,
6387                                                         num_bytes,
6388                                                         search_start,
6389                                                         &max_extent_size);
6390                                 if (offset) {
6391                                         /* we found one, proceed */
6392                                         spin_unlock(&last_ptr->refill_lock);
6393                                         trace_btrfs_reserve_extent_cluster(root,
6394                                                 block_group, search_start,
6395                                                 num_bytes);
6396                                         goto checks;
6397                                 }
6398                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
6399                                    && !failed_cluster_refill) {
6400                                 spin_unlock(&last_ptr->refill_lock);
6401
6402                                 failed_cluster_refill = true;
6403                                 wait_block_group_cache_progress(block_group,
6404                                        num_bytes + empty_cluster + empty_size);
6405                                 goto have_block_group;
6406                         }
6407
6408                         /*
6409                          * at this point we either didn't find a cluster
6410                          * or we weren't able to allocate a block from our
6411                          * cluster.  Free the cluster we've been trying
6412                          * to use, and go to the next block group
6413                          */
6414                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
6415                         spin_unlock(&last_ptr->refill_lock);
6416                         goto loop;
6417                 }
6418
6419 unclustered_alloc:
6420                 spin_lock(&block_group->free_space_ctl->tree_lock);
6421                 if (cached &&
6422                     block_group->free_space_ctl->free_space <
6423                     num_bytes + empty_cluster + empty_size) {
6424                         if (block_group->free_space_ctl->free_space >
6425                             max_extent_size)
6426                                 max_extent_size =
6427                                         block_group->free_space_ctl->free_space;
6428                         spin_unlock(&block_group->free_space_ctl->tree_lock);
6429                         goto loop;
6430                 }
6431                 spin_unlock(&block_group->free_space_ctl->tree_lock);
6432
6433                 offset = btrfs_find_space_for_alloc(block_group, search_start,
6434                                                     num_bytes, empty_size,
6435                                                     &max_extent_size);
6436                 /*
6437                  * If we didn't find a chunk, and we haven't failed on this
6438                  * block group before, and this block group is in the middle of
6439                  * caching and we are ok with waiting, then go ahead and wait
6440                  * for progress to be made, and set failed_alloc to true.
6441                  *
6442                  * If failed_alloc is true then we've already waited on this
6443                  * block group once and should move on to the next block group.
6444                  */
6445                 if (!offset && !failed_alloc && !cached &&
6446                     loop > LOOP_CACHING_NOWAIT) {
6447                         wait_block_group_cache_progress(block_group,
6448                                                 num_bytes + empty_size);
6449                         failed_alloc = true;
6450                         goto have_block_group;
6451                 } else if (!offset) {
6452                         if (!cached)
6453                                 have_caching_bg = true;
6454                         goto loop;
6455                 }
6456 checks:
6457                 search_start = stripe_align(root, block_group,
6458                                             offset, num_bytes);
6459
6460                 /* move on to the next group */
6461                 if (search_start + num_bytes >
6462                     block_group->key.objectid + block_group->key.offset) {
6463                         btrfs_add_free_space(block_group, offset, num_bytes);
6464                         goto loop;
6465                 }
6466
6467                 if (offset < search_start)
6468                         btrfs_add_free_space(block_group, offset,
6469                                              search_start - offset);
6470                 BUG_ON(offset > search_start);
6471
6472                 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6473                                                   alloc_type);
6474                 if (ret == -EAGAIN) {
6475                         btrfs_add_free_space(block_group, offset, num_bytes);
6476                         goto loop;
6477                 }
6478
6479                 /* we are all good, lets return */
6480                 ins->objectid = search_start;
6481                 ins->offset = num_bytes;
6482
6483                 trace_btrfs_reserve_extent(orig_root, block_group,
6484                                            search_start, num_bytes);
6485                 btrfs_put_block_group(block_group);
6486                 break;
6487 loop:
6488                 failed_cluster_refill = false;
6489                 failed_alloc = false;
6490                 BUG_ON(index != get_block_group_index(block_group));
6491                 btrfs_put_block_group(block_group);
6492         }
6493         up_read(&space_info->groups_sem);
6494
6495         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6496                 goto search;
6497
6498         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6499                 goto search;
6500
6501         /*
6502          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6503          *                      caching kthreads as we move along
6504          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6505          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6506          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6507          *                      again
6508          */
6509         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6510                 index = 0;
6511                 loop++;
6512                 if (loop == LOOP_ALLOC_CHUNK) {
6513                         struct btrfs_trans_handle *trans;
6514
6515                         trans = btrfs_join_transaction(root);
6516                         if (IS_ERR(trans)) {
6517                                 ret = PTR_ERR(trans);
6518                                 goto out;
6519                         }
6520
6521                         ret = do_chunk_alloc(trans, root, flags,
6522                                              CHUNK_ALLOC_FORCE);
6523                         /*
6524                          * Do not bail out on ENOSPC since we
6525                          * can do more things.
6526                          */
6527                         if (ret < 0 && ret != -ENOSPC)
6528                                 btrfs_abort_transaction(trans,
6529                                                         root, ret);
6530                         else
6531                                 ret = 0;
6532                         btrfs_end_transaction(trans, root);
6533                         if (ret)
6534                                 goto out;
6535                 }
6536
6537                 if (loop == LOOP_NO_EMPTY_SIZE) {
6538                         empty_size = 0;
6539                         empty_cluster = 0;
6540                 }
6541
6542                 goto search;
6543         } else if (!ins->objectid) {
6544                 ret = -ENOSPC;
6545         } else if (ins->objectid) {
6546                 ret = 0;
6547         }
6548 out:
6549         if (ret == -ENOSPC)
6550                 ins->offset = max_extent_size;
6551         return ret;
6552 }
6553
6554 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6555                             int dump_block_groups)
6556 {
6557         struct btrfs_block_group_cache *cache;
6558         int index = 0;
6559
6560         spin_lock(&info->lock);
6561         printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6562                info->flags,
6563                info->total_bytes - info->bytes_used - info->bytes_pinned -
6564                info->bytes_reserved - info->bytes_readonly,
6565                (info->full) ? "" : "not ");
6566         printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6567                "reserved=%llu, may_use=%llu, readonly=%llu\n",
6568                info->total_bytes, info->bytes_used, info->bytes_pinned,
6569                info->bytes_reserved, info->bytes_may_use,
6570                info->bytes_readonly);
6571         spin_unlock(&info->lock);
6572
6573         if (!dump_block_groups)
6574                 return;
6575
6576         down_read(&info->groups_sem);
6577 again:
6578         list_for_each_entry(cache, &info->block_groups[index], list) {
6579                 spin_lock(&cache->lock);
6580                 printk(KERN_INFO "BTRFS: "
6581                            "block group %llu has %llu bytes, "
6582                            "%llu used %llu pinned %llu reserved %s\n",
6583                        cache->key.objectid, cache->key.offset,
6584                        btrfs_block_group_used(&cache->item), cache->pinned,
6585                        cache->reserved, cache->ro ? "[readonly]" : "");
6586                 btrfs_dump_free_space(cache, bytes);
6587                 spin_unlock(&cache->lock);
6588         }
6589         if (++index < BTRFS_NR_RAID_TYPES)
6590                 goto again;
6591         up_read(&info->groups_sem);
6592 }
6593
6594 int btrfs_reserve_extent(struct btrfs_root *root,
6595                          u64 num_bytes, u64 min_alloc_size,
6596                          u64 empty_size, u64 hint_byte,
6597                          struct btrfs_key *ins, int is_data)
6598 {
6599         bool final_tried = false;
6600         u64 flags;
6601         int ret;
6602
6603         flags = btrfs_get_alloc_profile(root, is_data);
6604 again:
6605         WARN_ON(num_bytes < root->sectorsize);
6606         ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6607                                flags);
6608
6609         if (ret == -ENOSPC) {
6610                 if (!final_tried && ins->offset) {
6611                         num_bytes = min(num_bytes >> 1, ins->offset);
6612                         num_bytes = round_down(num_bytes, root->sectorsize);
6613                         num_bytes = max(num_bytes, min_alloc_size);
6614                         if (num_bytes == min_alloc_size)
6615                                 final_tried = true;
6616                         goto again;
6617                 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6618                         struct btrfs_space_info *sinfo;
6619
6620                         sinfo = __find_space_info(root->fs_info, flags);
6621                         btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6622                                 flags, num_bytes);
6623                         if (sinfo)
6624                                 dump_space_info(sinfo, num_bytes, 1);
6625                 }
6626         }
6627
6628         return ret;
6629 }
6630
6631 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6632                                         u64 start, u64 len, int pin)
6633 {
6634         struct btrfs_block_group_cache *cache;
6635         int ret = 0;
6636
6637         cache = btrfs_lookup_block_group(root->fs_info, start);
6638         if (!cache) {
6639                 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6640                         start);
6641                 return -ENOSPC;
6642         }
6643
6644         if (btrfs_test_opt(root, DISCARD))
6645                 ret = btrfs_discard_extent(root, start, len, NULL);
6646
6647         if (pin)
6648                 pin_down_extent(root, cache, start, len, 1);
6649         else {
6650                 btrfs_add_free_space(cache, start, len);
6651                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6652         }
6653         btrfs_put_block_group(cache);
6654
6655         trace_btrfs_reserved_extent_free(root, start, len);
6656
6657         return ret;
6658 }
6659
6660 int btrfs_free_reserved_extent(struct btrfs_root *root,
6661                                         u64 start, u64 len)
6662 {
6663         return __btrfs_free_reserved_extent(root, start, len, 0);
6664 }
6665
6666 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6667                                        u64 start, u64 len)
6668 {
6669         return __btrfs_free_reserved_extent(root, start, len, 1);
6670 }
6671
6672 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6673                                       struct btrfs_root *root,
6674                                       u64 parent, u64 root_objectid,
6675                                       u64 flags, u64 owner, u64 offset,
6676                                       struct btrfs_key *ins, int ref_mod)
6677 {
6678         int ret;
6679         struct btrfs_fs_info *fs_info = root->fs_info;
6680         struct btrfs_extent_item *extent_item;
6681         struct btrfs_extent_inline_ref *iref;
6682         struct btrfs_path *path;
6683         struct extent_buffer *leaf;
6684         int type;
6685         u32 size;
6686
6687         if (parent > 0)
6688                 type = BTRFS_SHARED_DATA_REF_KEY;
6689         else
6690                 type = BTRFS_EXTENT_DATA_REF_KEY;
6691
6692         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6693
6694         path = btrfs_alloc_path();
6695         if (!path)
6696                 return -ENOMEM;
6697
6698         path->leave_spinning = 1;
6699         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6700                                       ins, size);
6701         if (ret) {
6702                 btrfs_free_path(path);
6703                 return ret;
6704         }
6705
6706         leaf = path->nodes[0];
6707         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6708                                      struct btrfs_extent_item);
6709         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6710         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6711         btrfs_set_extent_flags(leaf, extent_item,
6712                                flags | BTRFS_EXTENT_FLAG_DATA);
6713
6714         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6715         btrfs_set_extent_inline_ref_type(leaf, iref, type);
6716         if (parent > 0) {
6717                 struct btrfs_shared_data_ref *ref;
6718                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6719                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6720                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6721         } else {
6722                 struct btrfs_extent_data_ref *ref;
6723                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6724                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6725                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6726                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6727                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6728         }
6729
6730         btrfs_mark_buffer_dirty(path->nodes[0]);
6731         btrfs_free_path(path);
6732
6733         ret = update_block_group(root, ins->objectid, ins->offset, 1);
6734         if (ret) { /* -ENOENT, logic error */
6735                 btrfs_err(fs_info, "update block group failed for %llu %llu",
6736                         ins->objectid, ins->offset);
6737                 BUG();
6738         }
6739         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6740         return ret;
6741 }
6742
6743 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6744                                      struct btrfs_root *root,
6745                                      u64 parent, u64 root_objectid,
6746                                      u64 flags, struct btrfs_disk_key *key,
6747                                      int level, struct btrfs_key *ins)
6748 {
6749         int ret;
6750         struct btrfs_fs_info *fs_info = root->fs_info;
6751         struct btrfs_extent_item *extent_item;
6752         struct btrfs_tree_block_info *block_info;
6753         struct btrfs_extent_inline_ref *iref;
6754         struct btrfs_path *path;
6755         struct extent_buffer *leaf;
6756         u32 size = sizeof(*extent_item) + sizeof(*iref);
6757         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6758                                                  SKINNY_METADATA);
6759
6760         if (!skinny_metadata)
6761                 size += sizeof(*block_info);
6762
6763         path = btrfs_alloc_path();
6764         if (!path) {
6765                 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6766                                                    root->leafsize);
6767                 return -ENOMEM;
6768         }
6769
6770         path->leave_spinning = 1;
6771         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6772                                       ins, size);
6773         if (ret) {
6774                 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6775                                                    root->leafsize);
6776                 btrfs_free_path(path);
6777                 return ret;
6778         }
6779
6780         leaf = path->nodes[0];
6781         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6782                                      struct btrfs_extent_item);
6783         btrfs_set_extent_refs(leaf, extent_item, 1);
6784         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6785         btrfs_set_extent_flags(leaf, extent_item,
6786                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6787
6788         if (skinny_metadata) {
6789                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6790         } else {
6791                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6792                 btrfs_set_tree_block_key(leaf, block_info, key);
6793                 btrfs_set_tree_block_level(leaf, block_info, level);
6794                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6795         }
6796
6797         if (parent > 0) {
6798                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6799                 btrfs_set_extent_inline_ref_type(leaf, iref,
6800                                                  BTRFS_SHARED_BLOCK_REF_KEY);
6801                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6802         } else {
6803                 btrfs_set_extent_inline_ref_type(leaf, iref,
6804                                                  BTRFS_TREE_BLOCK_REF_KEY);
6805                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6806         }
6807
6808         btrfs_mark_buffer_dirty(leaf);
6809         btrfs_free_path(path);
6810
6811         ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6812         if (ret) { /* -ENOENT, logic error */
6813                 btrfs_err(fs_info, "update block group failed for %llu %llu",
6814                         ins->objectid, ins->offset);
6815                 BUG();
6816         }
6817
6818         trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6819         return ret;
6820 }
6821
6822 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6823                                      struct btrfs_root *root,
6824                                      u64 root_objectid, u64 owner,
6825                                      u64 offset, struct btrfs_key *ins)
6826 {
6827         int ret;
6828
6829         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6830
6831         ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6832                                          ins->offset, 0,
6833                                          root_objectid, owner, offset,
6834                                          BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6835         return ret;
6836 }
6837
6838 /*
6839  * this is used by the tree logging recovery code.  It records that
6840  * an extent has been allocated and makes sure to clear the free
6841  * space cache bits as well
6842  */
6843 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6844                                    struct btrfs_root *root,
6845                                    u64 root_objectid, u64 owner, u64 offset,
6846                                    struct btrfs_key *ins)
6847 {
6848         int ret;
6849         struct btrfs_block_group_cache *block_group;
6850
6851         /*
6852          * Mixed block groups will exclude before processing the log so we only
6853          * need to do the exlude dance if this fs isn't mixed.
6854          */
6855         if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6856                 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6857                 if (ret)
6858                         return ret;
6859         }
6860
6861         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6862         if (!block_group)
6863                 return -EINVAL;
6864
6865         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6866                                           RESERVE_ALLOC_NO_ACCOUNT);
6867         BUG_ON(ret); /* logic error */
6868         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6869                                          0, owner, offset, ins, 1);
6870         btrfs_put_block_group(block_group);
6871         return ret;
6872 }
6873
6874 static struct extent_buffer *
6875 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6876                       u64 bytenr, u32 blocksize, int level)
6877 {
6878         struct extent_buffer *buf;
6879
6880         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6881         if (!buf)
6882                 return ERR_PTR(-ENOMEM);
6883         btrfs_set_header_generation(buf, trans->transid);
6884         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6885         btrfs_tree_lock(buf);
6886         clean_tree_block(trans, root, buf);
6887         clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6888
6889         btrfs_set_lock_blocking(buf);
6890         btrfs_set_buffer_uptodate(buf);
6891
6892         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6893                 /*
6894                  * we allow two log transactions at a time, use different
6895                  * EXENT bit to differentiate dirty pages.
6896                  */
6897                 if (root->log_transid % 2 == 0)
6898                         set_extent_dirty(&root->dirty_log_pages, buf->start,
6899                                         buf->start + buf->len - 1, GFP_NOFS);
6900                 else
6901                         set_extent_new(&root->dirty_log_pages, buf->start,
6902                                         buf->start + buf->len - 1, GFP_NOFS);
6903         } else {
6904                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6905                          buf->start + buf->len - 1, GFP_NOFS);
6906         }
6907         trans->blocks_used++;
6908         /* this returns a buffer locked for blocking */
6909         return buf;
6910 }
6911
6912 static struct btrfs_block_rsv *
6913 use_block_rsv(struct btrfs_trans_handle *trans,
6914               struct btrfs_root *root, u32 blocksize)
6915 {
6916         struct btrfs_block_rsv *block_rsv;
6917         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6918         int ret;
6919         bool global_updated = false;
6920
6921         block_rsv = get_block_rsv(trans, root);
6922
6923         if (unlikely(block_rsv->size == 0))
6924                 goto try_reserve;
6925 again:
6926         ret = block_rsv_use_bytes(block_rsv, blocksize);
6927         if (!ret)
6928                 return block_rsv;
6929
6930         if (block_rsv->failfast)
6931                 return ERR_PTR(ret);
6932
6933         if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6934                 global_updated = true;
6935                 update_global_block_rsv(root->fs_info);
6936                 goto again;
6937         }
6938
6939         if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6940                 static DEFINE_RATELIMIT_STATE(_rs,
6941                                 DEFAULT_RATELIMIT_INTERVAL * 10,
6942                                 /*DEFAULT_RATELIMIT_BURST*/ 1);
6943                 if (__ratelimit(&_rs))
6944                         WARN(1, KERN_DEBUG
6945                                 "BTRFS: block rsv returned %d\n", ret);
6946         }
6947 try_reserve:
6948         ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6949                                      BTRFS_RESERVE_NO_FLUSH);
6950         if (!ret)
6951                 return block_rsv;
6952         /*
6953          * If we couldn't reserve metadata bytes try and use some from
6954          * the global reserve if its space type is the same as the global
6955          * reservation.
6956          */
6957         if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6958             block_rsv->space_info == global_rsv->space_info) {
6959                 ret = block_rsv_use_bytes(global_rsv, blocksize);
6960                 if (!ret)
6961                         return global_rsv;
6962         }
6963         return ERR_PTR(ret);
6964 }
6965
6966 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6967                             struct btrfs_block_rsv *block_rsv, u32 blocksize)
6968 {
6969         block_rsv_add_bytes(block_rsv, blocksize, 0);
6970         block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6971 }
6972
6973 /*
6974  * finds a free extent and does all the dirty work required for allocation
6975  * returns the key for the extent through ins, and a tree buffer for
6976  * the first block of the extent through buf.
6977  *
6978  * returns the tree buffer or NULL.
6979  */
6980 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6981                                         struct btrfs_root *root, u32 blocksize,
6982                                         u64 parent, u64 root_objectid,
6983                                         struct btrfs_disk_key *key, int level,
6984                                         u64 hint, u64 empty_size)
6985 {
6986         struct btrfs_key ins;
6987         struct btrfs_block_rsv *block_rsv;
6988         struct extent_buffer *buf;
6989         u64 flags = 0;
6990         int ret;
6991         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6992                                                  SKINNY_METADATA);
6993
6994         block_rsv = use_block_rsv(trans, root, blocksize);
6995         if (IS_ERR(block_rsv))
6996                 return ERR_CAST(block_rsv);
6997
6998         ret = btrfs_reserve_extent(root, blocksize, blocksize,
6999                                    empty_size, hint, &ins, 0);
7000         if (ret) {
7001                 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7002                 return ERR_PTR(ret);
7003         }
7004
7005         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7006                                     blocksize, level);
7007         BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7008
7009         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7010                 if (parent == 0)
7011                         parent = ins.objectid;
7012                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7013         } else
7014                 BUG_ON(parent > 0);
7015
7016         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7017                 struct btrfs_delayed_extent_op *extent_op;
7018                 extent_op = btrfs_alloc_delayed_extent_op();
7019                 BUG_ON(!extent_op); /* -ENOMEM */
7020                 if (key)
7021                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
7022                 else
7023                         memset(&extent_op->key, 0, sizeof(extent_op->key));
7024                 extent_op->flags_to_set = flags;
7025                 if (skinny_metadata)
7026                         extent_op->update_key = 0;
7027                 else
7028                         extent_op->update_key = 1;
7029                 extent_op->update_flags = 1;
7030                 extent_op->is_data = 0;
7031                 extent_op->level = level;
7032
7033                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7034                                         ins.objectid,
7035                                         ins.offset, parent, root_objectid,
7036                                         level, BTRFS_ADD_DELAYED_EXTENT,
7037                                         extent_op, 0);
7038                 BUG_ON(ret); /* -ENOMEM */
7039         }
7040         return buf;
7041 }
7042
7043 struct walk_control {
7044         u64 refs[BTRFS_MAX_LEVEL];
7045         u64 flags[BTRFS_MAX_LEVEL];
7046         struct btrfs_key update_progress;
7047         int stage;
7048         int level;
7049         int shared_level;
7050         int update_ref;
7051         int keep_locks;
7052         int reada_slot;
7053         int reada_count;
7054         int for_reloc;
7055 };
7056
7057 #define DROP_REFERENCE  1
7058 #define UPDATE_BACKREF  2
7059
7060 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7061                                      struct btrfs_root *root,
7062                                      struct walk_control *wc,
7063                                      struct btrfs_path *path)
7064 {
7065         u64 bytenr;
7066         u64 generation;
7067         u64 refs;
7068         u64 flags;
7069         u32 nritems;
7070         u32 blocksize;
7071         struct btrfs_key key;
7072         struct extent_buffer *eb;
7073         int ret;
7074         int slot;
7075         int nread = 0;
7076
7077         if (path->slots[wc->level] < wc->reada_slot) {
7078                 wc->reada_count = wc->reada_count * 2 / 3;
7079                 wc->reada_count = max(wc->reada_count, 2);
7080         } else {
7081                 wc->reada_count = wc->reada_count * 3 / 2;
7082                 wc->reada_count = min_t(int, wc->reada_count,
7083                                         BTRFS_NODEPTRS_PER_BLOCK(root));
7084         }
7085
7086         eb = path->nodes[wc->level];
7087         nritems = btrfs_header_nritems(eb);
7088         blocksize = btrfs_level_size(root, wc->level - 1);
7089
7090         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7091                 if (nread >= wc->reada_count)
7092                         break;
7093
7094                 cond_resched();
7095                 bytenr = btrfs_node_blockptr(eb, slot);
7096                 generation = btrfs_node_ptr_generation(eb, slot);
7097
7098                 if (slot == path->slots[wc->level])
7099                         goto reada;
7100
7101                 if (wc->stage == UPDATE_BACKREF &&
7102                     generation <= root->root_key.offset)
7103                         continue;
7104
7105                 /* We don't lock the tree block, it's OK to be racy here */
7106                 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7107                                                wc->level - 1, 1, &refs,
7108                                                &flags);
7109                 /* We don't care about errors in readahead. */
7110                 if (ret < 0)
7111                         continue;
7112                 BUG_ON(refs == 0);
7113
7114                 if (wc->stage == DROP_REFERENCE) {
7115                         if (refs == 1)
7116                                 goto reada;
7117
7118                         if (wc->level == 1 &&
7119                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7120                                 continue;
7121                         if (!wc->update_ref ||
7122                             generation <= root->root_key.offset)
7123                                 continue;
7124                         btrfs_node_key_to_cpu(eb, &key, slot);
7125                         ret = btrfs_comp_cpu_keys(&key,
7126                                                   &wc->update_progress);
7127                         if (ret < 0)
7128                                 continue;
7129                 } else {
7130                         if (wc->level == 1 &&
7131                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7132                                 continue;
7133                 }
7134 reada:
7135                 ret = readahead_tree_block(root, bytenr, blocksize,
7136                                            generation);
7137                 if (ret)
7138                         break;
7139                 nread++;
7140         }
7141         wc->reada_slot = slot;
7142 }
7143
7144 /*
7145  * helper to process tree block while walking down the tree.
7146  *
7147  * when wc->stage == UPDATE_BACKREF, this function updates
7148  * back refs for pointers in the block.
7149  *
7150  * NOTE: return value 1 means we should stop walking down.
7151  */
7152 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7153                                    struct btrfs_root *root,
7154                                    struct btrfs_path *path,
7155                                    struct walk_control *wc, int lookup_info)
7156 {
7157         int level = wc->level;
7158         struct extent_buffer *eb = path->nodes[level];
7159         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7160         int ret;
7161
7162         if (wc->stage == UPDATE_BACKREF &&
7163             btrfs_header_owner(eb) != root->root_key.objectid)
7164                 return 1;
7165
7166         /*
7167          * when reference count of tree block is 1, it won't increase
7168          * again. once full backref flag is set, we never clear it.
7169          */
7170         if (lookup_info &&
7171             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7172              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7173                 BUG_ON(!path->locks[level]);
7174                 ret = btrfs_lookup_extent_info(trans, root,
7175                                                eb->start, level, 1,
7176                                                &wc->refs[level],
7177                                                &wc->flags[level]);
7178                 BUG_ON(ret == -ENOMEM);
7179                 if (ret)
7180                         return ret;
7181                 BUG_ON(wc->refs[level] == 0);
7182         }
7183
7184         if (wc->stage == DROP_REFERENCE) {
7185                 if (wc->refs[level] > 1)
7186                         return 1;
7187
7188                 if (path->locks[level] && !wc->keep_locks) {
7189                         btrfs_tree_unlock_rw(eb, path->locks[level]);
7190                         path->locks[level] = 0;
7191                 }
7192                 return 0;
7193         }
7194
7195         /* wc->stage == UPDATE_BACKREF */
7196         if (!(wc->flags[level] & flag)) {
7197                 BUG_ON(!path->locks[level]);
7198                 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7199                 BUG_ON(ret); /* -ENOMEM */
7200                 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7201                 BUG_ON(ret); /* -ENOMEM */
7202                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7203                                                   eb->len, flag,
7204                                                   btrfs_header_level(eb), 0);
7205                 BUG_ON(ret); /* -ENOMEM */
7206                 wc->flags[level] |= flag;
7207         }
7208
7209         /*
7210          * the block is shared by multiple trees, so it's not good to
7211          * keep the tree lock
7212          */
7213         if (path->locks[level] && level > 0) {
7214                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7215                 path->locks[level] = 0;
7216         }
7217         return 0;
7218 }
7219
7220 /*
7221  * helper to process tree block pointer.
7222  *
7223  * when wc->stage == DROP_REFERENCE, this function checks
7224  * reference count of the block pointed to. if the block
7225  * is shared and we need update back refs for the subtree
7226  * rooted at the block, this function changes wc->stage to
7227  * UPDATE_BACKREF. if the block is shared and there is no
7228  * need to update back, this function drops the reference
7229  * to the block.
7230  *
7231  * NOTE: return value 1 means we should stop walking down.
7232  */
7233 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7234                                  struct btrfs_root *root,
7235                                  struct btrfs_path *path,
7236                                  struct walk_control *wc, int *lookup_info)
7237 {
7238         u64 bytenr;
7239         u64 generation;
7240         u64 parent;
7241         u32 blocksize;
7242         struct btrfs_key key;
7243         struct extent_buffer *next;
7244         int level = wc->level;
7245         int reada = 0;
7246         int ret = 0;
7247
7248         generation = btrfs_node_ptr_generation(path->nodes[level],
7249                                                path->slots[level]);
7250         /*
7251          * if the lower level block was created before the snapshot
7252          * was created, we know there is no need to update back refs
7253          * for the subtree
7254          */
7255         if (wc->stage == UPDATE_BACKREF &&
7256             generation <= root->root_key.offset) {
7257                 *lookup_info = 1;
7258                 return 1;
7259         }
7260
7261         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7262         blocksize = btrfs_level_size(root, level - 1);
7263
7264         next = btrfs_find_tree_block(root, bytenr, blocksize);
7265         if (!next) {
7266                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7267                 if (!next)
7268                         return -ENOMEM;
7269                 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7270                                                level - 1);
7271                 reada = 1;
7272         }
7273         btrfs_tree_lock(next);
7274         btrfs_set_lock_blocking(next);
7275
7276         ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7277                                        &wc->refs[level - 1],
7278                                        &wc->flags[level - 1]);
7279         if (ret < 0) {
7280                 btrfs_tree_unlock(next);
7281                 return ret;
7282         }
7283
7284         if (unlikely(wc->refs[level - 1] == 0)) {
7285                 btrfs_err(root->fs_info, "Missing references.");
7286                 BUG();
7287         }
7288         *lookup_info = 0;
7289
7290         if (wc->stage == DROP_REFERENCE) {
7291                 if (wc->refs[level - 1] > 1) {
7292                         if (level == 1 &&
7293                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7294                                 goto skip;
7295
7296                         if (!wc->update_ref ||
7297                             generation <= root->root_key.offset)
7298                                 goto skip;
7299
7300                         btrfs_node_key_to_cpu(path->nodes[level], &key,
7301                                               path->slots[level]);
7302                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7303                         if (ret < 0)
7304                                 goto skip;
7305
7306                         wc->stage = UPDATE_BACKREF;
7307                         wc->shared_level = level - 1;
7308                 }
7309         } else {
7310                 if (level == 1 &&
7311                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7312                         goto skip;
7313         }
7314
7315         if (!btrfs_buffer_uptodate(next, generation, 0)) {
7316                 btrfs_tree_unlock(next);
7317                 free_extent_buffer(next);
7318                 next = NULL;
7319                 *lookup_info = 1;
7320         }
7321
7322         if (!next) {
7323                 if (reada && level == 1)
7324                         reada_walk_down(trans, root, wc, path);
7325                 next = read_tree_block(root, bytenr, blocksize, generation);
7326                 if (!next || !extent_buffer_uptodate(next)) {
7327                         free_extent_buffer(next);
7328                         return -EIO;
7329                 }
7330                 btrfs_tree_lock(next);
7331                 btrfs_set_lock_blocking(next);
7332         }
7333
7334         level--;
7335         BUG_ON(level != btrfs_header_level(next));
7336         path->nodes[level] = next;
7337         path->slots[level] = 0;
7338         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7339         wc->level = level;
7340         if (wc->level == 1)
7341                 wc->reada_slot = 0;
7342         return 0;
7343 skip:
7344         wc->refs[level - 1] = 0;
7345         wc->flags[level - 1] = 0;
7346         if (wc->stage == DROP_REFERENCE) {
7347                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7348                         parent = path->nodes[level]->start;
7349                 } else {
7350                         BUG_ON(root->root_key.objectid !=
7351                                btrfs_header_owner(path->nodes[level]));
7352                         parent = 0;
7353                 }
7354
7355                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7356                                 root->root_key.objectid, level - 1, 0, 0);
7357                 BUG_ON(ret); /* -ENOMEM */
7358         }
7359         btrfs_tree_unlock(next);
7360         free_extent_buffer(next);
7361         *lookup_info = 1;
7362         return 1;
7363 }
7364
7365 /*
7366  * helper to process tree block while walking up the tree.
7367  *
7368  * when wc->stage == DROP_REFERENCE, this function drops
7369  * reference count on the block.
7370  *
7371  * when wc->stage == UPDATE_BACKREF, this function changes
7372  * wc->stage back to DROP_REFERENCE if we changed wc->stage
7373  * to UPDATE_BACKREF previously while processing the block.
7374  *
7375  * NOTE: return value 1 means we should stop walking up.
7376  */
7377 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7378                                  struct btrfs_root *root,
7379                                  struct btrfs_path *path,
7380                                  struct walk_control *wc)
7381 {
7382         int ret;
7383         int level = wc->level;
7384         struct extent_buffer *eb = path->nodes[level];
7385         u64 parent = 0;
7386
7387         if (wc->stage == UPDATE_BACKREF) {
7388                 BUG_ON(wc->shared_level < level);
7389                 if (level < wc->shared_level)
7390                         goto out;
7391
7392                 ret = find_next_key(path, level + 1, &wc->update_progress);
7393                 if (ret > 0)
7394                         wc->update_ref = 0;
7395
7396                 wc->stage = DROP_REFERENCE;
7397                 wc->shared_level = -1;
7398                 path->slots[level] = 0;
7399
7400                 /*
7401                  * check reference count again if the block isn't locked.
7402                  * we should start walking down the tree again if reference
7403                  * count is one.
7404                  */
7405                 if (!path->locks[level]) {
7406                         BUG_ON(level == 0);
7407                         btrfs_tree_lock(eb);
7408                         btrfs_set_lock_blocking(eb);
7409                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7410
7411                         ret = btrfs_lookup_extent_info(trans, root,
7412                                                        eb->start, level, 1,
7413                                                        &wc->refs[level],
7414                                                        &wc->flags[level]);
7415                         if (ret < 0) {
7416                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7417                                 path->locks[level] = 0;
7418                                 return ret;
7419                         }
7420                         BUG_ON(wc->refs[level] == 0);
7421                         if (wc->refs[level] == 1) {
7422                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7423                                 path->locks[level] = 0;
7424                                 return 1;
7425                         }
7426                 }
7427         }
7428
7429         /* wc->stage == DROP_REFERENCE */
7430         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7431
7432         if (wc->refs[level] == 1) {
7433                 if (level == 0) {
7434                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7435                                 ret = btrfs_dec_ref(trans, root, eb, 1,
7436                                                     wc->for_reloc);
7437                         else
7438                                 ret = btrfs_dec_ref(trans, root, eb, 0,
7439                                                     wc->for_reloc);
7440                         BUG_ON(ret); /* -ENOMEM */
7441                 }
7442                 /* make block locked assertion in clean_tree_block happy */
7443                 if (!path->locks[level] &&
7444                     btrfs_header_generation(eb) == trans->transid) {
7445                         btrfs_tree_lock(eb);
7446                         btrfs_set_lock_blocking(eb);
7447                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7448                 }
7449                 clean_tree_block(trans, root, eb);
7450         }
7451
7452         if (eb == root->node) {
7453                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7454                         parent = eb->start;
7455                 else
7456                         BUG_ON(root->root_key.objectid !=
7457                                btrfs_header_owner(eb));
7458         } else {
7459                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7460                         parent = path->nodes[level + 1]->start;
7461                 else
7462                         BUG_ON(root->root_key.objectid !=
7463                                btrfs_header_owner(path->nodes[level + 1]));
7464         }
7465
7466         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7467 out:
7468         wc->refs[level] = 0;
7469         wc->flags[level] = 0;
7470         return 0;
7471 }
7472
7473 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7474                                    struct btrfs_root *root,
7475                                    struct btrfs_path *path,
7476                                    struct walk_control *wc)
7477 {
7478         int level = wc->level;
7479         int lookup_info = 1;
7480         int ret;
7481
7482         while (level >= 0) {
7483                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7484                 if (ret > 0)
7485                         break;
7486
7487                 if (level == 0)
7488                         break;
7489
7490                 if (path->slots[level] >=
7491                     btrfs_header_nritems(path->nodes[level]))
7492                         break;
7493
7494                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7495                 if (ret > 0) {
7496                         path->slots[level]++;
7497                         continue;
7498                 } else if (ret < 0)
7499                         return ret;
7500                 level = wc->level;
7501         }
7502         return 0;
7503 }
7504
7505 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7506                                  struct btrfs_root *root,
7507                                  struct btrfs_path *path,
7508                                  struct walk_control *wc, int max_level)
7509 {
7510         int level = wc->level;
7511         int ret;
7512
7513         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7514         while (level < max_level && path->nodes[level]) {
7515                 wc->level = level;
7516                 if (path->slots[level] + 1 <
7517                     btrfs_header_nritems(path->nodes[level])) {
7518                         path->slots[level]++;
7519                         return 0;
7520                 } else {
7521                         ret = walk_up_proc(trans, root, path, wc);
7522                         if (ret > 0)
7523                                 return 0;
7524
7525                         if (path->locks[level]) {
7526                                 btrfs_tree_unlock_rw(path->nodes[level],
7527                                                      path->locks[level]);
7528                                 path->locks[level] = 0;
7529                         }
7530                         free_extent_buffer(path->nodes[level]);
7531                         path->nodes[level] = NULL;
7532                         level++;
7533                 }
7534         }
7535         return 1;
7536 }
7537
7538 /*
7539  * drop a subvolume tree.
7540  *
7541  * this function traverses the tree freeing any blocks that only
7542  * referenced by the tree.
7543  *
7544  * when a shared tree block is found. this function decreases its
7545  * reference count by one. if update_ref is true, this function
7546  * also make sure backrefs for the shared block and all lower level
7547  * blocks are properly updated.
7548  *
7549  * If called with for_reloc == 0, may exit early with -EAGAIN
7550  */
7551 int btrfs_drop_snapshot(struct btrfs_root *root,
7552                          struct btrfs_block_rsv *block_rsv, int update_ref,
7553                          int for_reloc)
7554 {
7555         struct btrfs_path *path;
7556         struct btrfs_trans_handle *trans;
7557         struct btrfs_root *tree_root = root->fs_info->tree_root;
7558         struct btrfs_root_item *root_item = &root->root_item;
7559         struct walk_control *wc;
7560         struct btrfs_key key;
7561         int err = 0;
7562         int ret;
7563         int level;
7564         bool root_dropped = false;
7565
7566         path = btrfs_alloc_path();
7567         if (!path) {
7568                 err = -ENOMEM;
7569                 goto out;
7570         }
7571
7572         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7573         if (!wc) {
7574                 btrfs_free_path(path);
7575                 err = -ENOMEM;
7576                 goto out;
7577         }
7578
7579         trans = btrfs_start_transaction(tree_root, 0);
7580         if (IS_ERR(trans)) {
7581                 err = PTR_ERR(trans);
7582                 goto out_free;
7583         }
7584
7585         if (block_rsv)
7586                 trans->block_rsv = block_rsv;
7587
7588         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7589                 level = btrfs_header_level(root->node);
7590                 path->nodes[level] = btrfs_lock_root_node(root);
7591                 btrfs_set_lock_blocking(path->nodes[level]);
7592                 path->slots[level] = 0;
7593                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7594                 memset(&wc->update_progress, 0,
7595                        sizeof(wc->update_progress));
7596         } else {
7597                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7598                 memcpy(&wc->update_progress, &key,
7599                        sizeof(wc->update_progress));
7600
7601                 level = root_item->drop_level;
7602                 BUG_ON(level == 0);
7603                 path->lowest_level = level;
7604                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7605                 path->lowest_level = 0;
7606                 if (ret < 0) {
7607                         err = ret;
7608                         goto out_end_trans;
7609                 }
7610                 WARN_ON(ret > 0);
7611
7612                 /*
7613                  * unlock our path, this is safe because only this
7614                  * function is allowed to delete this snapshot
7615                  */
7616                 btrfs_unlock_up_safe(path, 0);
7617
7618                 level = btrfs_header_level(root->node);
7619                 while (1) {
7620                         btrfs_tree_lock(path->nodes[level]);
7621                         btrfs_set_lock_blocking(path->nodes[level]);
7622                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7623
7624                         ret = btrfs_lookup_extent_info(trans, root,
7625                                                 path->nodes[level]->start,
7626                                                 level, 1, &wc->refs[level],
7627                                                 &wc->flags[level]);
7628                         if (ret < 0) {
7629                                 err = ret;
7630                                 goto out_end_trans;
7631                         }
7632                         BUG_ON(wc->refs[level] == 0);
7633
7634                         if (level == root_item->drop_level)
7635                                 break;
7636
7637                         btrfs_tree_unlock(path->nodes[level]);
7638                         path->locks[level] = 0;
7639                         WARN_ON(wc->refs[level] != 1);
7640                         level--;
7641                 }
7642         }
7643
7644         wc->level = level;
7645         wc->shared_level = -1;
7646         wc->stage = DROP_REFERENCE;
7647         wc->update_ref = update_ref;
7648         wc->keep_locks = 0;
7649         wc->for_reloc = for_reloc;
7650         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7651
7652         while (1) {
7653
7654                 ret = walk_down_tree(trans, root, path, wc);
7655                 if (ret < 0) {
7656                         err = ret;
7657                         break;
7658                 }
7659
7660                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7661                 if (ret < 0) {
7662                         err = ret;
7663                         break;
7664                 }
7665
7666                 if (ret > 0) {
7667                         BUG_ON(wc->stage != DROP_REFERENCE);
7668                         break;
7669                 }
7670
7671                 if (wc->stage == DROP_REFERENCE) {
7672                         level = wc->level;
7673                         btrfs_node_key(path->nodes[level],
7674                                        &root_item->drop_progress,
7675                                        path->slots[level]);
7676                         root_item->drop_level = level;
7677                 }
7678
7679                 BUG_ON(wc->level == 0);
7680                 if (btrfs_should_end_transaction(trans, tree_root) ||
7681                     (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7682                         ret = btrfs_update_root(trans, tree_root,
7683                                                 &root->root_key,
7684                                                 root_item);
7685                         if (ret) {
7686                                 btrfs_abort_transaction(trans, tree_root, ret);
7687                                 err = ret;
7688                                 goto out_end_trans;
7689                         }
7690
7691                         btrfs_end_transaction_throttle(trans, tree_root);
7692                         if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7693                                 pr_debug("BTRFS: drop snapshot early exit\n");
7694                                 err = -EAGAIN;
7695                                 goto out_free;
7696                         }
7697
7698                         trans = btrfs_start_transaction(tree_root, 0);
7699                         if (IS_ERR(trans)) {
7700                                 err = PTR_ERR(trans);
7701                                 goto out_free;
7702                         }
7703                         if (block_rsv)
7704                                 trans->block_rsv = block_rsv;
7705                 }
7706         }
7707         btrfs_release_path(path);
7708         if (err)
7709                 goto out_end_trans;
7710
7711         ret = btrfs_del_root(trans, tree_root, &root->root_key);
7712         if (ret) {
7713                 btrfs_abort_transaction(trans, tree_root, ret);
7714                 goto out_end_trans;
7715         }
7716
7717         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7718                 ret = btrfs_find_root(tree_root, &root->root_key, path,
7719                                       NULL, NULL);
7720                 if (ret < 0) {
7721                         btrfs_abort_transaction(trans, tree_root, ret);
7722                         err = ret;
7723                         goto out_end_trans;
7724                 } else if (ret > 0) {
7725                         /* if we fail to delete the orphan item this time
7726                          * around, it'll get picked up the next time.
7727                          *
7728                          * The most common failure here is just -ENOENT.
7729                          */
7730                         btrfs_del_orphan_item(trans, tree_root,
7731                                               root->root_key.objectid);
7732                 }
7733         }
7734
7735         if (root->in_radix) {
7736                 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7737         } else {
7738                 free_extent_buffer(root->node);
7739                 free_extent_buffer(root->commit_root);
7740                 btrfs_put_fs_root(root);
7741         }
7742         root_dropped = true;
7743 out_end_trans:
7744         btrfs_end_transaction_throttle(trans, tree_root);
7745 out_free:
7746         kfree(wc);
7747         btrfs_free_path(path);
7748 out:
7749         /*
7750          * So if we need to stop dropping the snapshot for whatever reason we
7751          * need to make sure to add it back to the dead root list so that we
7752          * keep trying to do the work later.  This also cleans up roots if we
7753          * don't have it in the radix (like when we recover after a power fail
7754          * or unmount) so we don't leak memory.
7755          */
7756         if (!for_reloc && root_dropped == false)
7757                 btrfs_add_dead_root(root);
7758         if (err && err != -EAGAIN)
7759                 btrfs_std_error(root->fs_info, err);
7760         return err;
7761 }
7762
7763 /*
7764  * drop subtree rooted at tree block 'node'.
7765  *
7766  * NOTE: this function will unlock and release tree block 'node'
7767  * only used by relocation code
7768  */
7769 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7770                         struct btrfs_root *root,
7771                         struct extent_buffer *node,
7772                         struct extent_buffer *parent)
7773 {
7774         struct btrfs_path *path;
7775         struct walk_control *wc;
7776         int level;
7777         int parent_level;
7778         int ret = 0;
7779         int wret;
7780
7781         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7782
7783         path = btrfs_alloc_path();
7784         if (!path)
7785                 return -ENOMEM;
7786
7787         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7788         if (!wc) {
7789                 btrfs_free_path(path);
7790                 return -ENOMEM;
7791         }
7792
7793         btrfs_assert_tree_locked(parent);
7794         parent_level = btrfs_header_level(parent);
7795         extent_buffer_get(parent);
7796         path->nodes[parent_level] = parent;
7797         path->slots[parent_level] = btrfs_header_nritems(parent);
7798
7799         btrfs_assert_tree_locked(node);
7800         level = btrfs_header_level(node);
7801         path->nodes[level] = node;
7802         path->slots[level] = 0;
7803         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7804
7805         wc->refs[parent_level] = 1;
7806         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7807         wc->level = level;
7808         wc->shared_level = -1;
7809         wc->stage = DROP_REFERENCE;
7810         wc->update_ref = 0;
7811         wc->keep_locks = 1;
7812         wc->for_reloc = 1;
7813         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7814
7815         while (1) {
7816                 wret = walk_down_tree(trans, root, path, wc);
7817                 if (wret < 0) {
7818                         ret = wret;
7819                         break;
7820                 }
7821
7822                 wret = walk_up_tree(trans, root, path, wc, parent_level);
7823                 if (wret < 0)
7824                         ret = wret;
7825                 if (wret != 0)
7826                         break;
7827         }
7828
7829         kfree(wc);
7830         btrfs_free_path(path);
7831         return ret;
7832 }
7833
7834 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7835 {
7836         u64 num_devices;
7837         u64 stripped;
7838
7839         /*
7840          * if restripe for this chunk_type is on pick target profile and
7841          * return, otherwise do the usual balance
7842          */
7843         stripped = get_restripe_target(root->fs_info, flags);
7844         if (stripped)
7845                 return extended_to_chunk(stripped);
7846
7847         /*
7848          * we add in the count of missing devices because we want
7849          * to make sure that any RAID levels on a degraded FS
7850          * continue to be honored.
7851          */
7852         num_devices = root->fs_info->fs_devices->rw_devices +
7853                 root->fs_info->fs_devices->missing_devices;
7854
7855         stripped = BTRFS_BLOCK_GROUP_RAID0 |
7856                 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7857                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7858
7859         if (num_devices == 1) {
7860                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7861                 stripped = flags & ~stripped;
7862
7863                 /* turn raid0 into single device chunks */
7864                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7865                         return stripped;
7866
7867                 /* turn mirroring into duplication */
7868                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7869                              BTRFS_BLOCK_GROUP_RAID10))
7870                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7871         } else {
7872                 /* they already had raid on here, just return */
7873                 if (flags & stripped)
7874                         return flags;
7875
7876                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7877                 stripped = flags & ~stripped;
7878
7879                 /* switch duplicated blocks with raid1 */
7880                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7881                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7882
7883                 /* this is drive concat, leave it alone */
7884         }
7885
7886         return flags;
7887 }
7888
7889 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7890 {
7891         struct btrfs_space_info *sinfo = cache->space_info;
7892         u64 num_bytes;
7893         u64 min_allocable_bytes;
7894         int ret = -ENOSPC;
7895
7896
7897         /*
7898          * We need some metadata space and system metadata space for
7899          * allocating chunks in some corner cases until we force to set
7900          * it to be readonly.
7901          */
7902         if ((sinfo->flags &
7903              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7904             !force)
7905                 min_allocable_bytes = 1 * 1024 * 1024;
7906         else
7907                 min_allocable_bytes = 0;
7908
7909         spin_lock(&sinfo->lock);
7910         spin_lock(&cache->lock);
7911
7912         if (cache->ro) {
7913                 ret = 0;
7914                 goto out;
7915         }
7916
7917         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7918                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7919
7920         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7921             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7922             min_allocable_bytes <= sinfo->total_bytes) {
7923                 sinfo->bytes_readonly += num_bytes;
7924                 cache->ro = 1;
7925                 ret = 0;
7926         }
7927 out:
7928         spin_unlock(&cache->lock);
7929         spin_unlock(&sinfo->lock);
7930         return ret;
7931 }
7932
7933 int btrfs_set_block_group_ro(struct btrfs_root *root,
7934                              struct btrfs_block_group_cache *cache)
7935
7936 {
7937         struct btrfs_trans_handle *trans;
7938         u64 alloc_flags;
7939         int ret;
7940
7941         BUG_ON(cache->ro);
7942
7943         trans = btrfs_join_transaction(root);
7944         if (IS_ERR(trans))
7945                 return PTR_ERR(trans);
7946
7947         alloc_flags = update_block_group_flags(root, cache->flags);
7948         if (alloc_flags != cache->flags) {
7949                 ret = do_chunk_alloc(trans, root, alloc_flags,
7950                                      CHUNK_ALLOC_FORCE);
7951                 if (ret < 0)
7952                         goto out;
7953         }
7954
7955         ret = set_block_group_ro(cache, 0);
7956         if (!ret)
7957                 goto out;
7958         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7959         ret = do_chunk_alloc(trans, root, alloc_flags,
7960                              CHUNK_ALLOC_FORCE);
7961         if (ret < 0)
7962                 goto out;
7963         ret = set_block_group_ro(cache, 0);
7964 out:
7965         btrfs_end_transaction(trans, root);
7966         return ret;
7967 }
7968
7969 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7970                             struct btrfs_root *root, u64 type)
7971 {
7972         u64 alloc_flags = get_alloc_profile(root, type);
7973         return do_chunk_alloc(trans, root, alloc_flags,
7974                               CHUNK_ALLOC_FORCE);
7975 }
7976
7977 /*
7978  * helper to account the unused space of all the readonly block group in the
7979  * list. takes mirrors into account.
7980  */
7981 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7982 {
7983         struct btrfs_block_group_cache *block_group;
7984         u64 free_bytes = 0;
7985         int factor;
7986
7987         list_for_each_entry(block_group, groups_list, list) {
7988                 spin_lock(&block_group->lock);
7989
7990                 if (!block_group->ro) {
7991                         spin_unlock(&block_group->lock);
7992                         continue;
7993                 }
7994
7995                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7996                                           BTRFS_BLOCK_GROUP_RAID10 |
7997                                           BTRFS_BLOCK_GROUP_DUP))
7998                         factor = 2;
7999                 else
8000                         factor = 1;
8001
8002                 free_bytes += (block_group->key.offset -
8003                                btrfs_block_group_used(&block_group->item)) *
8004                                factor;
8005
8006                 spin_unlock(&block_group->lock);
8007         }
8008
8009         return free_bytes;
8010 }
8011
8012 /*
8013  * helper to account the unused space of all the readonly block group in the
8014  * space_info. takes mirrors into account.
8015  */
8016 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8017 {
8018         int i;
8019         u64 free_bytes = 0;
8020
8021         spin_lock(&sinfo->lock);
8022
8023         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8024                 if (!list_empty(&sinfo->block_groups[i]))
8025                         free_bytes += __btrfs_get_ro_block_group_free_space(
8026                                                 &sinfo->block_groups[i]);
8027
8028         spin_unlock(&sinfo->lock);
8029
8030         return free_bytes;
8031 }
8032
8033 void btrfs_set_block_group_rw(struct btrfs_root *root,
8034                               struct btrfs_block_group_cache *cache)
8035 {
8036         struct btrfs_space_info *sinfo = cache->space_info;
8037         u64 num_bytes;
8038
8039         BUG_ON(!cache->ro);
8040
8041         spin_lock(&sinfo->lock);
8042         spin_lock(&cache->lock);
8043         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8044                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8045         sinfo->bytes_readonly -= num_bytes;
8046         cache->ro = 0;
8047         spin_unlock(&cache->lock);
8048         spin_unlock(&sinfo->lock);
8049 }
8050
8051 /*
8052  * checks to see if its even possible to relocate this block group.
8053  *
8054  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8055  * ok to go ahead and try.
8056  */
8057 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8058 {
8059         struct btrfs_block_group_cache *block_group;
8060         struct btrfs_space_info *space_info;
8061         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8062         struct btrfs_device *device;
8063         struct btrfs_trans_handle *trans;
8064         u64 min_free;
8065         u64 dev_min = 1;
8066         u64 dev_nr = 0;
8067         u64 target;
8068         int index;
8069         int full = 0;
8070         int ret = 0;
8071
8072         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8073
8074         /* odd, couldn't find the block group, leave it alone */
8075         if (!block_group)
8076                 return -1;
8077
8078         min_free = btrfs_block_group_used(&block_group->item);
8079
8080         /* no bytes used, we're good */
8081         if (!min_free)
8082                 goto out;
8083
8084         space_info = block_group->space_info;
8085         spin_lock(&space_info->lock);
8086
8087         full = space_info->full;
8088
8089         /*
8090          * if this is the last block group we have in this space, we can't
8091          * relocate it unless we're able to allocate a new chunk below.
8092          *
8093          * Otherwise, we need to make sure we have room in the space to handle
8094          * all of the extents from this block group.  If we can, we're good
8095          */
8096         if ((space_info->total_bytes != block_group->key.offset) &&
8097             (space_info->bytes_used + space_info->bytes_reserved +
8098              space_info->bytes_pinned + space_info->bytes_readonly +
8099              min_free < space_info->total_bytes)) {
8100                 spin_unlock(&space_info->lock);
8101                 goto out;
8102         }
8103         spin_unlock(&space_info->lock);
8104
8105         /*
8106          * ok we don't have enough space, but maybe we have free space on our
8107          * devices to allocate new chunks for relocation, so loop through our
8108          * alloc devices and guess if we have enough space.  if this block
8109          * group is going to be restriped, run checks against the target
8110          * profile instead of the current one.
8111          */
8112         ret = -1;
8113
8114         /*
8115          * index:
8116          *      0: raid10
8117          *      1: raid1
8118          *      2: dup
8119          *      3: raid0
8120          *      4: single
8121          */
8122         target = get_restripe_target(root->fs_info, block_group->flags);
8123         if (target) {
8124                 index = __get_raid_index(extended_to_chunk(target));
8125         } else {
8126                 /*
8127                  * this is just a balance, so if we were marked as full
8128                  * we know there is no space for a new chunk
8129                  */
8130                 if (full)
8131                         goto out;
8132
8133                 index = get_block_group_index(block_group);
8134         }
8135
8136         if (index == BTRFS_RAID_RAID10) {
8137                 dev_min = 4;
8138                 /* Divide by 2 */
8139                 min_free >>= 1;
8140         } else if (index == BTRFS_RAID_RAID1) {
8141                 dev_min = 2;
8142         } else if (index == BTRFS_RAID_DUP) {
8143                 /* Multiply by 2 */
8144                 min_free <<= 1;
8145         } else if (index == BTRFS_RAID_RAID0) {
8146                 dev_min = fs_devices->rw_devices;
8147                 do_div(min_free, dev_min);
8148         }
8149
8150         /* We need to do this so that we can look at pending chunks */
8151         trans = btrfs_join_transaction(root);
8152         if (IS_ERR(trans)) {
8153                 ret = PTR_ERR(trans);
8154                 goto out;
8155         }
8156
8157         mutex_lock(&root->fs_info->chunk_mutex);
8158         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8159                 u64 dev_offset;
8160
8161                 /*
8162                  * check to make sure we can actually find a chunk with enough
8163                  * space to fit our block group in.
8164                  */
8165                 if (device->total_bytes > device->bytes_used + min_free &&
8166                     !device->is_tgtdev_for_dev_replace) {
8167                         ret = find_free_dev_extent(trans, device, min_free,
8168                                                    &dev_offset, NULL);
8169                         if (!ret)
8170                                 dev_nr++;
8171
8172                         if (dev_nr >= dev_min)
8173                                 break;
8174
8175                         ret = -1;
8176                 }
8177         }
8178         mutex_unlock(&root->fs_info->chunk_mutex);
8179         btrfs_end_transaction(trans, root);
8180 out:
8181         btrfs_put_block_group(block_group);
8182         return ret;
8183 }
8184
8185 static int find_first_block_group(struct btrfs_root *root,
8186                 struct btrfs_path *path, struct btrfs_key *key)
8187 {
8188         int ret = 0;
8189         struct btrfs_key found_key;
8190         struct extent_buffer *leaf;
8191         int slot;
8192
8193         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8194         if (ret < 0)
8195                 goto out;
8196
8197         while (1) {
8198                 slot = path->slots[0];
8199                 leaf = path->nodes[0];
8200                 if (slot >= btrfs_header_nritems(leaf)) {
8201                         ret = btrfs_next_leaf(root, path);
8202                         if (ret == 0)
8203                                 continue;
8204                         if (ret < 0)
8205                                 goto out;
8206                         break;
8207                 }
8208                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8209
8210                 if (found_key.objectid >= key->objectid &&
8211                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8212                         ret = 0;
8213                         goto out;
8214                 }
8215                 path->slots[0]++;
8216         }
8217 out:
8218         return ret;
8219 }
8220
8221 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8222 {
8223         struct btrfs_block_group_cache *block_group;
8224         u64 last = 0;
8225
8226         while (1) {
8227                 struct inode *inode;
8228
8229                 block_group = btrfs_lookup_first_block_group(info, last);
8230                 while (block_group) {
8231                         spin_lock(&block_group->lock);
8232                         if (block_group->iref)
8233                                 break;
8234                         spin_unlock(&block_group->lock);
8235                         block_group = next_block_group(info->tree_root,
8236                                                        block_group);
8237                 }
8238                 if (!block_group) {
8239                         if (last == 0)
8240                                 break;
8241                         last = 0;
8242                         continue;
8243                 }
8244
8245                 inode = block_group->inode;
8246                 block_group->iref = 0;
8247                 block_group->inode = NULL;
8248                 spin_unlock(&block_group->lock);
8249                 iput(inode);
8250                 last = block_group->key.objectid + block_group->key.offset;
8251                 btrfs_put_block_group(block_group);
8252         }
8253 }
8254
8255 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8256 {
8257         struct btrfs_block_group_cache *block_group;
8258         struct btrfs_space_info *space_info;
8259         struct btrfs_caching_control *caching_ctl;
8260         struct rb_node *n;
8261
8262         down_write(&info->extent_commit_sem);
8263         while (!list_empty(&info->caching_block_groups)) {
8264                 caching_ctl = list_entry(info->caching_block_groups.next,
8265                                          struct btrfs_caching_control, list);
8266                 list_del(&caching_ctl->list);
8267                 put_caching_control(caching_ctl);
8268         }
8269         up_write(&info->extent_commit_sem);
8270
8271         spin_lock(&info->block_group_cache_lock);
8272         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8273                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8274                                        cache_node);
8275                 rb_erase(&block_group->cache_node,
8276                          &info->block_group_cache_tree);
8277                 spin_unlock(&info->block_group_cache_lock);
8278
8279                 down_write(&block_group->space_info->groups_sem);
8280                 list_del(&block_group->list);
8281                 up_write(&block_group->space_info->groups_sem);
8282
8283                 if (block_group->cached == BTRFS_CACHE_STARTED)
8284                         wait_block_group_cache_done(block_group);
8285
8286                 /*
8287                  * We haven't cached this block group, which means we could
8288                  * possibly have excluded extents on this block group.
8289                  */
8290                 if (block_group->cached == BTRFS_CACHE_NO ||
8291                     block_group->cached == BTRFS_CACHE_ERROR)
8292                         free_excluded_extents(info->extent_root, block_group);
8293
8294                 btrfs_remove_free_space_cache(block_group);
8295                 btrfs_put_block_group(block_group);
8296
8297                 spin_lock(&info->block_group_cache_lock);
8298         }
8299         spin_unlock(&info->block_group_cache_lock);
8300
8301         /* now that all the block groups are freed, go through and
8302          * free all the space_info structs.  This is only called during
8303          * the final stages of unmount, and so we know nobody is
8304          * using them.  We call synchronize_rcu() once before we start,
8305          * just to be on the safe side.
8306          */
8307         synchronize_rcu();
8308
8309         release_global_block_rsv(info);
8310
8311         while (!list_empty(&info->space_info)) {
8312                 int i;
8313
8314                 space_info = list_entry(info->space_info.next,
8315                                         struct btrfs_space_info,
8316                                         list);
8317                 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8318                         if (WARN_ON(space_info->bytes_pinned > 0 ||
8319                             space_info->bytes_reserved > 0 ||
8320                             space_info->bytes_may_use > 0)) {
8321                                 dump_space_info(space_info, 0, 0);
8322                         }
8323                 }
8324                 list_del(&space_info->list);
8325                 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8326                         struct kobject *kobj;
8327                         kobj = &space_info->block_group_kobjs[i];
8328                         if (kobj->parent) {
8329                                 kobject_del(kobj);
8330                                 kobject_put(kobj);
8331                         }
8332                 }
8333                 kobject_del(&space_info->kobj);
8334                 kobject_put(&space_info->kobj);
8335         }
8336         return 0;
8337 }
8338
8339 static void __link_block_group(struct btrfs_space_info *space_info,
8340                                struct btrfs_block_group_cache *cache)
8341 {
8342         int index = get_block_group_index(cache);
8343
8344         down_write(&space_info->groups_sem);
8345         if (list_empty(&space_info->block_groups[index])) {
8346                 struct kobject *kobj = &space_info->block_group_kobjs[index];
8347                 int ret;
8348
8349                 kobject_get(&space_info->kobj); /* put in release */
8350                 ret = kobject_add(kobj, &space_info->kobj, "%s",
8351                                   get_raid_name(index));
8352                 if (ret) {
8353                         pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8354                         kobject_put(&space_info->kobj);
8355                 }
8356         }
8357         list_add_tail(&cache->list, &space_info->block_groups[index]);
8358         up_write(&space_info->groups_sem);
8359 }
8360
8361 static struct btrfs_block_group_cache *
8362 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8363 {
8364         struct btrfs_block_group_cache *cache;
8365
8366         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8367         if (!cache)
8368                 return NULL;
8369
8370         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8371                                         GFP_NOFS);
8372         if (!cache->free_space_ctl) {
8373                 kfree(cache);
8374                 return NULL;
8375         }
8376
8377         cache->key.objectid = start;
8378         cache->key.offset = size;
8379         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8380
8381         cache->sectorsize = root->sectorsize;
8382         cache->fs_info = root->fs_info;
8383         cache->full_stripe_len = btrfs_full_stripe_len(root,
8384                                                &root->fs_info->mapping_tree,
8385                                                start);
8386         atomic_set(&cache->count, 1);
8387         spin_lock_init(&cache->lock);
8388         INIT_LIST_HEAD(&cache->list);
8389         INIT_LIST_HEAD(&cache->cluster_list);
8390         INIT_LIST_HEAD(&cache->new_bg_list);
8391         btrfs_init_free_space_ctl(cache);
8392
8393         return cache;
8394 }
8395
8396 int btrfs_read_block_groups(struct btrfs_root *root)
8397 {
8398         struct btrfs_path *path;
8399         int ret;
8400         struct btrfs_block_group_cache *cache;
8401         struct btrfs_fs_info *info = root->fs_info;
8402         struct btrfs_space_info *space_info;
8403         struct btrfs_key key;
8404         struct btrfs_key found_key;
8405         struct extent_buffer *leaf;
8406         int need_clear = 0;
8407         u64 cache_gen;
8408
8409         root = info->extent_root;
8410         key.objectid = 0;
8411         key.offset = 0;
8412         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8413         path = btrfs_alloc_path();
8414         if (!path)
8415                 return -ENOMEM;
8416         path->reada = 1;
8417
8418         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8419         if (btrfs_test_opt(root, SPACE_CACHE) &&
8420             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8421                 need_clear = 1;
8422         if (btrfs_test_opt(root, CLEAR_CACHE))
8423                 need_clear = 1;
8424
8425         while (1) {
8426                 ret = find_first_block_group(root, path, &key);
8427                 if (ret > 0)
8428                         break;
8429                 if (ret != 0)
8430                         goto error;
8431
8432                 leaf = path->nodes[0];
8433                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8434
8435                 cache = btrfs_create_block_group_cache(root, found_key.objectid,
8436                                                        found_key.offset);
8437                 if (!cache) {
8438                         ret = -ENOMEM;
8439                         goto error;
8440                 }
8441
8442                 if (need_clear) {
8443                         /*
8444                          * When we mount with old space cache, we need to
8445                          * set BTRFS_DC_CLEAR and set dirty flag.
8446                          *
8447                          * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8448                          *    truncate the old free space cache inode and
8449                          *    setup a new one.
8450                          * b) Setting 'dirty flag' makes sure that we flush
8451                          *    the new space cache info onto disk.
8452                          */
8453                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8454                         if (btrfs_test_opt(root, SPACE_CACHE))
8455                                 cache->dirty = 1;
8456                 }
8457
8458                 read_extent_buffer(leaf, &cache->item,
8459                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8460                                    sizeof(cache->item));
8461                 cache->flags = btrfs_block_group_flags(&cache->item);
8462
8463                 key.objectid = found_key.objectid + found_key.offset;
8464                 btrfs_release_path(path);
8465
8466                 /*
8467                  * We need to exclude the super stripes now so that the space
8468                  * info has super bytes accounted for, otherwise we'll think
8469                  * we have more space than we actually do.
8470                  */
8471                 ret = exclude_super_stripes(root, cache);
8472                 if (ret) {
8473                         /*
8474                          * We may have excluded something, so call this just in
8475                          * case.
8476                          */
8477                         free_excluded_extents(root, cache);
8478                         btrfs_put_block_group(cache);
8479                         goto error;
8480                 }
8481
8482                 /*
8483                  * check for two cases, either we are full, and therefore
8484                  * don't need to bother with the caching work since we won't
8485                  * find any space, or we are empty, and we can just add all
8486                  * the space in and be done with it.  This saves us _alot_ of
8487                  * time, particularly in the full case.
8488                  */
8489                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8490                         cache->last_byte_to_unpin = (u64)-1;
8491                         cache->cached = BTRFS_CACHE_FINISHED;
8492                         free_excluded_extents(root, cache);
8493                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8494                         cache->last_byte_to_unpin = (u64)-1;
8495                         cache->cached = BTRFS_CACHE_FINISHED;
8496                         add_new_free_space(cache, root->fs_info,
8497                                            found_key.objectid,
8498                                            found_key.objectid +
8499                                            found_key.offset);
8500                         free_excluded_extents(root, cache);
8501                 }
8502
8503                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8504                 if (ret) {
8505                         btrfs_remove_free_space_cache(cache);
8506                         btrfs_put_block_group(cache);
8507                         goto error;
8508                 }
8509
8510                 ret = update_space_info(info, cache->flags, found_key.offset,
8511                                         btrfs_block_group_used(&cache->item),
8512                                         &space_info);
8513                 if (ret) {
8514                         btrfs_remove_free_space_cache(cache);
8515                         spin_lock(&info->block_group_cache_lock);
8516                         rb_erase(&cache->cache_node,
8517                                  &info->block_group_cache_tree);
8518                         spin_unlock(&info->block_group_cache_lock);
8519                         btrfs_put_block_group(cache);
8520                         goto error;
8521                 }
8522
8523                 cache->space_info = space_info;
8524                 spin_lock(&cache->space_info->lock);
8525                 cache->space_info->bytes_readonly += cache->bytes_super;
8526                 spin_unlock(&cache->space_info->lock);
8527
8528                 __link_block_group(space_info, cache);
8529
8530                 set_avail_alloc_bits(root->fs_info, cache->flags);
8531                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8532                         set_block_group_ro(cache, 1);
8533         }
8534
8535         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8536                 if (!(get_alloc_profile(root, space_info->flags) &
8537                       (BTRFS_BLOCK_GROUP_RAID10 |
8538                        BTRFS_BLOCK_GROUP_RAID1 |
8539                        BTRFS_BLOCK_GROUP_RAID5 |
8540                        BTRFS_BLOCK_GROUP_RAID6 |
8541                        BTRFS_BLOCK_GROUP_DUP)))
8542                         continue;
8543                 /*
8544                  * avoid allocating from un-mirrored block group if there are
8545                  * mirrored block groups.
8546                  */
8547                 list_for_each_entry(cache,
8548                                 &space_info->block_groups[BTRFS_RAID_RAID0],
8549                                 list)
8550                         set_block_group_ro(cache, 1);
8551                 list_for_each_entry(cache,
8552                                 &space_info->block_groups[BTRFS_RAID_SINGLE],
8553                                 list)
8554                         set_block_group_ro(cache, 1);
8555         }
8556
8557         init_global_block_rsv(info);
8558         ret = 0;
8559 error:
8560         btrfs_free_path(path);
8561         return ret;
8562 }
8563
8564 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8565                                        struct btrfs_root *root)
8566 {
8567         struct btrfs_block_group_cache *block_group, *tmp;
8568         struct btrfs_root *extent_root = root->fs_info->extent_root;
8569         struct btrfs_block_group_item item;
8570         struct btrfs_key key;
8571         int ret = 0;
8572
8573         list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8574                                  new_bg_list) {
8575                 list_del_init(&block_group->new_bg_list);
8576
8577                 if (ret)
8578                         continue;
8579
8580                 spin_lock(&block_group->lock);
8581                 memcpy(&item, &block_group->item, sizeof(item));
8582                 memcpy(&key, &block_group->key, sizeof(key));
8583                 spin_unlock(&block_group->lock);
8584
8585                 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8586                                         sizeof(item));
8587                 if (ret)
8588                         btrfs_abort_transaction(trans, extent_root, ret);
8589                 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8590                                                key.objectid, key.offset);
8591                 if (ret)
8592                         btrfs_abort_transaction(trans, extent_root, ret);
8593         }
8594 }
8595
8596 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8597                            struct btrfs_root *root, u64 bytes_used,
8598                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8599                            u64 size)
8600 {
8601         int ret;
8602         struct btrfs_root *extent_root;
8603         struct btrfs_block_group_cache *cache;
8604
8605         extent_root = root->fs_info->extent_root;
8606
8607         root->fs_info->last_trans_log_full_commit = trans->transid;
8608
8609         cache = btrfs_create_block_group_cache(root, chunk_offset, size);
8610         if (!cache)
8611                 return -ENOMEM;
8612
8613         btrfs_set_block_group_used(&cache->item, bytes_used);
8614         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8615         btrfs_set_block_group_flags(&cache->item, type);
8616
8617         cache->flags = type;
8618         cache->last_byte_to_unpin = (u64)-1;
8619         cache->cached = BTRFS_CACHE_FINISHED;
8620         ret = exclude_super_stripes(root, cache);
8621         if (ret) {
8622                 /*
8623                  * We may have excluded something, so call this just in
8624                  * case.
8625                  */
8626                 free_excluded_extents(root, cache);
8627                 btrfs_put_block_group(cache);
8628                 return ret;
8629         }
8630
8631         add_new_free_space(cache, root->fs_info, chunk_offset,
8632                            chunk_offset + size);
8633
8634         free_excluded_extents(root, cache);
8635
8636         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8637         if (ret) {
8638                 btrfs_remove_free_space_cache(cache);
8639                 btrfs_put_block_group(cache);
8640                 return ret;
8641         }
8642
8643         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8644                                 &cache->space_info);
8645         if (ret) {
8646                 btrfs_remove_free_space_cache(cache);
8647                 spin_lock(&root->fs_info->block_group_cache_lock);
8648                 rb_erase(&cache->cache_node,
8649                          &root->fs_info->block_group_cache_tree);
8650                 spin_unlock(&root->fs_info->block_group_cache_lock);
8651                 btrfs_put_block_group(cache);
8652                 return ret;
8653         }
8654         update_global_block_rsv(root->fs_info);
8655
8656         spin_lock(&cache->space_info->lock);
8657         cache->space_info->bytes_readonly += cache->bytes_super;
8658         spin_unlock(&cache->space_info->lock);
8659
8660         __link_block_group(cache->space_info, cache);
8661
8662         list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8663
8664         set_avail_alloc_bits(extent_root->fs_info, type);
8665
8666         return 0;
8667 }
8668
8669 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8670 {
8671         u64 extra_flags = chunk_to_extended(flags) &
8672                                 BTRFS_EXTENDED_PROFILE_MASK;
8673
8674         write_seqlock(&fs_info->profiles_lock);
8675         if (flags & BTRFS_BLOCK_GROUP_DATA)
8676                 fs_info->avail_data_alloc_bits &= ~extra_flags;
8677         if (flags & BTRFS_BLOCK_GROUP_METADATA)
8678                 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8679         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8680                 fs_info->avail_system_alloc_bits &= ~extra_flags;
8681         write_sequnlock(&fs_info->profiles_lock);
8682 }
8683
8684 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8685                              struct btrfs_root *root, u64 group_start)
8686 {
8687         struct btrfs_path *path;
8688         struct btrfs_block_group_cache *block_group;
8689         struct btrfs_free_cluster *cluster;
8690         struct btrfs_root *tree_root = root->fs_info->tree_root;
8691         struct btrfs_key key;
8692         struct inode *inode;
8693         int ret;
8694         int index;
8695         int factor;
8696
8697         root = root->fs_info->extent_root;
8698
8699         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8700         BUG_ON(!block_group);
8701         BUG_ON(!block_group->ro);
8702
8703         /*
8704          * Free the reserved super bytes from this block group before
8705          * remove it.
8706          */
8707         free_excluded_extents(root, block_group);
8708
8709         memcpy(&key, &block_group->key, sizeof(key));
8710         index = get_block_group_index(block_group);
8711         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8712                                   BTRFS_BLOCK_GROUP_RAID1 |
8713                                   BTRFS_BLOCK_GROUP_RAID10))
8714                 factor = 2;
8715         else
8716                 factor = 1;
8717
8718         /* make sure this block group isn't part of an allocation cluster */
8719         cluster = &root->fs_info->data_alloc_cluster;
8720         spin_lock(&cluster->refill_lock);
8721         btrfs_return_cluster_to_free_space(block_group, cluster);
8722         spin_unlock(&cluster->refill_lock);
8723
8724         /*
8725          * make sure this block group isn't part of a metadata
8726          * allocation cluster
8727          */
8728         cluster = &root->fs_info->meta_alloc_cluster;
8729         spin_lock(&cluster->refill_lock);
8730         btrfs_return_cluster_to_free_space(block_group, cluster);
8731         spin_unlock(&cluster->refill_lock);
8732
8733         path = btrfs_alloc_path();
8734         if (!path) {
8735                 ret = -ENOMEM;
8736                 goto out;
8737         }
8738
8739         inode = lookup_free_space_inode(tree_root, block_group, path);
8740         if (!IS_ERR(inode)) {
8741                 ret = btrfs_orphan_add(trans, inode);
8742                 if (ret) {
8743                         btrfs_add_delayed_iput(inode);
8744                         goto out;
8745                 }
8746                 clear_nlink(inode);
8747                 /* One for the block groups ref */
8748                 spin_lock(&block_group->lock);
8749                 if (block_group->iref) {
8750                         block_group->iref = 0;
8751                         block_group->inode = NULL;
8752                         spin_unlock(&block_group->lock);
8753                         iput(inode);
8754                 } else {
8755                         spin_unlock(&block_group->lock);
8756                 }
8757                 /* One for our lookup ref */
8758                 btrfs_add_delayed_iput(inode);
8759         }
8760
8761         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8762         key.offset = block_group->key.objectid;
8763         key.type = 0;
8764
8765         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8766         if (ret < 0)
8767                 goto out;
8768         if (ret > 0)
8769                 btrfs_release_path(path);
8770         if (ret == 0) {
8771                 ret = btrfs_del_item(trans, tree_root, path);
8772                 if (ret)
8773                         goto out;
8774                 btrfs_release_path(path);
8775         }
8776
8777         spin_lock(&root->fs_info->block_group_cache_lock);
8778         rb_erase(&block_group->cache_node,
8779                  &root->fs_info->block_group_cache_tree);
8780
8781         if (root->fs_info->first_logical_byte == block_group->key.objectid)
8782                 root->fs_info->first_logical_byte = (u64)-1;
8783         spin_unlock(&root->fs_info->block_group_cache_lock);
8784
8785         down_write(&block_group->space_info->groups_sem);
8786         /*
8787          * we must use list_del_init so people can check to see if they
8788          * are still on the list after taking the semaphore
8789          */
8790         list_del_init(&block_group->list);
8791         if (list_empty(&block_group->space_info->block_groups[index])) {
8792                 kobject_del(&block_group->space_info->block_group_kobjs[index]);
8793                 kobject_put(&block_group->space_info->block_group_kobjs[index]);
8794                 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8795         }
8796         up_write(&block_group->space_info->groups_sem);
8797
8798         if (block_group->cached == BTRFS_CACHE_STARTED)
8799                 wait_block_group_cache_done(block_group);
8800
8801         btrfs_remove_free_space_cache(block_group);
8802
8803         spin_lock(&block_group->space_info->lock);
8804         block_group->space_info->total_bytes -= block_group->key.offset;
8805         block_group->space_info->bytes_readonly -= block_group->key.offset;
8806         block_group->space_info->disk_total -= block_group->key.offset * factor;
8807         spin_unlock(&block_group->space_info->lock);
8808
8809         memcpy(&key, &block_group->key, sizeof(key));
8810
8811         btrfs_clear_space_info_full(root->fs_info);
8812
8813         btrfs_put_block_group(block_group);
8814         btrfs_put_block_group(block_group);
8815
8816         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8817         if (ret > 0)
8818                 ret = -EIO;
8819         if (ret < 0)
8820                 goto out;
8821
8822         ret = btrfs_del_item(trans, root, path);
8823 out:
8824         btrfs_free_path(path);
8825         return ret;
8826 }
8827
8828 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8829 {
8830         struct btrfs_space_info *space_info;
8831         struct btrfs_super_block *disk_super;
8832         u64 features;
8833         u64 flags;
8834         int mixed = 0;
8835         int ret;
8836
8837         disk_super = fs_info->super_copy;
8838         if (!btrfs_super_root(disk_super))
8839                 return 1;
8840
8841         features = btrfs_super_incompat_flags(disk_super);
8842         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8843                 mixed = 1;
8844
8845         flags = BTRFS_BLOCK_GROUP_SYSTEM;
8846         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8847         if (ret)
8848                 goto out;
8849
8850         if (mixed) {
8851                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8852                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8853         } else {
8854                 flags = BTRFS_BLOCK_GROUP_METADATA;
8855                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8856                 if (ret)
8857                         goto out;
8858
8859                 flags = BTRFS_BLOCK_GROUP_DATA;
8860                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8861         }
8862 out:
8863         return ret;
8864 }
8865
8866 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8867 {
8868         return unpin_extent_range(root, start, end);
8869 }
8870
8871 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8872                                u64 num_bytes, u64 *actual_bytes)
8873 {
8874         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8875 }
8876
8877 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8878 {
8879         struct btrfs_fs_info *fs_info = root->fs_info;
8880         struct btrfs_block_group_cache *cache = NULL;
8881         u64 group_trimmed;
8882         u64 start;
8883         u64 end;
8884         u64 trimmed = 0;
8885         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8886         int ret = 0;
8887
8888         /*
8889          * try to trim all FS space, our block group may start from non-zero.
8890          */
8891         if (range->len == total_bytes)
8892                 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8893         else
8894                 cache = btrfs_lookup_block_group(fs_info, range->start);
8895
8896         while (cache) {
8897                 if (cache->key.objectid >= (range->start + range->len)) {
8898                         btrfs_put_block_group(cache);
8899                         break;
8900                 }
8901
8902                 start = max(range->start, cache->key.objectid);
8903                 end = min(range->start + range->len,
8904                                 cache->key.objectid + cache->key.offset);
8905
8906                 if (end - start >= range->minlen) {
8907                         if (!block_group_cache_done(cache)) {
8908                                 ret = cache_block_group(cache, 0);
8909                                 if (ret) {
8910                                         btrfs_put_block_group(cache);
8911                                         break;
8912                                 }
8913                                 ret = wait_block_group_cache_done(cache);
8914                                 if (ret) {
8915                                         btrfs_put_block_group(cache);
8916                                         break;
8917                                 }
8918                         }
8919                         ret = btrfs_trim_block_group(cache,
8920                                                      &group_trimmed,
8921                                                      start,
8922                                                      end,
8923                                                      range->minlen);
8924
8925                         trimmed += group_trimmed;
8926                         if (ret) {
8927                                 btrfs_put_block_group(cache);
8928                                 break;
8929                         }
8930                 }
8931
8932                 cache = next_block_group(fs_info->tree_root, cache);
8933         }
8934
8935         range->len = trimmed;
8936         return ret;
8937 }