fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <linux/slab.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/percpu_counter.h>
  28 #include "hash.h"
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "print-tree.h"
  32 #include "transaction.h"
  33 #include "volumes.h"
  34 #include "raid56.h"
  35 #include "locking.h"
  36 #include "free-space-cache.h"
  37 #include "math.h"
  38
  39 #undef SCRAMBLE_DELAYED_REFS
  40
  41 /*
  42  * control flags for do_chunk_alloc's force field
  43  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  44  * if we really need one.
  45  *
  46  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  47  * if we have very few chunks already allocated.  This is
  48  * used as part of the clustering code to help make sure
  49  * we have a good pool of storage to cluster in, without
  50  * filling the FS with empty chunks
  51  *
  52  * CHUNK_ALLOC_FORCE means it must try to allocate one
  53  *
  54  */
  55 enum {
  56         CHUNK_ALLOC_NO_FORCE = 0,
  57         CHUNK_ALLOC_LIMITED = 1,
  58         CHUNK_ALLOC_FORCE = 2,
  59 };
  60
  61 /*
  62  * Control how reservations are dealt with.
  63  *
  64  * RESERVE_FREE - freeing a reservation.
  65  * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
  66  *   ENOSPC accounting
  67  * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
  68  *   bytes_may_use as the ENOSPC accounting is done elsewhere
  69  */
  70 enum {
  71         RESERVE_FREE = 0,
  72         RESERVE_ALLOC = 1,
  73         RESERVE_ALLOC_NO_ACCOUNT = 2,
  74 };
  75
  76 static int update_block_group(struct btrfs_root *root,
  77                               u64 bytenr, u64 num_bytes, int alloc);
  78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  79                                 struct btrfs_root *root,
  80                                 u64 bytenr, u64 num_bytes, u64 parent,
  81                                 u64 root_objectid, u64 owner_objectid,
  82                                 u64 owner_offset, int refs_to_drop,
  83                                 struct btrfs_delayed_extent_op *extra_op);
  84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  85                                     struct extent_buffer *leaf,
  86                                     struct btrfs_extent_item *ei);
  87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  88                                       struct btrfs_root *root,
  89                                       u64 parent, u64 root_objectid,
  90                                       u64 flags, u64 owner, u64 offset,
  91                                       struct btrfs_key *ins, int ref_mod);
  92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  93                                      struct btrfs_root *root,
  94                                      u64 parent, u64 root_objectid,
  95                                      u64 flags, struct btrfs_disk_key *key,
  96                                      int level, struct btrfs_key *ins);
  97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  98                           struct btrfs_root *extent_root, u64 flags,
  99                           int force);
 100 static int find_next_key(struct btrfs_path *path, int level,
 101                          struct btrfs_key *key);
 102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
 103                             int dump_block_groups);
 104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
 105                                        u64 num_bytes, int reserve);
 106 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
 107                                u64 num_bytes);
 108 int btrfs_pin_extent(struct btrfs_root *root,
 109                      u64 bytenr, u64 num_bytes, int reserved);
 110
 111 static noinline int
 112 block_group_cache_done(struct btrfs_block_group_cache *cache)
 113 {
 114         smp_mb();
 115         return cache->cached == BTRFS_CACHE_FINISHED ||
 116                 cache->cached == BTRFS_CACHE_ERROR;
 117 }
 118
 119 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 120 {
 121         return (cache->flags & bits) == bits;
 122 }
 123
 124 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 125 {
 126         atomic_inc(&cache->count);
 127 }
 128
 129 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 130 {
 131         if (atomic_dec_and_test(&cache->count)) {
 132                 WARN_ON(cache->pinned > 0);
 133                 WARN_ON(cache->reserved > 0);
 134                 kfree(cache->free_space_ctl);
 135                 kfree(cache);
 136         }
 137 }
 138
 139 /*
 140  * this adds the block group to the fs_info rb tree for the block group
 141  * cache
 142  */
 143 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 144                                 struct btrfs_block_group_cache *block_group)
 145 {
 146         struct rb_node **p;
 147         struct rb_node *parent = NULL;
 148         struct btrfs_block_group_cache *cache;
 149
 150         spin_lock(&info->block_group_cache_lock);
 151         p = &info->block_group_cache_tree.rb_node;
 152
 153         while (*p) {
 154                 parent = *p;
 155                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 156                                  cache_node);
 157                 if (block_group->key.objectid < cache->key.objectid) {
 158                         p = &(*p)->rb_left;
 159                 } else if (block_group->key.objectid > cache->key.objectid) {
 160                         p = &(*p)->rb_right;
 161                 } else {
 162                         spin_unlock(&info->block_group_cache_lock);
 163                         return -EEXIST;
 164                 }
 165         }
 166
 167         rb_link_node(&block_group->cache_node, parent, p);
 168         rb_insert_color(&block_group->cache_node,
 169                         &info->block_group_cache_tree);
 170
 171         if (info->first_logical_byte > block_group->key.objectid)
 172                 info->first_logical_byte = block_group->key.objectid;
 173
 174         spin_unlock(&info->block_group_cache_lock);
 175
 176         return 0;
 177 }
 178
 179 /*
 180  * This will return the block group at or after bytenr if contains is 0, else
 181  * it will return the block group that contains the bytenr
 182  */
 183 static struct btrfs_block_group_cache *
 184 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 185                               int contains)
 186 {
 187         struct btrfs_block_group_cache *cache, *ret = NULL;
 188         struct rb_node *n;
 189         u64 end, start;
 190
 191         spin_lock(&info->block_group_cache_lock);
 192         n = info->block_group_cache_tree.rb_node;
 193
 194         while (n) {
 195                 cache = rb_entry(n, struct btrfs_block_group_cache,
 196                                  cache_node);
 197                 end = cache->key.objectid + cache->key.offset - 1;
 198                 start = cache->key.objectid;
 199
 200                 if (bytenr < start) {
 201                         if (!contains && (!ret || start < ret->key.objectid))
 202                                 ret = cache;
 203                         n = n->rb_left;
 204                 } else if (bytenr > start) {
 205                         if (contains && bytenr <= end) {
 206                                 ret = cache;
 207                                 break;
 208                         }
 209                         n = n->rb_right;
 210                 } else {
 211                         ret = cache;
 212                         break;
 213                 }
 214         }
 215         if (ret) {
 216                 btrfs_get_block_group(ret);
 217                 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
 218                         info->first_logical_byte = ret->key.objectid;
 219         }
 220         spin_unlock(&info->block_group_cache_lock);
 221
 222         return ret;
 223 }
 224
 225 static int add_excluded_extent(struct btrfs_root *root,
 226                                u64 start, u64 num_bytes)
 227 {
 228         u64 end = start + num_bytes - 1;
 229         set_extent_bits(&root->fs_info->freed_extents[0],
 230                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 231         set_extent_bits(&root->fs_info->freed_extents[1],
 232                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 233         return 0;
 234 }
 235
 236 static void free_excluded_extents(struct btrfs_root *root,
 237                                   struct btrfs_block_group_cache *cache)
 238 {
 239         u64 start, end;
 240
 241         start = cache->key.objectid;
 242         end = start + cache->key.offset - 1;
 243
 244         clear_extent_bits(&root->fs_info->freed_extents[0],
 245                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 246         clear_extent_bits(&root->fs_info->freed_extents[1],
 247                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 248 }
 249
 250 static int exclude_super_stripes(struct btrfs_root *root,
 251                                  struct btrfs_block_group_cache *cache)
 252 {
 253         u64 bytenr;
 254         u64 *logical;
 255         int stripe_len;
 256         int i, nr, ret;
 257
 258         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 259                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 260                 cache->bytes_super += stripe_len;
 261                 ret = add_excluded_extent(root, cache->key.objectid,
 262                                           stripe_len);
 263                 if (ret)
 264                         return ret;
 265         }
 266
 267         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 268                 bytenr = btrfs_sb_offset(i);
 269                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 270                                        cache->key.objectid, bytenr,
 271                                        0, &logical, &nr, &stripe_len);
 272                 if (ret)
 273                         return ret;
 274
 275                 while (nr--) {
 276                         u64 start, len;
 277
 278                         if (logical[nr] > cache->key.objectid +
 279                             cache->key.offset)
 280                                 continue;
 281
 282                         if (logical[nr] + stripe_len <= cache->key.objectid)
 283                                 continue;
 284
 285                         start = logical[nr];
 286                         if (start < cache->key.objectid) {
 287                                 start = cache->key.objectid;
 288                                 len = (logical[nr] + stripe_len) - start;
 289                         } else {
 290                                 len = min_t(u64, stripe_len,
 291                                             cache->key.objectid +
 292                                             cache->key.offset - start);
 293                         }
 294
 295                         cache->bytes_super += len;
 296                         ret = add_excluded_extent(root, start, len);
 297                         if (ret) {
 298                                 kfree(logical);
 299                                 return ret;
 300                         }
 301                 }
 302
 303                 kfree(logical);
 304         }
 305         return 0;
 306 }
 307
 308 static struct btrfs_caching_control *
 309 get_caching_control(struct btrfs_block_group_cache *cache)
 310 {
 311         struct btrfs_caching_control *ctl;
 312
 313         spin_lock(&cache->lock);
 314         if (cache->cached != BTRFS_CACHE_STARTED) {
 315                 spin_unlock(&cache->lock);
 316                 return NULL;
 317         }
 318
 319         /* We're loading it the fast way, so we don't have a caching_ctl. */
 320         if (!cache->caching_ctl) {
 321                 spin_unlock(&cache->lock);
 322                 return NULL;
 323         }
 324
 325         ctl = cache->caching_ctl;
 326         atomic_inc(&ctl->count);
 327         spin_unlock(&cache->lock);
 328         return ctl;
 329 }
 330
 331 static void put_caching_control(struct btrfs_caching_control *ctl)
 332 {
 333         if (atomic_dec_and_test(&ctl->count))
 334                 kfree(ctl);
 335 }
 336
 337 /*
 338  * this is only called by cache_block_group, since we could have freed extents
 339  * we need to check the pinned_extents for any extents that can't be used yet
 340  * since their free space will be released as soon as the transaction commits.
 341  */
 342 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 343                               struct btrfs_fs_info *info, u64 start, u64 end)
 344 {
 345         u64 extent_start, extent_end, size, total_added = 0;
 346         int ret;
 347
 348         while (start < end) {
 349                 ret = find_first_extent_bit(info->pinned_extents, start,
 350                                             &extent_start, &extent_end,
 351                                             EXTENT_DIRTY | EXTENT_UPTODATE,
 352                                             NULL);
 353                 if (ret)
 354                         break;
 355
 356                 if (extent_start <= start) {
 357                         start = extent_end + 1;
 358                 } else if (extent_start > start && extent_start < end) {
 359                         size = extent_start - start;
 360                         total_added += size;
 361                         ret = btrfs_add_free_space(block_group, start,
 362                                                    size);
 363                         BUG_ON(ret); /* -ENOMEM or logic error */
 364                         start = extent_end + 1;
 365                 } else {
 366                         break;
 367                 }
 368         }
 369
 370         if (start < end) {
 371                 size = end - start;
 372                 total_added += size;
 373                 ret = btrfs_add_free_space(block_group, start, size);
 374                 BUG_ON(ret); /* -ENOMEM or logic error */
 375         }
 376
 377         return total_added;
 378 }
 379
 380 static noinline void caching_thread(struct btrfs_work *work)
 381 {
 382         struct btrfs_block_group_cache *block_group;
 383         struct btrfs_fs_info *fs_info;
 384         struct btrfs_caching_control *caching_ctl;
 385         struct btrfs_root *extent_root;
 386         struct btrfs_path *path;
 387         struct extent_buffer *leaf;
 388         struct btrfs_key key;
 389         u64 total_found = 0;
 390         u64 last = 0;
 391         u32 nritems;
 392         int ret = -ENOMEM;
 393
 394         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 395         block_group = caching_ctl->block_group;
 396         fs_info = block_group->fs_info;
 397         extent_root = fs_info->extent_root;
 398
 399         path = btrfs_alloc_path();
 400         if (!path)
 401                 goto out;
 402
 403         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 404
 405         /*
 406          * We don't want to deadlock with somebody trying to allocate a new
 407          * extent for the extent root while also trying to search the extent
 408          * root to add free space.  So we skip locking and search the commit
 409          * root, since its read-only
 410          */
 411         path->skip_locking = 1;
 412         path->search_commit_root = 1;
 413         path->reada = 1;
 414
 415         key.objectid = last;
 416         key.offset = 0;
 417         key.type = BTRFS_EXTENT_ITEM_KEY;
 418 again:
 419         mutex_lock(&caching_ctl->mutex);
 420         /* need to make sure the commit_root doesn't disappear */
 421         down_read(&fs_info->extent_commit_sem);
 422
 423 next:
 424         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 425         if (ret < 0)
 426                 goto err;
 427
 428         leaf = path->nodes[0];
 429         nritems = btrfs_header_nritems(leaf);
 430
 431         while (1) {
 432                 if (btrfs_fs_closing(fs_info) > 1) {
 433                         last = (u64)-1;
 434                         break;
 435                 }
 436
 437                 if (path->slots[0] < nritems) {
 438                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 439                 } else {
 440                         ret = find_next_key(path, 0, &key);
 441                         if (ret)
 442                                 break;
 443
 444                         if (need_resched()) {
 445                                 caching_ctl->progress = last;
 446                                 btrfs_release_path(path);
 447                                 up_read(&fs_info->extent_commit_sem);
 448                                 mutex_unlock(&caching_ctl->mutex);
 449                                 cond_resched();
 450                                 goto again;
 451                         }
 452
 453                         ret = btrfs_next_leaf(extent_root, path);
 454                         if (ret < 0)
 455                                 goto err;
 456                         if (ret)
 457                                 break;
 458                         leaf = path->nodes[0];
 459                         nritems = btrfs_header_nritems(leaf);
 460                         continue;
 461                 }
 462
 463                 if (key.objectid < last) {
 464                         key.objectid = last;
 465                         key.offset = 0;
 466                         key.type = BTRFS_EXTENT_ITEM_KEY;
 467
 468                         caching_ctl->progress = last;
 469                         btrfs_release_path(path);
 470                         goto next;
 471                 }
 472
 473                 if (key.objectid < block_group->key.objectid) {
 474                         path->slots[0]++;
 475                         continue;
 476                 }
 477
 478                 if (key.objectid >= block_group->key.objectid +
 479                     block_group->key.offset)
 480                         break;
 481
 482                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
 483                     key.type == BTRFS_METADATA_ITEM_KEY) {
 484                         total_found += add_new_free_space(block_group,
 485                                                           fs_info, last,
 486                                                           key.objectid);
 487                         if (key.type == BTRFS_METADATA_ITEM_KEY)
 488                                 last = key.objectid +
 489                                         fs_info->tree_root->leafsize;
 490                         else
 491                                 last = key.objectid + key.offset;
 492
 493                         if (total_found > (1024 * 1024 * 2)) {
 494                                 total_found = 0;
 495                                 wake_up(&caching_ctl->wait);
 496                         }
 497                 }
 498                 path->slots[0]++;
 499         }
 500         ret = 0;
 501
 502         total_found += add_new_free_space(block_group, fs_info, last,
 503                                           block_group->key.objectid +
 504                                           block_group->key.offset);
 505         caching_ctl->progress = (u64)-1;
 506
 507         spin_lock(&block_group->lock);
 508         block_group->caching_ctl = NULL;
 509         block_group->cached = BTRFS_CACHE_FINISHED;
 510         spin_unlock(&block_group->lock);
 511
 512 err:
 513         btrfs_free_path(path);
 514         up_read(&fs_info->extent_commit_sem);
 515
 516         free_excluded_extents(extent_root, block_group);
 517
 518         mutex_unlock(&caching_ctl->mutex);
 519 out:
 520         if (ret) {
 521                 spin_lock(&block_group->lock);
 522                 block_group->caching_ctl = NULL;
 523                 block_group->cached = BTRFS_CACHE_ERROR;
 524                 spin_unlock(&block_group->lock);
 525         }
 526         wake_up(&caching_ctl->wait);
 527
 528         put_caching_control(caching_ctl);
 529         btrfs_put_block_group(block_group);
 530 }
 531
 532 static int cache_block_group(struct btrfs_block_group_cache *cache,
 533                              int load_cache_only)
 534 {
 535         DEFINE_WAIT(wait);
 536         struct btrfs_fs_info *fs_info = cache->fs_info;
 537         struct btrfs_caching_control *caching_ctl;
 538         int ret = 0;
 539
 540         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 541         if (!caching_ctl)
 542                 return -ENOMEM;
 543
 544         INIT_LIST_HEAD(&caching_ctl->list);
 545         mutex_init(&caching_ctl->mutex);
 546         init_waitqueue_head(&caching_ctl->wait);
 547         caching_ctl->block_group = cache;
 548         caching_ctl->progress = cache->key.objectid;
 549         atomic_set(&caching_ctl->count, 1);
 550         caching_ctl->work.func = caching_thread;
 551
 552         spin_lock(&cache->lock);
 553         /*
 554          * This should be a rare occasion, but this could happen I think in the
 555          * case where one thread starts to load the space cache info, and then
 556          * some other thread starts a transaction commit which tries to do an
 557          * allocation while the other thread is still loading the space cache
 558          * info.  The previous loop should have kept us from choosing this block
 559          * group, but if we've moved to the state where we will wait on caching
 560          * block groups we need to first check if we're doing a fast load here,
 561          * so we can wait for it to finish, otherwise we could end up allocating
 562          * from a block group who's cache gets evicted for one reason or
 563          * another.
 564          */
 565         while (cache->cached == BTRFS_CACHE_FAST) {
 566                 struct btrfs_caching_control *ctl;
 567
 568                 ctl = cache->caching_ctl;
 569                 atomic_inc(&ctl->count);
 570                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
 571                 spin_unlock(&cache->lock);
 572
 573                 schedule();
 574
 575                 finish_wait(&ctl->wait, &wait);
 576                 put_caching_control(ctl);
 577                 spin_lock(&cache->lock);
 578         }
 579
 580         if (cache->cached != BTRFS_CACHE_NO) {
 581                 spin_unlock(&cache->lock);
 582                 kfree(caching_ctl);
 583                 return 0;
 584         }
 585         WARN_ON(cache->caching_ctl);
 586         cache->caching_ctl = caching_ctl;
 587         cache->cached = BTRFS_CACHE_FAST;
 588         spin_unlock(&cache->lock);
 589
 590         if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
 591                 ret = load_free_space_cache(fs_info, cache);
 592
 593                 spin_lock(&cache->lock);
 594                 if (ret == 1) {
 595                         cache->caching_ctl = NULL;
 596                         cache->cached = BTRFS_CACHE_FINISHED;
 597                         cache->last_byte_to_unpin = (u64)-1;
 598                 } else {
 599                         if (load_cache_only) {
 600                                 cache->caching_ctl = NULL;
 601                                 cache->cached = BTRFS_CACHE_NO;
 602                         } else {
 603                                 cache->cached = BTRFS_CACHE_STARTED;
 604                         }
 605                 }
 606                 spin_unlock(&cache->lock);
 607                 wake_up(&caching_ctl->wait);
 608                 if (ret == 1) {
 609                         put_caching_control(caching_ctl);
 610                         free_excluded_extents(fs_info->extent_root, cache);
 611                         return 0;
 612                 }
 613         } else {
 614                 /*
 615                  * We are not going to do the fast caching, set cached to the
 616                  * appropriate value and wakeup any waiters.
 617                  */
 618                 spin_lock(&cache->lock);
 619                 if (load_cache_only) {
 620                         cache->caching_ctl = NULL;
 621                         cache->cached = BTRFS_CACHE_NO;
 622                 } else {
 623                         cache->cached = BTRFS_CACHE_STARTED;
 624                 }
 625                 spin_unlock(&cache->lock);
 626                 wake_up(&caching_ctl->wait);
 627         }
 628
 629         if (load_cache_only) {
 630                 put_caching_control(caching_ctl);
 631                 return 0;
 632         }
 633
 634         down_write(&fs_info->extent_commit_sem);
 635         atomic_inc(&caching_ctl->count);
 636         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 637         up_write(&fs_info->extent_commit_sem);
 638
 639         btrfs_get_block_group(cache);
 640
 641         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 642
 643         return ret;
 644 }
 645
 646 /*
 647  * return the block group that starts at or after bytenr
 648  */
 649 static struct btrfs_block_group_cache *
 650 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 651 {
 652         struct btrfs_block_group_cache *cache;
 653
 654         cache = block_group_cache_tree_search(info, bytenr, 0);
 655
 656         return cache;
 657 }
 658
 659 /*
 660  * return the block group that contains the given bytenr
 661  */
 662 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 663                                                  struct btrfs_fs_info *info,
 664                                                  u64 bytenr)
 665 {
 666         struct btrfs_block_group_cache *cache;
 667
 668         cache = block_group_cache_tree_search(info, bytenr, 1);
 669
 670         return cache;
 671 }
 672
 673 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 674                                                   u64 flags)
 675 {
 676         struct list_head *head = &info->space_info;
 677         struct btrfs_space_info *found;
 678
 679         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 680
 681         rcu_read_lock();
 682         list_for_each_entry_rcu(found, head, list) {
 683                 if (found->flags & flags) {
 684                         rcu_read_unlock();
 685                         return found;
 686                 }
 687         }
 688         rcu_read_unlock();
 689         return NULL;
 690 }
 691
 692 /*
 693  * after adding space to the filesystem, we need to clear the full flags
 694  * on all the space infos.
 695  */
 696 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 697 {
 698         struct list_head *head = &info->space_info;
 699         struct btrfs_space_info *found;
 700
 701         rcu_read_lock();
 702         list_for_each_entry_rcu(found, head, list)
 703                 found->full = 0;
 704         rcu_read_unlock();
 705 }
 706
 707 /* simple helper to search for an existing extent at a given offset */
 708 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 709 {
 710         int ret;
 711         struct btrfs_key key;
 712         struct btrfs_path *path;
 713
 714         path = btrfs_alloc_path();
 715         if (!path)
 716                 return -ENOMEM;
 717
 718         key.objectid = start;
 719         key.offset = len;
 720         key.type = BTRFS_EXTENT_ITEM_KEY;
 721         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 722                                 0, 0);
 723         if (ret > 0) {
 724                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 725                 if (key.objectid == start &&
 726                     key.type == BTRFS_METADATA_ITEM_KEY)
 727                         ret = 0;
 728         }
 729         btrfs_free_path(path);
 730         return ret;
 731 }
 732
 733 /*
 734  * helper function to lookup reference count and flags of a tree block.
 735  *
 736  * the head node for delayed ref is used to store the sum of all the
 737  * reference count modifications queued up in the rbtree. the head
 738  * node may also store the extent flags to set. This way you can check
 739  * to see what the reference count and extent flags would be if all of
 740  * the delayed refs are not processed.
 741  */
 742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 743                              struct btrfs_root *root, u64 bytenr,
 744                              u64 offset, int metadata, u64 *refs, u64 *flags)
 745 {
 746         struct btrfs_delayed_ref_head *head;
 747         struct btrfs_delayed_ref_root *delayed_refs;
 748         struct btrfs_path *path;
 749         struct btrfs_extent_item *ei;
 750         struct extent_buffer *leaf;
 751         struct btrfs_key key;
 752         u32 item_size;
 753         u64 num_refs;
 754         u64 extent_flags;
 755         int ret;
 756
 757         /*
 758          * If we don't have skinny metadata, don't bother doing anything
 759          * different
 760          */
 761         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
 762                 offset = root->leafsize;
 763                 metadata = 0;
 764         }
 765
 766         path = btrfs_alloc_path();
 767         if (!path)
 768                 return -ENOMEM;
 769
 770         if (metadata) {
 771                 key.objectid = bytenr;
 772                 key.type = BTRFS_METADATA_ITEM_KEY;
 773                 key.offset = offset;
 774         } else {
 775                 key.objectid = bytenr;
 776                 key.type = BTRFS_EXTENT_ITEM_KEY;
 777                 key.offset = offset;
 778         }
 779
 780         if (!trans) {
 781                 path->skip_locking = 1;
 782                 path->search_commit_root = 1;
 783         }
 784 again:
 785         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 786                                 &key, path, 0, 0);
 787         if (ret < 0)
 788                 goto out_free;
 789
 790         if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
 791                 metadata = 0;
 792                 if (path->slots[0]) {
 793                         path->slots[0]--;
 794                         btrfs_item_key_to_cpu(path->nodes[0], &key,
 795                                               path->slots[0]);
 796                         if (key.objectid == bytenr &&
 797                             key.type == BTRFS_EXTENT_ITEM_KEY &&
 798                             key.offset == root->leafsize)
 799                                 ret = 0;
 800                 }
 801                 if (ret) {
 802                         key.objectid = bytenr;
 803                         key.type = BTRFS_EXTENT_ITEM_KEY;
 804                         key.offset = root->leafsize;
 805                         btrfs_release_path(path);
 806                         goto again;
 807                 }
 808         }
 809
 810         if (ret == 0) {
 811                 leaf = path->nodes[0];
 812                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 813                 if (item_size >= sizeof(*ei)) {
 814                         ei = btrfs_item_ptr(leaf, path->slots[0],
 815                                             struct btrfs_extent_item);
 816                         num_refs = btrfs_extent_refs(leaf, ei);
 817                         extent_flags = btrfs_extent_flags(leaf, ei);
 818                 } else {
 819 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 820                         struct btrfs_extent_item_v0 *ei0;
 821                         BUG_ON(item_size != sizeof(*ei0));
 822                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 823                                              struct btrfs_extent_item_v0);
 824                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 825                         /* FIXME: this isn't correct for data */
 826                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 827 #else
 828                         BUG();
 829 #endif
 830                 }
 831                 BUG_ON(num_refs == 0);
 832         } else {
 833                 num_refs = 0;
 834                 extent_flags = 0;
 835                 ret = 0;
 836         }
 837
 838         if (!trans)
 839                 goto out;
 840
 841         delayed_refs = &trans->transaction->delayed_refs;
 842         spin_lock(&delayed_refs->lock);
 843         head = btrfs_find_delayed_ref_head(trans, bytenr);
 844         if (head) {
 845                 if (!mutex_trylock(&head->mutex)) {
 846                         atomic_inc(&head->node.refs);
 847                         spin_unlock(&delayed_refs->lock);
 848
 849                         btrfs_release_path(path);
 850
 851                         /*
 852                          * Mutex was contended, block until it's released and try
 853                          * again
 854                          */
 855                         mutex_lock(&head->mutex);
 856                         mutex_unlock(&head->mutex);
 857                         btrfs_put_delayed_ref(&head->node);
 858                         goto again;
 859                 }
 860                 if (head->extent_op && head->extent_op->update_flags)
 861                         extent_flags |= head->extent_op->flags_to_set;
 862                 else
 863                         BUG_ON(num_refs == 0);
 864
 865                 num_refs += head->node.ref_mod;
 866                 mutex_unlock(&head->mutex);
 867         }
 868         spin_unlock(&delayed_refs->lock);
 869 out:
 870         WARN_ON(num_refs == 0);
 871         if (refs)
 872                 *refs = num_refs;
 873         if (flags)
 874                 *flags = extent_flags;
 875 out_free:
 876         btrfs_free_path(path);
 877         return ret;
 878 }
 879
 880 /*
 881  * Back reference rules.  Back refs have three main goals:
 882  *
 883  * 1) differentiate between all holders of references to an extent so that
 884  *    when a reference is dropped we can make sure it was a valid reference
 885  *    before freeing the extent.
 886  *
 887  * 2) Provide enough information to quickly find the holders of an extent
 888  *    if we notice a given block is corrupted or bad.
 889  *
 890  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 891  *    maintenance.  This is actually the same as #2, but with a slightly
 892  *    different use case.
 893  *
 894  * There are two kinds of back refs. The implicit back refs is optimized
 895  * for pointers in non-shared tree blocks. For a given pointer in a block,
 896  * back refs of this kind provide information about the block's owner tree
 897  * and the pointer's key. These information allow us to find the block by
 898  * b-tree searching. The full back refs is for pointers in tree blocks not
 899  * referenced by their owner trees. The location of tree block is recorded
 900  * in the back refs. Actually the full back refs is generic, and can be
 901  * used in all cases the implicit back refs is used. The major shortcoming
 902  * of the full back refs is its overhead. Every time a tree block gets
 903  * COWed, we have to update back refs entry for all pointers in it.
 904  *
 905  * For a newly allocated tree block, we use implicit back refs for
 906  * pointers in it. This means most tree related operations only involve
 907  * implicit back refs. For a tree block created in old transaction, the
 908  * only way to drop a reference to it is COW it. So we can detect the
 909  * event that tree block loses its owner tree's reference and do the
 910  * back refs conversion.
 911  *
 912  * When a tree block is COW'd through a tree, there are four cases:
 913  *
 914  * The reference count of the block is one and the tree is the block's
 915  * owner tree. Nothing to do in this case.
 916  *
 917  * The reference count of the block is one and the tree is not the
 918  * block's owner tree. In this case, full back refs is used for pointers
 919  * in the block. Remove these full back refs, add implicit back refs for
 920  * every pointers in the new block.
 921  *
 922  * The reference count of the block is greater than one and the tree is
 923  * the block's owner tree. In this case, implicit back refs is used for
 924  * pointers in the block. Add full back refs for every pointers in the
 925  * block, increase lower level extents' reference counts. The original
 926  * implicit back refs are entailed to the new block.
 927  *
 928  * The reference count of the block is greater than one and the tree is
 929  * not the block's owner tree. Add implicit back refs for every pointer in
 930  * the new block, increase lower level extents' reference count.
 931  *
 932  * Back Reference Key composing:
 933  *
 934  * The key objectid corresponds to the first byte in the extent,
 935  * The key type is used to differentiate between types of back refs.
 936  * There are different meanings of the key offset for different types
 937  * of back refs.
 938  *
 939  * File extents can be referenced by:
 940  *
 941  * - multiple snapshots, subvolumes, or different generations in one subvol
 942  * - different files inside a single subvolume
 943  * - different offsets inside a file (bookend extents in file.c)
 944  *
 945  * The extent ref structure for the implicit back refs has fields for:
 946  *
 947  * - Objectid of the subvolume root
 948  * - objectid of the file holding the reference
 949  * - original offset in the file
 950  * - how many bookend extents
 951  *
 952  * The key offset for the implicit back refs is hash of the first
 953  * three fields.
 954  *
 955  * The extent ref structure for the full back refs has field for:
 956  *
 957  * - number of pointers in the tree leaf
 958  *
 959  * The key offset for the implicit back refs is the first byte of
 960  * the tree leaf
 961  *
 962  * When a file extent is allocated, The implicit back refs is used.
 963  * the fields are filled in:
 964  *
 965  *     (root_key.objectid, inode objectid, offset in file, 1)
 966  *
 967  * When a file extent is removed file truncation, we find the
 968  * corresponding implicit back refs and check the following fields:
 969  *
 970  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 971  *
 972  * Btree extents can be referenced by:
 973  *
 974  * - Different subvolumes
 975  *
 976  * Both the implicit back refs and the full back refs for tree blocks
 977  * only consist of key. The key offset for the implicit back refs is
 978  * objectid of block's owner tree. The key offset for the full back refs
 979  * is the first byte of parent block.
 980  *
 981  * When implicit back refs is used, information about the lowest key and
 982  * level of the tree block are required. These information are stored in
 983  * tree block info structure.
 984  */
 985
 986 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 987 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 988                                   struct btrfs_root *root,
 989                                   struct btrfs_path *path,
 990                                   u64 owner, u32 extra_size)
 991 {
 992         struct btrfs_extent_item *item;
 993         struct btrfs_extent_item_v0 *ei0;
 994         struct btrfs_extent_ref_v0 *ref0;
 995         struct btrfs_tree_block_info *bi;
 996         struct extent_buffer *leaf;
 997         struct btrfs_key key;
 998         struct btrfs_key found_key;
 999         u32 new_size = sizeof(*item);
1000         u64 refs;
1001         int ret;
1002
1003         leaf = path->nodes[0];
1004         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1005
1006         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1007         ei0 = btrfs_item_ptr(leaf, path->slots[0],
1008                              struct btrfs_extent_item_v0);
1009         refs = btrfs_extent_refs_v0(leaf, ei0);
1010
1011         if (owner == (u64)-1) {
1012                 while (1) {
1013                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1014                                 ret = btrfs_next_leaf(root, path);
1015                                 if (ret < 0)
1016                                         return ret;
1017                                 BUG_ON(ret > 0); /* Corruption */
1018                                 leaf = path->nodes[0];
1019                         }
1020                         btrfs_item_key_to_cpu(leaf, &found_key,
1021                                               path->slots[0]);
1022                         BUG_ON(key.objectid != found_key.objectid);
1023                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1024                                 path->slots[0]++;
1025                                 continue;
1026                         }
1027                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1028                                               struct btrfs_extent_ref_v0);
1029                         owner = btrfs_ref_objectid_v0(leaf, ref0);
1030                         break;
1031                 }
1032         }
1033         btrfs_release_path(path);
1034
1035         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1036                 new_size += sizeof(*bi);
1037
1038         new_size -= sizeof(*ei0);
1039         ret = btrfs_search_slot(trans, root, &key, path,
1040                                 new_size + extra_size, 1);
1041         if (ret < 0)
1042                 return ret;
1043         BUG_ON(ret); /* Corruption */
1044
1045         btrfs_extend_item(root, path, new_size);
1046
1047         leaf = path->nodes[0];
1048         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1049         btrfs_set_extent_refs(leaf, item, refs);
1050         /* FIXME: get real generation */
1051         btrfs_set_extent_generation(leaf, item, 0);
1052         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1053                 btrfs_set_extent_flags(leaf, item,
1054                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1055                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1056                 bi = (struct btrfs_tree_block_info *)(item + 1);
1057                 /* FIXME: get first key of the block */
1058                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1059                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1060         } else {
1061                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1062         }
1063         btrfs_mark_buffer_dirty(leaf);
1064         return 0;
1065 }
1066 #endif
1067
1068 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1069 {
1070         u32 high_crc = ~(u32)0;
1071         u32 low_crc = ~(u32)0;
1072         __le64 lenum;
1073
1074         lenum = cpu_to_le64(root_objectid);
1075         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1076         lenum = cpu_to_le64(owner);
1077         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1078         lenum = cpu_to_le64(offset);
1079         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1080
1081         return ((u64)high_crc << 31) ^ (u64)low_crc;
1082 }
1083
1084 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1085                                      struct btrfs_extent_data_ref *ref)
1086 {
1087         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1088                                     btrfs_extent_data_ref_objectid(leaf, ref),
1089                                     btrfs_extent_data_ref_offset(leaf, ref));
1090 }
1091
1092 static int match_extent_data_ref(struct extent_buffer *leaf,
1093                                  struct btrfs_extent_data_ref *ref,
1094                                  u64 root_objectid, u64 owner, u64 offset)
1095 {
1096         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1097             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1098             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1099                 return 0;
1100         return 1;
1101 }
1102
1103 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1104                                            struct btrfs_root *root,
1105                                            struct btrfs_path *path,
1106                                            u64 bytenr, u64 parent,
1107                                            u64 root_objectid,
1108                                            u64 owner, u64 offset)
1109 {
1110         struct btrfs_key key;
1111         struct btrfs_extent_data_ref *ref;
1112         struct extent_buffer *leaf;
1113         u32 nritems;
1114         int ret;
1115         int recow;
1116         int err = -ENOENT;
1117
1118         key.objectid = bytenr;
1119         if (parent) {
1120                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1121                 key.offset = parent;
1122         } else {
1123                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1124                 key.offset = hash_extent_data_ref(root_objectid,
1125                                                   owner, offset);
1126         }
1127 again:
1128         recow = 0;
1129         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1130         if (ret < 0) {
1131                 err = ret;
1132                 goto fail;
1133         }
1134
1135         if (parent) {
1136                 if (!ret)
1137                         return 0;
1138 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1139                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1140                 btrfs_release_path(path);
1141                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1142                 if (ret < 0) {
1143                         err = ret;
1144                         goto fail;
1145                 }
1146                 if (!ret)
1147                         return 0;
1148 #endif
1149                 goto fail;
1150         }
1151
1152         leaf = path->nodes[0];
1153         nritems = btrfs_header_nritems(leaf);
1154         while (1) {
1155                 if (path->slots[0] >= nritems) {
1156                         ret = btrfs_next_leaf(root, path);
1157                         if (ret < 0)
1158                                 err = ret;
1159                         if (ret)
1160                                 goto fail;
1161
1162                         leaf = path->nodes[0];
1163                         nritems = btrfs_header_nritems(leaf);
1164                         recow = 1;
1165                 }
1166
1167                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1168                 if (key.objectid != bytenr ||
1169                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1170                         goto fail;
1171
1172                 ref = btrfs_item_ptr(leaf, path->slots[0],
1173                                      struct btrfs_extent_data_ref);
1174
1175                 if (match_extent_data_ref(leaf, ref, root_objectid,
1176                                           owner, offset)) {
1177                         if (recow) {
1178                                 btrfs_release_path(path);
1179                                 goto again;
1180                         }
1181                         err = 0;
1182                         break;
1183                 }
1184                 path->slots[0]++;
1185         }
1186 fail:
1187         return err;
1188 }
1189
1190 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1191                                            struct btrfs_root *root,
1192                                            struct btrfs_path *path,
1193                                            u64 bytenr, u64 parent,
1194                                            u64 root_objectid, u64 owner,
1195                                            u64 offset, int refs_to_add)
1196 {
1197         struct btrfs_key key;
1198         struct extent_buffer *leaf;
1199         u32 size;
1200         u32 num_refs;
1201         int ret;
1202
1203         key.objectid = bytenr;
1204         if (parent) {
1205                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1206                 key.offset = parent;
1207                 size = sizeof(struct btrfs_shared_data_ref);
1208         } else {
1209                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1210                 key.offset = hash_extent_data_ref(root_objectid,
1211                                                   owner, offset);
1212                 size = sizeof(struct btrfs_extent_data_ref);
1213         }
1214
1215         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1216         if (ret && ret != -EEXIST)
1217                 goto fail;
1218
1219         leaf = path->nodes[0];
1220         if (parent) {
1221                 struct btrfs_shared_data_ref *ref;
1222                 ref = btrfs_item_ptr(leaf, path->slots[0],
1223                                      struct btrfs_shared_data_ref);
1224                 if (ret == 0) {
1225                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1226                 } else {
1227                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1228                         num_refs += refs_to_add;
1229                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1230                 }
1231         } else {
1232                 struct btrfs_extent_data_ref *ref;
1233                 while (ret == -EEXIST) {
1234                         ref = btrfs_item_ptr(leaf, path->slots[0],
1235                                              struct btrfs_extent_data_ref);
1236                         if (match_extent_data_ref(leaf, ref, root_objectid,
1237                                                   owner, offset))
1238                                 break;
1239                         btrfs_release_path(path);
1240                         key.offset++;
1241                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1242                                                       size);
1243                         if (ret && ret != -EEXIST)
1244                                 goto fail;
1245
1246                         leaf = path->nodes[0];
1247                 }
1248                 ref = btrfs_item_ptr(leaf, path->slots[0],
1249                                      struct btrfs_extent_data_ref);
1250                 if (ret == 0) {
1251                         btrfs_set_extent_data_ref_root(leaf, ref,
1252                                                        root_objectid);
1253                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1254                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1255                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1256                 } else {
1257                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1258                         num_refs += refs_to_add;
1259                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1260                 }
1261         }
1262         btrfs_mark_buffer_dirty(leaf);
1263         ret = 0;
1264 fail:
1265         btrfs_release_path(path);
1266         return ret;
1267 }
1268
1269 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1270                                            struct btrfs_root *root,
1271                                            struct btrfs_path *path,
1272                                            int refs_to_drop)
1273 {
1274         struct btrfs_key key;
1275         struct btrfs_extent_data_ref *ref1 = NULL;
1276         struct btrfs_shared_data_ref *ref2 = NULL;
1277         struct extent_buffer *leaf;
1278         u32 num_refs = 0;
1279         int ret = 0;
1280
1281         leaf = path->nodes[0];
1282         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1283
1284         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1285                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1286                                       struct btrfs_extent_data_ref);
1287                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1288         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1289                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1290                                       struct btrfs_shared_data_ref);
1291                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1292 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1293         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1294                 struct btrfs_extent_ref_v0 *ref0;
1295                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1296                                       struct btrfs_extent_ref_v0);
1297                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1298 #endif
1299         } else {
1300                 BUG();
1301         }
1302
1303         BUG_ON(num_refs < refs_to_drop);
1304         num_refs -= refs_to_drop;
1305
1306         if (num_refs == 0) {
1307                 ret = btrfs_del_item(trans, root, path);
1308         } else {
1309                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1310                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1311                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1312                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1313 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1314                 else {
1315                         struct btrfs_extent_ref_v0 *ref0;
1316                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1317                                         struct btrfs_extent_ref_v0);
1318                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1319                 }
1320 #endif
1321                 btrfs_mark_buffer_dirty(leaf);
1322         }
1323         return ret;
1324 }
1325
1326 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1327                                           struct btrfs_path *path,
1328                                           struct btrfs_extent_inline_ref *iref)
1329 {
1330         struct btrfs_key key;
1331         struct extent_buffer *leaf;
1332         struct btrfs_extent_data_ref *ref1;
1333         struct btrfs_shared_data_ref *ref2;
1334         u32 num_refs = 0;
1335
1336         leaf = path->nodes[0];
1337         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1338         if (iref) {
1339                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1340                     BTRFS_EXTENT_DATA_REF_KEY) {
1341                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1342                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1343                 } else {
1344                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1345                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1346                 }
1347         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1348                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1349                                       struct btrfs_extent_data_ref);
1350                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1351         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1352                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1353                                       struct btrfs_shared_data_ref);
1354                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1355 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1356         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1357                 struct btrfs_extent_ref_v0 *ref0;
1358                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1359                                       struct btrfs_extent_ref_v0);
1360                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1361 #endif
1362         } else {
1363                 WARN_ON(1);
1364         }
1365         return num_refs;
1366 }
1367
1368 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1369                                           struct btrfs_root *root,
1370                                           struct btrfs_path *path,
1371                                           u64 bytenr, u64 parent,
1372                                           u64 root_objectid)
1373 {
1374         struct btrfs_key key;
1375         int ret;
1376
1377         key.objectid = bytenr;
1378         if (parent) {
1379                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1380                 key.offset = parent;
1381         } else {
1382                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1383                 key.offset = root_objectid;
1384         }
1385
1386         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1387         if (ret > 0)
1388                 ret = -ENOENT;
1389 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1390         if (ret == -ENOENT && parent) {
1391                 btrfs_release_path(path);
1392                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1393                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1394                 if (ret > 0)
1395                         ret = -ENOENT;
1396         }
1397 #endif
1398         return ret;
1399 }
1400
1401 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1402                                           struct btrfs_root *root,
1403                                           struct btrfs_path *path,
1404                                           u64 bytenr, u64 parent,
1405                                           u64 root_objectid)
1406 {
1407         struct btrfs_key key;
1408         int ret;
1409
1410         key.objectid = bytenr;
1411         if (parent) {
1412                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1413                 key.offset = parent;
1414         } else {
1415                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1416                 key.offset = root_objectid;
1417         }
1418
1419         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1420         btrfs_release_path(path);
1421         return ret;
1422 }
1423
1424 static inline int extent_ref_type(u64 parent, u64 owner)
1425 {
1426         int type;
1427         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1428                 if (parent > 0)
1429                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1430                 else
1431                         type = BTRFS_TREE_BLOCK_REF_KEY;
1432         } else {
1433                 if (parent > 0)
1434                         type = BTRFS_SHARED_DATA_REF_KEY;
1435                 else
1436                         type = BTRFS_EXTENT_DATA_REF_KEY;
1437         }
1438         return type;
1439 }
1440
1441 static int find_next_key(struct btrfs_path *path, int level,
1442                          struct btrfs_key *key)
1443
1444 {
1445         for (; level < BTRFS_MAX_LEVEL; level++) {
1446                 if (!path->nodes[level])
1447                         break;
1448                 if (path->slots[level] + 1 >=
1449                     btrfs_header_nritems(path->nodes[level]))
1450                         continue;
1451                 if (level == 0)
1452                         btrfs_item_key_to_cpu(path->nodes[level], key,
1453                                               path->slots[level] + 1);
1454                 else
1455                         btrfs_node_key_to_cpu(path->nodes[level], key,
1456                                               path->slots[level] + 1);
1457                 return 0;
1458         }
1459         return 1;
1460 }
1461
1462 /*
1463  * look for inline back ref. if back ref is found, *ref_ret is set
1464  * to the address of inline back ref, and 0 is returned.
1465  *
1466  * if back ref isn't found, *ref_ret is set to the address where it
1467  * should be inserted, and -ENOENT is returned.
1468  *
1469  * if insert is true and there are too many inline back refs, the path
1470  * points to the extent item, and -EAGAIN is returned.
1471  *
1472  * NOTE: inline back refs are ordered in the same way that back ref
1473  *       items in the tree are ordered.
1474  */
1475 static noinline_for_stack
1476 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1477                                  struct btrfs_root *root,
1478                                  struct btrfs_path *path,
1479                                  struct btrfs_extent_inline_ref **ref_ret,
1480                                  u64 bytenr, u64 num_bytes,
1481                                  u64 parent, u64 root_objectid,
1482                                  u64 owner, u64 offset, int insert)
1483 {
1484         struct btrfs_key key;
1485         struct extent_buffer *leaf;
1486         struct btrfs_extent_item *ei;
1487         struct btrfs_extent_inline_ref *iref;
1488         u64 flags;
1489         u64 item_size;
1490         unsigned long ptr;
1491         unsigned long end;
1492         int extra_size;
1493         int type;
1494         int want;
1495         int ret;
1496         int err = 0;
1497         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1498                                                  SKINNY_METADATA);
1499
1500         key.objectid = bytenr;
1501         key.type = BTRFS_EXTENT_ITEM_KEY;
1502         key.offset = num_bytes;
1503
1504         want = extent_ref_type(parent, owner);
1505         if (insert) {
1506                 extra_size = btrfs_extent_inline_ref_size(want);
1507                 path->keep_locks = 1;
1508         } else
1509                 extra_size = -1;
1510
1511         /*
1512          * Owner is our parent level, so we can just add one to get the level
1513          * for the block we are interested in.
1514          */
1515         if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1516                 key.type = BTRFS_METADATA_ITEM_KEY;
1517                 key.offset = owner;
1518         }
1519
1520 again:
1521         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1522         if (ret < 0) {
1523                 err = ret;
1524                 goto out;
1525         }
1526
1527         /*
1528          * We may be a newly converted file system which still has the old fat
1529          * extent entries for metadata, so try and see if we have one of those.
1530          */
1531         if (ret > 0 && skinny_metadata) {
1532                 skinny_metadata = false;
1533                 if (path->slots[0]) {
1534                         path->slots[0]--;
1535                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1536                                               path->slots[0]);
1537                         if (key.objectid == bytenr &&
1538                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1539                             key.offset == num_bytes)
1540                                 ret = 0;
1541                 }
1542                 if (ret) {
1543                         key.type = BTRFS_EXTENT_ITEM_KEY;
1544                         key.offset = num_bytes;
1545                         btrfs_release_path(path);
1546                         goto again;
1547                 }
1548         }
1549
1550         if (ret && !insert) {
1551                 err = -ENOENT;
1552                 goto out;
1553         } else if (WARN_ON(ret)) {
1554                 err = -EIO;
1555                 goto out;
1556         }
1557
1558         leaf = path->nodes[0];
1559         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1560 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1561         if (item_size < sizeof(*ei)) {
1562                 if (!insert) {
1563                         err = -ENOENT;
1564                         goto out;
1565                 }
1566                 ret = convert_extent_item_v0(trans, root, path, owner,
1567                                              extra_size);
1568                 if (ret < 0) {
1569                         err = ret;
1570                         goto out;
1571                 }
1572                 leaf = path->nodes[0];
1573                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1574         }
1575 #endif
1576         BUG_ON(item_size < sizeof(*ei));
1577
1578         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1579         flags = btrfs_extent_flags(leaf, ei);
1580
1581         ptr = (unsigned long)(ei + 1);
1582         end = (unsigned long)ei + item_size;
1583
1584         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1585                 ptr += sizeof(struct btrfs_tree_block_info);
1586                 BUG_ON(ptr > end);
1587         }
1588
1589         err = -ENOENT;
1590         while (1) {
1591                 if (ptr >= end) {
1592                         WARN_ON(ptr > end);
1593                         break;
1594                 }
1595                 iref = (struct btrfs_extent_inline_ref *)ptr;
1596                 type = btrfs_extent_inline_ref_type(leaf, iref);
1597                 if (want < type)
1598                         break;
1599                 if (want > type) {
1600                         ptr += btrfs_extent_inline_ref_size(type);
1601                         continue;
1602                 }
1603
1604                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1605                         struct btrfs_extent_data_ref *dref;
1606                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1607                         if (match_extent_data_ref(leaf, dref, root_objectid,
1608                                                   owner, offset)) {
1609                                 err = 0;
1610                                 break;
1611                         }
1612                         if (hash_extent_data_ref_item(leaf, dref) <
1613                             hash_extent_data_ref(root_objectid, owner, offset))
1614                                 break;
1615                 } else {
1616                         u64 ref_offset;
1617                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1618                         if (parent > 0) {
1619                                 if (parent == ref_offset) {
1620                                         err = 0;
1621                                         break;
1622                                 }
1623                                 if (ref_offset < parent)
1624                                         break;
1625                         } else {
1626                                 if (root_objectid == ref_offset) {
1627                                         err = 0;
1628                                         break;
1629                                 }
1630                                 if (ref_offset < root_objectid)
1631                                         break;
1632                         }
1633                 }
1634                 ptr += btrfs_extent_inline_ref_size(type);
1635         }
1636         if (err == -ENOENT && insert) {
1637                 if (item_size + extra_size >=
1638                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1639                         err = -EAGAIN;
1640                         goto out;
1641                 }
1642                 /*
1643                  * To add new inline back ref, we have to make sure
1644                  * there is no corresponding back ref item.
1645                  * For simplicity, we just do not add new inline back
1646                  * ref if there is any kind of item for this block
1647                  */
1648                 if (find_next_key(path, 0, &key) == 0 &&
1649                     key.objectid == bytenr &&
1650                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1651                         err = -EAGAIN;
1652                         goto out;
1653                 }
1654         }
1655         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1656 out:
1657         if (insert) {
1658                 path->keep_locks = 0;
1659                 btrfs_unlock_up_safe(path, 1);
1660         }
1661         return err;
1662 }
1663
1664 /*
1665  * helper to add new inline back ref
1666  */
1667 static noinline_for_stack
1668 void setup_inline_extent_backref(struct btrfs_root *root,
1669                                  struct btrfs_path *path,
1670                                  struct btrfs_extent_inline_ref *iref,
1671                                  u64 parent, u64 root_objectid,
1672                                  u64 owner, u64 offset, int refs_to_add,
1673                                  struct btrfs_delayed_extent_op *extent_op)
1674 {
1675         struct extent_buffer *leaf;
1676         struct btrfs_extent_item *ei;
1677         unsigned long ptr;
1678         unsigned long end;
1679         unsigned long item_offset;
1680         u64 refs;
1681         int size;
1682         int type;
1683
1684         leaf = path->nodes[0];
1685         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1686         item_offset = (unsigned long)iref - (unsigned long)ei;
1687
1688         type = extent_ref_type(parent, owner);
1689         size = btrfs_extent_inline_ref_size(type);
1690
1691         btrfs_extend_item(root, path, size);
1692
1693         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1694         refs = btrfs_extent_refs(leaf, ei);
1695         refs += refs_to_add;
1696         btrfs_set_extent_refs(leaf, ei, refs);
1697         if (extent_op)
1698                 __run_delayed_extent_op(extent_op, leaf, ei);
1699
1700         ptr = (unsigned long)ei + item_offset;
1701         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1702         if (ptr < end - size)
1703                 memmove_extent_buffer(leaf, ptr + size, ptr,
1704                                       end - size - ptr);
1705
1706         iref = (struct btrfs_extent_inline_ref *)ptr;
1707         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1708         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1709                 struct btrfs_extent_data_ref *dref;
1710                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1711                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1712                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1713                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1714                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1715         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1716                 struct btrfs_shared_data_ref *sref;
1717                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1718                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1719                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1720         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1721                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1722         } else {
1723                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1724         }
1725         btrfs_mark_buffer_dirty(leaf);
1726 }
1727
1728 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1729                                  struct btrfs_root *root,
1730                                  struct btrfs_path *path,
1731                                  struct btrfs_extent_inline_ref **ref_ret,
1732                                  u64 bytenr, u64 num_bytes, u64 parent,
1733                                  u64 root_objectid, u64 owner, u64 offset)
1734 {
1735         int ret;
1736
1737         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1738                                            bytenr, num_bytes, parent,
1739                                            root_objectid, owner, offset, 0);
1740         if (ret != -ENOENT)
1741                 return ret;
1742
1743         btrfs_release_path(path);
1744         *ref_ret = NULL;
1745
1746         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1747                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1748                                             root_objectid);
1749         } else {
1750                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1751                                              root_objectid, owner, offset);
1752         }
1753         return ret;
1754 }
1755
1756 /*
1757  * helper to update/remove inline back ref
1758  */
1759 static noinline_for_stack
1760 void update_inline_extent_backref(struct btrfs_root *root,
1761                                   struct btrfs_path *path,
1762                                   struct btrfs_extent_inline_ref *iref,
1763                                   int refs_to_mod,
1764                                   struct btrfs_delayed_extent_op *extent_op)
1765 {
1766         struct extent_buffer *leaf;
1767         struct btrfs_extent_item *ei;
1768         struct btrfs_extent_data_ref *dref = NULL;
1769         struct btrfs_shared_data_ref *sref = NULL;
1770         unsigned long ptr;
1771         unsigned long end;
1772         u32 item_size;
1773         int size;
1774         int type;
1775         u64 refs;
1776
1777         leaf = path->nodes[0];
1778         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1779         refs = btrfs_extent_refs(leaf, ei);
1780         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1781         refs += refs_to_mod;
1782         btrfs_set_extent_refs(leaf, ei, refs);
1783         if (extent_op)
1784                 __run_delayed_extent_op(extent_op, leaf, ei);
1785
1786         type = btrfs_extent_inline_ref_type(leaf, iref);
1787
1788         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1789                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1790                 refs = btrfs_extent_data_ref_count(leaf, dref);
1791         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1792                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1793                 refs = btrfs_shared_data_ref_count(leaf, sref);
1794         } else {
1795                 refs = 1;
1796                 BUG_ON(refs_to_mod != -1);
1797         }
1798
1799         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1800         refs += refs_to_mod;
1801
1802         if (refs > 0) {
1803                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1804                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1805                 else
1806                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1807         } else {
1808                 size =  btrfs_extent_inline_ref_size(type);
1809                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1810                 ptr = (unsigned long)iref;
1811                 end = (unsigned long)ei + item_size;
1812                 if (ptr + size < end)
1813                         memmove_extent_buffer(leaf, ptr, ptr + size,
1814                                               end - ptr - size);
1815                 item_size -= size;
1816                 btrfs_truncate_item(root, path, item_size, 1);
1817         }
1818         btrfs_mark_buffer_dirty(leaf);
1819 }
1820
1821 static noinline_for_stack
1822 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1823                                  struct btrfs_root *root,
1824                                  struct btrfs_path *path,
1825                                  u64 bytenr, u64 num_bytes, u64 parent,
1826                                  u64 root_objectid, u64 owner,
1827                                  u64 offset, int refs_to_add,
1828                                  struct btrfs_delayed_extent_op *extent_op)
1829 {
1830         struct btrfs_extent_inline_ref *iref;
1831         int ret;
1832
1833         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1834                                            bytenr, num_bytes, parent,
1835                                            root_objectid, owner, offset, 1);
1836         if (ret == 0) {
1837                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1838                 update_inline_extent_backref(root, path, iref,
1839                                              refs_to_add, extent_op);
1840         } else if (ret == -ENOENT) {
1841                 setup_inline_extent_backref(root, path, iref, parent,
1842                                             root_objectid, owner, offset,
1843                                             refs_to_add, extent_op);
1844                 ret = 0;
1845         }
1846         return ret;
1847 }
1848
1849 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1850                                  struct btrfs_root *root,
1851                                  struct btrfs_path *path,
1852                                  u64 bytenr, u64 parent, u64 root_objectid,
1853                                  u64 owner, u64 offset, int refs_to_add)
1854 {
1855         int ret;
1856         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1857                 BUG_ON(refs_to_add != 1);
1858                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1859                                             parent, root_objectid);
1860         } else {
1861                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1862                                              parent, root_objectid,
1863                                              owner, offset, refs_to_add);
1864         }
1865         return ret;
1866 }
1867
1868 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1869                                  struct btrfs_root *root,
1870                                  struct btrfs_path *path,
1871                                  struct btrfs_extent_inline_ref *iref,
1872                                  int refs_to_drop, int is_data)
1873 {
1874         int ret = 0;
1875
1876         BUG_ON(!is_data && refs_to_drop != 1);
1877         if (iref) {
1878                 update_inline_extent_backref(root, path, iref,
1879                                              -refs_to_drop, NULL);
1880         } else if (is_data) {
1881                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1882         } else {
1883                 ret = btrfs_del_item(trans, root, path);
1884         }
1885         return ret;
1886 }
1887
1888 static int btrfs_issue_discard(struct block_device *bdev,
1889                                 u64 start, u64 len)
1890 {
1891         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1892 }
1893
1894 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1895                                 u64 num_bytes, u64 *actual_bytes)
1896 {
1897         int ret;
1898         u64 discarded_bytes = 0;
1899         struct btrfs_bio *bbio = NULL;
1900
1901
1902         /* Tell the block device(s) that the sectors can be discarded */
1903         ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1904                               bytenr, &num_bytes, &bbio, 0);
1905         /* Error condition is -ENOMEM */
1906         if (!ret) {
1907                 struct btrfs_bio_stripe *stripe = bbio->stripes;
1908                 int i;
1909
1910
1911                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1912                         if (!stripe->dev->can_discard)
1913                                 continue;
1914
1915                         ret = btrfs_issue_discard(stripe->dev->bdev,
1916                                                   stripe->physical,
1917                                                   stripe->length);
1918                         if (!ret)
1919                                 discarded_bytes += stripe->length;
1920                         else if (ret != -EOPNOTSUPP)
1921                                 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1922
1923                         /*
1924                          * Just in case we get back EOPNOTSUPP for some reason,
1925                          * just ignore the return value so we don't screw up
1926                          * people calling discard_extent.
1927                          */
1928                         ret = 0;
1929                 }
1930                 kfree(bbio);
1931         }
1932
1933         if (actual_bytes)
1934                 *actual_bytes = discarded_bytes;
1935
1936
1937         if (ret == -EOPNOTSUPP)
1938                 ret = 0;
1939         return ret;
1940 }
1941
1942 /* Can return -ENOMEM */
1943 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1944                          struct btrfs_root *root,
1945                          u64 bytenr, u64 num_bytes, u64 parent,
1946                          u64 root_objectid, u64 owner, u64 offset, int for_cow)
1947 {
1948         int ret;
1949         struct btrfs_fs_info *fs_info = root->fs_info;
1950
1951         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1952                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1953
1954         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1955                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1956                                         num_bytes,
1957                                         parent, root_objectid, (int)owner,
1958                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1959         } else {
1960                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1961                                         num_bytes,
1962                                         parent, root_objectid, owner, offset,
1963                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1964         }
1965         return ret;
1966 }
1967
1968 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1969                                   struct btrfs_root *root,
1970                                   u64 bytenr, u64 num_bytes,
1971                                   u64 parent, u64 root_objectid,
1972                                   u64 owner, u64 offset, int refs_to_add,
1973                                   struct btrfs_delayed_extent_op *extent_op)
1974 {
1975         struct btrfs_path *path;
1976         struct extent_buffer *leaf;
1977         struct btrfs_extent_item *item;
1978         u64 refs;
1979         int ret;
1980
1981         path = btrfs_alloc_path();
1982         if (!path)
1983                 return -ENOMEM;
1984
1985         path->reada = 1;
1986         path->leave_spinning = 1;
1987         /* this will setup the path even if it fails to insert the back ref */
1988         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1989                                            path, bytenr, num_bytes, parent,
1990                                            root_objectid, owner, offset,
1991                                            refs_to_add, extent_op);
1992         if (ret != -EAGAIN)
1993                 goto out;
1994
1995         leaf = path->nodes[0];
1996         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1997         refs = btrfs_extent_refs(leaf, item);
1998         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1999         if (extent_op)
2000                 __run_delayed_extent_op(extent_op, leaf, item);
2001
2002         btrfs_mark_buffer_dirty(leaf);
2003         btrfs_release_path(path);
2004
2005         path->reada = 1;
2006         path->leave_spinning = 1;
2007
2008         /* now insert the actual backref */
2009         ret = insert_extent_backref(trans, root->fs_info->extent_root,
2010                                     path, bytenr, parent, root_objectid,
2011                                     owner, offset, refs_to_add);
2012         if (ret)
2013                 btrfs_abort_transaction(trans, root, ret);
2014 out:
2015         btrfs_free_path(path);
2016         return ret;
2017 }
2018
2019 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2020                                 struct btrfs_root *root,
2021                                 struct btrfs_delayed_ref_node *node,
2022                                 struct btrfs_delayed_extent_op *extent_op,
2023                                 int insert_reserved)
2024 {
2025         int ret = 0;
2026         struct btrfs_delayed_data_ref *ref;
2027         struct btrfs_key ins;
2028         u64 parent = 0;
2029         u64 ref_root = 0;
2030         u64 flags = 0;
2031
2032         ins.objectid = node->bytenr;
2033         ins.offset = node->num_bytes;
2034         ins.type = BTRFS_EXTENT_ITEM_KEY;
2035
2036         ref = btrfs_delayed_node_to_data_ref(node);
2037         trace_run_delayed_data_ref(node, ref, node->action);
2038
2039         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2040                 parent = ref->parent;
2041         else
2042                 ref_root = ref->root;
2043
2044         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2045                 if (extent_op)
2046                         flags |= extent_op->flags_to_set;
2047                 ret = alloc_reserved_file_extent(trans, root,
2048                                                  parent, ref_root, flags,
2049                                                  ref->objectid, ref->offset,
2050                                                  &ins, node->ref_mod);
2051         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2052                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2053                                              node->num_bytes, parent,
2054                                              ref_root, ref->objectid,
2055                                              ref->offset, node->ref_mod,
2056                                              extent_op);
2057         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2058                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2059                                           node->num_bytes, parent,
2060                                           ref_root, ref->objectid,
2061                                           ref->offset, node->ref_mod,
2062                                           extent_op);
2063         } else {
2064                 BUG();
2065         }
2066         return ret;
2067 }
2068
2069 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2070                                     struct extent_buffer *leaf,
2071                                     struct btrfs_extent_item *ei)
2072 {
2073         u64 flags = btrfs_extent_flags(leaf, ei);
2074         if (extent_op->update_flags) {
2075                 flags |= extent_op->flags_to_set;
2076                 btrfs_set_extent_flags(leaf, ei, flags);
2077         }
2078
2079         if (extent_op->update_key) {
2080                 struct btrfs_tree_block_info *bi;
2081                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2082                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2083                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2084         }
2085 }
2086
2087 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2088                                  struct btrfs_root *root,
2089                                  struct btrfs_delayed_ref_node *node,
2090                                  struct btrfs_delayed_extent_op *extent_op)
2091 {
2092         struct btrfs_key key;
2093         struct btrfs_path *path;
2094         struct btrfs_extent_item *ei;
2095         struct extent_buffer *leaf;
2096         u32 item_size;
2097         int ret;
2098         int err = 0;
2099         int metadata = !extent_op->is_data;
2100
2101         if (trans->aborted)
2102                 return 0;
2103
2104         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2105                 metadata = 0;
2106
2107         path = btrfs_alloc_path();
2108         if (!path)
2109                 return -ENOMEM;
2110
2111         key.objectid = node->bytenr;
2112
2113         if (metadata) {
2114                 key.type = BTRFS_METADATA_ITEM_KEY;
2115                 key.offset = extent_op->level;
2116         } else {
2117                 key.type = BTRFS_EXTENT_ITEM_KEY;
2118                 key.offset = node->num_bytes;
2119         }
2120
2121 again:
2122         path->reada = 1;
2123         path->leave_spinning = 1;
2124         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2125                                 path, 0, 1);
2126         if (ret < 0) {
2127                 err = ret;
2128                 goto out;
2129         }
2130         if (ret > 0) {
2131                 if (metadata) {
2132                         if (path->slots[0] > 0) {
2133                                 path->slots[0]--;
2134                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
2135                                                       path->slots[0]);
2136                                 if (key.objectid == node->bytenr &&
2137                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
2138                                     key.offset == node->num_bytes)
2139                                         ret = 0;
2140                         }
2141                         if (ret > 0) {
2142                                 btrfs_release_path(path);
2143                                 metadata = 0;
2144
2145                                 key.objectid = node->bytenr;
2146                                 key.offset = node->num_bytes;
2147                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2148                                 goto again;
2149                         }
2150                 } else {
2151                         err = -EIO;
2152                         goto out;
2153                 }
2154         }
2155
2156         leaf = path->nodes[0];
2157         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2158 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2159         if (item_size < sizeof(*ei)) {
2160                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2161                                              path, (u64)-1, 0);
2162                 if (ret < 0) {
2163                         err = ret;
2164                         goto out;
2165                 }
2166                 leaf = path->nodes[0];
2167                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2168         }
2169 #endif
2170         BUG_ON(item_size < sizeof(*ei));
2171         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2172         __run_delayed_extent_op(extent_op, leaf, ei);
2173
2174         btrfs_mark_buffer_dirty(leaf);
2175 out:
2176         btrfs_free_path(path);
2177         return err;
2178 }
2179
2180 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2181                                 struct btrfs_root *root,
2182                                 struct btrfs_delayed_ref_node *node,
2183                                 struct btrfs_delayed_extent_op *extent_op,
2184                                 int insert_reserved)
2185 {
2186         int ret = 0;
2187         struct btrfs_delayed_tree_ref *ref;
2188         struct btrfs_key ins;
2189         u64 parent = 0;
2190         u64 ref_root = 0;
2191         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2192                                                  SKINNY_METADATA);
2193
2194         ref = btrfs_delayed_node_to_tree_ref(node);
2195         trace_run_delayed_tree_ref(node, ref, node->action);
2196
2197         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2198                 parent = ref->parent;
2199         else
2200                 ref_root = ref->root;
2201
2202         ins.objectid = node->bytenr;
2203         if (skinny_metadata) {
2204                 ins.offset = ref->level;
2205                 ins.type = BTRFS_METADATA_ITEM_KEY;
2206         } else {
2207                 ins.offset = node->num_bytes;
2208                 ins.type = BTRFS_EXTENT_ITEM_KEY;
2209         }
2210
2211         BUG_ON(node->ref_mod != 1);
2212         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2213                 BUG_ON(!extent_op || !extent_op->update_flags);
2214                 ret = alloc_reserved_tree_block(trans, root,
2215                                                 parent, ref_root,
2216                                                 extent_op->flags_to_set,
2217                                                 &extent_op->key,
2218                                                 ref->level, &ins);
2219         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2220                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2221                                              node->num_bytes, parent, ref_root,
2222                                              ref->level, 0, 1, extent_op);
2223         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2224                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2225                                           node->num_bytes, parent, ref_root,
2226                                           ref->level, 0, 1, extent_op);
2227         } else {
2228                 BUG();
2229         }
2230         return ret;
2231 }
2232
2233 /* helper function to actually process a single delayed ref entry */
2234 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2235                                struct btrfs_root *root,
2236                                struct btrfs_delayed_ref_node *node,
2237                                struct btrfs_delayed_extent_op *extent_op,
2238                                int insert_reserved)
2239 {
2240         int ret = 0;
2241
2242         if (trans->aborted) {
2243                 if (insert_reserved)
2244                         btrfs_pin_extent(root, node->bytenr,
2245                                          node->num_bytes, 1);
2246                 return 0;
2247         }
2248
2249         if (btrfs_delayed_ref_is_head(node)) {
2250                 struct btrfs_delayed_ref_head *head;
2251                 /*
2252                  * we've hit the end of the chain and we were supposed
2253                  * to insert this extent into the tree.  But, it got
2254                  * deleted before we ever needed to insert it, so all
2255                  * we have to do is clean up the accounting
2256                  */
2257                 BUG_ON(extent_op);
2258                 head = btrfs_delayed_node_to_head(node);
2259                 trace_run_delayed_ref_head(node, head, node->action);
2260
2261                 if (insert_reserved) {
2262                         btrfs_pin_extent(root, node->bytenr,
2263                                          node->num_bytes, 1);
2264                         if (head->is_data) {
2265                                 ret = btrfs_del_csums(trans, root,
2266                                                       node->bytenr,
2267                                                       node->num_bytes);
2268                         }
2269                 }
2270                 return ret;
2271         }
2272
2273         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2274             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2275                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2276                                            insert_reserved);
2277         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2278                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2279                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2280                                            insert_reserved);
2281         else
2282                 BUG();
2283         return ret;
2284 }
2285
2286 static noinline struct btrfs_delayed_ref_node *
2287 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2288 {
2289         struct rb_node *node;
2290         struct btrfs_delayed_ref_node *ref;
2291         int action = BTRFS_ADD_DELAYED_REF;
2292 again:
2293         /*
2294          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2295          * this prevents ref count from going down to zero when
2296          * there still are pending delayed ref.
2297          */
2298         node = rb_prev(&head->node.rb_node);
2299         while (1) {
2300                 if (!node)
2301                         break;
2302                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2303                                 rb_node);
2304                 if (ref->bytenr != head->node.bytenr)
2305                         break;
2306                 if (ref->action == action)
2307                         return ref;
2308                 node = rb_prev(node);
2309         }
2310         if (action == BTRFS_ADD_DELAYED_REF) {
2311                 action = BTRFS_DROP_DELAYED_REF;
2312                 goto again;
2313         }
2314         return NULL;
2315 }
2316
2317 /*
2318  * Returns 0 on success or if called with an already aborted transaction.
2319  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2320  */
2321 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2322                                        struct btrfs_root *root,
2323                                        struct list_head *cluster)
2324 {
2325         struct btrfs_delayed_ref_root *delayed_refs;
2326         struct btrfs_delayed_ref_node *ref;
2327         struct btrfs_delayed_ref_head *locked_ref = NULL;
2328         struct btrfs_delayed_extent_op *extent_op;
2329         struct btrfs_fs_info *fs_info = root->fs_info;
2330         int ret;
2331         int count = 0;
2332         int must_insert_reserved = 0;
2333
2334         delayed_refs = &trans->transaction->delayed_refs;
2335         while (1) {
2336                 if (!locked_ref) {
2337                         /* pick a new head ref from the cluster list */
2338                         if (list_empty(cluster))
2339                                 break;
2340
2341                         locked_ref = list_entry(cluster->next,
2342                                      struct btrfs_delayed_ref_head, cluster);
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
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                 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2369                                          locked_ref);
2370
2371                 /*
2372                  * locked_ref is the head node, so we have to go one
2373                  * node back for any delayed ref updates
2374                  */
2375                 ref = select_delayed_ref(locked_ref);
2376
2377                 if (ref && ref->seq &&
2378                     btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2379                         /*
2380                          * there are still refs with lower seq numbers in the
2381                          * process of being added. Don't run this ref yet.
2382                          */
2383                         list_del_init(&locked_ref->cluster);
2384                         btrfs_delayed_ref_unlock(locked_ref);
2385                         locked_ref = NULL;
2386                         delayed_refs->num_heads_ready++;
2387                         spin_unlock(&delayed_refs->lock);
2388                         cond_resched();
2389                         spin_lock(&delayed_refs->lock);
2390                         continue;
2391                 }
2392
2393                 /*
2394                  * record the must insert reserved flag before we
2395                  * drop the spin lock.
2396                  */
2397                 must_insert_reserved = locked_ref->must_insert_reserved;
2398                 locked_ref->must_insert_reserved = 0;
2399
2400                 extent_op = locked_ref->extent_op;
2401                 locked_ref->extent_op = NULL;
2402
2403                 if (!ref) {
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(&delayed_refs->lock);
2417
2418                                 ret = run_delayed_extent_op(trans, root,
2419                                                             ref, extent_op);
2420                                 btrfs_free_delayed_extent_op(extent_op);
2421
2422                                 if (ret) {
2423                                         /*
2424                                          * Need to reset must_insert_reserved if
2425                                          * there was an error so the abort stuff
2426                                          * can cleanup the reserved space
2427                                          * properly.
2428                                          */
2429                                         if (must_insert_reserved)
2430                                                 locked_ref->must_insert_reserved = 1;
2431                                         btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2432                                         spin_lock(&delayed_refs->lock);
2433                                         btrfs_delayed_ref_unlock(locked_ref);
2434                                         return ret;
2435                                 }
2436
2437                                 goto next;
2438                         }
2439                 }
2440
2441                 ref->in_tree = 0;
2442                 rb_erase(&ref->rb_node, &delayed_refs->root);
2443                 delayed_refs->num_entries--;
2444                 if (!btrfs_delayed_ref_is_head(ref)) {
2445                         /*
2446                          * when we play the delayed ref, also correct the
2447                          * ref_mod on head
2448                          */
2449                         switch (ref->action) {
2450                         case BTRFS_ADD_DELAYED_REF:
2451                         case BTRFS_ADD_DELAYED_EXTENT:
2452                                 locked_ref->node.ref_mod -= ref->ref_mod;
2453                                 break;
2454                         case BTRFS_DROP_DELAYED_REF:
2455                                 locked_ref->node.ref_mod += ref->ref_mod;
2456                                 break;
2457                         default:
2458                                 WARN_ON(1);
2459                         }
2460                 } else {
2461                         list_del_init(&locked_ref->cluster);
2462                 }
2463                 spin_unlock(&delayed_refs->lock);
2464
2465                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2466                                           must_insert_reserved);
2467
2468                 btrfs_free_delayed_extent_op(extent_op);
2469                 if (ret) {
2470                         btrfs_delayed_ref_unlock(locked_ref);
2471                         btrfs_put_delayed_ref(ref);
2472                         btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2473                         spin_lock(&delayed_refs->lock);
2474                         return ret;
2475                 }
2476
2477                 /*
2478                  * If this node is a head, that means all the refs in this head
2479                  * have been dealt with, and we will pick the next head to deal
2480                  * with, so we must unlock the head and drop it from the cluster
2481                  * list before we release it.
2482                  */
2483                 if (btrfs_delayed_ref_is_head(ref)) {
2484                         btrfs_delayed_ref_unlock(locked_ref);
2485                         locked_ref = NULL;
2486                 }
2487                 btrfs_put_delayed_ref(ref);
2488                 count++;
2489 next:
2490                 cond_resched();
2491                 spin_lock(&delayed_refs->lock);
2492         }
2493         return count;
2494 }
2495
2496 #ifdef SCRAMBLE_DELAYED_REFS
2497 /*
2498  * Normally delayed refs get processed in ascending bytenr order. This
2499  * correlates in most cases to the order added. To expose dependencies on this
2500  * order, we start to process the tree in the middle instead of the beginning
2501  */
2502 static u64 find_middle(struct rb_root *root)
2503 {
2504         struct rb_node *n = root->rb_node;
2505         struct btrfs_delayed_ref_node *entry;
2506         int alt = 1;
2507         u64 middle;
2508         u64 first = 0, last = 0;
2509
2510         n = rb_first(root);
2511         if (n) {
2512                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2513                 first = entry->bytenr;
2514         }
2515         n = rb_last(root);
2516         if (n) {
2517                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2518                 last = entry->bytenr;
2519         }
2520         n = root->rb_node;
2521
2522         while (n) {
2523                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2524                 WARN_ON(!entry->in_tree);
2525
2526                 middle = entry->bytenr;
2527
2528                 if (alt)
2529                         n = n->rb_left;
2530                 else
2531                         n = n->rb_right;
2532
2533                 alt = 1 - alt;
2534         }
2535         return middle;
2536 }
2537 #endif
2538
2539 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2540                                          struct btrfs_fs_info *fs_info)
2541 {
2542         struct qgroup_update *qgroup_update;
2543         int ret = 0;
2544
2545         if (list_empty(&trans->qgroup_ref_list) !=
2546             !trans->delayed_ref_elem.seq) {
2547                 /* list without seq or seq without list */
2548                 btrfs_err(fs_info,
2549                         "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2550                         list_empty(&trans->qgroup_ref_list) ? "" : " not",
2551                         (u32)(trans->delayed_ref_elem.seq >> 32),
2552                         (u32)trans->delayed_ref_elem.seq);
2553                 BUG();
2554         }
2555
2556         if (!trans->delayed_ref_elem.seq)
2557                 return 0;
2558
2559         while (!list_empty(&trans->qgroup_ref_list)) {
2560                 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2561                                                  struct qgroup_update, list);
2562                 list_del(&qgroup_update->list);
2563                 if (!ret)
2564                         ret = btrfs_qgroup_account_ref(
2565                                         trans, fs_info, qgroup_update->node,
2566                                         qgroup_update->extent_op);
2567                 kfree(qgroup_update);
2568         }
2569
2570         btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2571
2572         return ret;
2573 }
2574
2575 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2576                       int count)
2577 {
2578         int val = atomic_read(&delayed_refs->ref_seq);
2579
2580         if (val < seq || val >= seq + count)
2581                 return 1;
2582         return 0;
2583 }
2584
2585 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2586 {
2587         u64 num_bytes;
2588
2589         num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2590                              sizeof(struct btrfs_extent_inline_ref));
2591         if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2592                 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2593
2594         /*
2595          * We don't ever fill up leaves all the way so multiply by 2 just to be
2596          * closer to what we're really going to want to ouse.
2597          */
2598         return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2599 }
2600
2601 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2602                                        struct btrfs_root *root)
2603 {
2604         struct btrfs_block_rsv *global_rsv;
2605         u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2606         u64 num_bytes;
2607         int ret = 0;
2608
2609         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2610         num_heads = heads_to_leaves(root, num_heads);
2611         if (num_heads > 1)
2612                 num_bytes += (num_heads - 1) * root->leafsize;
2613         num_bytes <<= 1;
2614         global_rsv = &root->fs_info->global_block_rsv;
2615
2616         /*
2617          * If we can't allocate any more chunks lets make sure we have _lots_ of
2618          * wiggle room since running delayed refs can create more delayed refs.
2619          */
2620         if (global_rsv->space_info->full)
2621                 num_bytes <<= 1;
2622
2623         spin_lock(&global_rsv->lock);
2624         if (global_rsv->reserved <= num_bytes)
2625                 ret = 1;
2626         spin_unlock(&global_rsv->lock);
2627         return ret;
2628 }
2629
2630 /*
2631  * this starts processing the delayed reference count updates and
2632  * extent insertions we have queued up so far.  count can be
2633  * 0, which means to process everything in the tree at the start
2634  * of the run (but not newly added entries), or it can be some target
2635  * number you'd like to process.
2636  *
2637  * Returns 0 on success or if called with an aborted transaction
2638  * Returns <0 on error and aborts the transaction
2639  */
2640 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2641                            struct btrfs_root *root, unsigned long count)
2642 {
2643         struct rb_node *node;
2644         struct btrfs_delayed_ref_root *delayed_refs;
2645         struct btrfs_delayed_ref_node *ref;
2646         struct list_head cluster;
2647         int ret;
2648         u64 delayed_start;
2649         int run_all = count == (unsigned long)-1;
2650         int run_most = 0;
2651         int loops;
2652
2653         /* We'll clean this up in btrfs_cleanup_transaction */
2654         if (trans->aborted)
2655                 return 0;
2656
2657         if (root == root->fs_info->extent_root)
2658                 root = root->fs_info->tree_root;
2659
2660         btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2661
2662         delayed_refs = &trans->transaction->delayed_refs;
2663         INIT_LIST_HEAD(&cluster);
2664         if (count == 0) {
2665                 count = delayed_refs->num_entries * 2;
2666                 run_most = 1;
2667         }
2668
2669         if (!run_all && !run_most) {
2670                 int old;
2671                 int seq = atomic_read(&delayed_refs->ref_seq);
2672
2673 progress:
2674                 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2675                 if (old) {
2676                         DEFINE_WAIT(__wait);
2677                         if (delayed_refs->flushing ||
2678                             !btrfs_should_throttle_delayed_refs(trans, root))
2679                                 return 0;
2680
2681                         prepare_to_wait(&delayed_refs->wait, &__wait,
2682                                         TASK_UNINTERRUPTIBLE);
2683
2684                         old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2685                         if (old) {
2686                                 schedule();
2687                                 finish_wait(&delayed_refs->wait, &__wait);
2688
2689                                 if (!refs_newer(delayed_refs, seq, 256))
2690                                         goto progress;
2691                                 else
2692                                         return 0;
2693                         } else {
2694                                 finish_wait(&delayed_refs->wait, &__wait);
2695                                 goto again;
2696                         }
2697                 }
2698
2699         } else {
2700                 atomic_inc(&delayed_refs->procs_running_refs);
2701         }
2702
2703 again:
2704         loops = 0;
2705         spin_lock(&delayed_refs->lock);
2706
2707 #ifdef SCRAMBLE_DELAYED_REFS
2708         delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2709 #endif
2710
2711         while (1) {
2712                 if (!(run_all || run_most) &&
2713                     !btrfs_should_throttle_delayed_refs(trans, root))
2714                         break;
2715
2716                 /*
2717                  * go find something we can process in the rbtree.  We start at
2718                  * the beginning of the tree, and then build a cluster
2719                  * of refs to process starting at the first one we are able to
2720                  * lock
2721                  */
2722                 delayed_start = delayed_refs->run_delayed_start;
2723                 ret = btrfs_find_ref_cluster(trans, &cluster,
2724                                              delayed_refs->run_delayed_start);
2725                 if (ret)
2726                         break;
2727
2728                 ret = run_clustered_refs(trans, root, &cluster);
2729                 if (ret < 0) {
2730                         btrfs_release_ref_cluster(&cluster);
2731                         spin_unlock(&delayed_refs->lock);
2732                         btrfs_abort_transaction(trans, root, ret);
2733                         atomic_dec(&delayed_refs->procs_running_refs);
2734                         wake_up(&delayed_refs->wait);
2735                         return ret;
2736                 }
2737
2738                 atomic_add(ret, &delayed_refs->ref_seq);
2739
2740                 count -= min_t(unsigned long, ret, count);
2741
2742                 if (count == 0)
2743                         break;
2744
2745                 if (delayed_start >= delayed_refs->run_delayed_start) {
2746                         if (loops == 0) {
2747                                 /*
2748                                  * btrfs_find_ref_cluster looped. let's do one
2749                                  * more cycle. if we don't run any delayed ref
2750                                  * during that cycle (because we can't because
2751                                  * all of them are blocked), bail out.
2752                                  */
2753                                 loops = 1;
2754                         } else {
2755                                 /*
2756                                  * no runnable refs left, stop trying
2757                                  */
2758                                 BUG_ON(run_all);
2759                                 break;
2760                         }
2761                 }
2762                 if (ret) {
2763                         /* refs were run, let's reset staleness detection */
2764                         loops = 0;
2765                 }
2766         }
2767
2768         if (run_all) {
2769                 if (!list_empty(&trans->new_bgs)) {
2770                         spin_unlock(&delayed_refs->lock);
2771                         btrfs_create_pending_block_groups(trans, root);
2772                         spin_lock(&delayed_refs->lock);
2773                 }
2774
2775                 node = rb_first(&delayed_refs->root);
2776                 if (!node)
2777                         goto out;
2778                 count = (unsigned long)-1;
2779
2780                 while (node) {
2781                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2782                                        rb_node);
2783                         if (btrfs_delayed_ref_is_head(ref)) {
2784                                 struct btrfs_delayed_ref_head *head;
2785
2786                                 head = btrfs_delayed_node_to_head(ref);
2787                                 atomic_inc(&ref->refs);
2788
2789                                 spin_unlock(&delayed_refs->lock);
2790                                 /*
2791                                  * Mutex was contended, block until it's
2792                                  * released and try again
2793                                  */
2794                                 mutex_lock(&head->mutex);
2795                                 mutex_unlock(&head->mutex);
2796
2797                                 btrfs_put_delayed_ref(ref);
2798                                 cond_resched();
2799                                 goto again;
2800                         }
2801                         node = rb_next(node);
2802                 }
2803                 spin_unlock(&delayed_refs->lock);
2804                 schedule_timeout(1);
2805                 goto again;
2806         }
2807 out:
2808         atomic_dec(&delayed_refs->procs_running_refs);
2809         smp_mb();
2810         if (waitqueue_active(&delayed_refs->wait))
2811                 wake_up(&delayed_refs->wait);
2812
2813         spin_unlock(&delayed_refs->lock);
2814         assert_qgroups_uptodate(trans);
2815         return 0;
2816 }
2817
2818 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2819                                 struct btrfs_root *root,
2820                                 u64 bytenr, u64 num_bytes, u64 flags,
2821                                 int level, int is_data)
2822 {
2823         struct btrfs_delayed_extent_op *extent_op;
2824         int ret;
2825
2826         extent_op = btrfs_alloc_delayed_extent_op();
2827         if (!extent_op)
2828                 return -ENOMEM;
2829
2830         extent_op->flags_to_set = flags;
2831         extent_op->update_flags = 1;
2832         extent_op->update_key = 0;
2833         extent_op->is_data = is_data ? 1 : 0;
2834         extent_op->level = level;
2835
2836         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2837                                           num_bytes, extent_op);
2838         if (ret)
2839                 btrfs_free_delayed_extent_op(extent_op);
2840         return ret;
2841 }
2842
2843 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2844                                       struct btrfs_root *root,
2845                                       struct btrfs_path *path,
2846                                       u64 objectid, u64 offset, u64 bytenr)
2847 {
2848         struct btrfs_delayed_ref_head *head;
2849         struct btrfs_delayed_ref_node *ref;
2850         struct btrfs_delayed_data_ref *data_ref;
2851         struct btrfs_delayed_ref_root *delayed_refs;
2852         struct rb_node *node;
2853         int ret = 0;
2854
2855         ret = -ENOENT;
2856         delayed_refs = &trans->transaction->delayed_refs;
2857         spin_lock(&delayed_refs->lock);
2858         head = btrfs_find_delayed_ref_head(trans, bytenr);
2859         if (!head)
2860                 goto out;
2861
2862         if (!mutex_trylock(&head->mutex)) {
2863                 atomic_inc(&head->node.refs);
2864                 spin_unlock(&delayed_refs->lock);
2865
2866                 btrfs_release_path(path);
2867
2868                 /*
2869                  * Mutex was contended, block until it's released and let
2870                  * caller try again
2871                  */
2872                 mutex_lock(&head->mutex);
2873                 mutex_unlock(&head->mutex);
2874                 btrfs_put_delayed_ref(&head->node);
2875                 return -EAGAIN;
2876         }
2877
2878         node = rb_prev(&head->node.rb_node);
2879         if (!node)
2880                 goto out_unlock;
2881
2882         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2883
2884         if (ref->bytenr != bytenr)
2885                 goto out_unlock;
2886
2887         ret = 1;
2888         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2889                 goto out_unlock;
2890
2891         data_ref = btrfs_delayed_node_to_data_ref(ref);
2892
2893         node = rb_prev(node);
2894         if (node) {
2895                 int seq = ref->seq;
2896
2897                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2898                 if (ref->bytenr == bytenr && ref->seq == seq)
2899                         goto out_unlock;
2900         }
2901
2902         if (data_ref->root != root->root_key.objectid ||
2903             data_ref->objectid != objectid || data_ref->offset != offset)
2904                 goto out_unlock;
2905
2906         ret = 0;
2907 out_unlock:
2908         mutex_unlock(&head->mutex);
2909 out:
2910         spin_unlock(&delayed_refs->lock);
2911         return ret;
2912 }
2913
2914 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2915                                         struct btrfs_root *root,
2916                                         struct btrfs_path *path,
2917                                         u64 objectid, u64 offset, u64 bytenr)
2918 {
2919         struct btrfs_root *extent_root = root->fs_info->extent_root;
2920         struct extent_buffer *leaf;
2921         struct btrfs_extent_data_ref *ref;
2922         struct btrfs_extent_inline_ref *iref;
2923         struct btrfs_extent_item *ei;
2924         struct btrfs_key key;
2925         u32 item_size;
2926         int ret;
2927
2928         key.objectid = bytenr;
2929         key.offset = (u64)-1;
2930         key.type = BTRFS_EXTENT_ITEM_KEY;
2931
2932         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2933         if (ret < 0)
2934                 goto out;
2935         BUG_ON(ret == 0); /* Corruption */
2936
2937         ret = -ENOENT;
2938         if (path->slots[0] == 0)
2939                 goto out;
2940
2941         path->slots[0]--;
2942         leaf = path->nodes[0];
2943         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2944
2945         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2946                 goto out;
2947
2948         ret = 1;
2949         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2950 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2951         if (item_size < sizeof(*ei)) {
2952                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2953                 goto out;
2954         }
2955 #endif
2956         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2957
2958         if (item_size != sizeof(*ei) +
2959             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2960                 goto out;
2961
2962         if (btrfs_extent_generation(leaf, ei) <=
2963             btrfs_root_last_snapshot(&root->root_item))
2964                 goto out;
2965
2966         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2967         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2968             BTRFS_EXTENT_DATA_REF_KEY)
2969                 goto out;
2970
2971         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2972         if (btrfs_extent_refs(leaf, ei) !=
2973             btrfs_extent_data_ref_count(leaf, ref) ||
2974             btrfs_extent_data_ref_root(leaf, ref) !=
2975             root->root_key.objectid ||
2976             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2977             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2978                 goto out;
2979
2980         ret = 0;
2981 out:
2982         return ret;
2983 }
2984
2985 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2986                           struct btrfs_root *root,
2987                           u64 objectid, u64 offset, u64 bytenr)
2988 {
2989         struct btrfs_path *path;
2990         int ret;
2991         int ret2;
2992
2993         path = btrfs_alloc_path();
2994         if (!path)
2995                 return -ENOENT;
2996
2997         do {
2998                 ret = check_committed_ref(trans, root, path, objectid,
2999                                           offset, bytenr);
3000                 if (ret && ret != -ENOENT)
3001                         goto out;
3002
3003                 ret2 = check_delayed_ref(trans, root, path, objectid,
3004                                          offset, bytenr);
3005         } while (ret2 == -EAGAIN);
3006
3007         if (ret2 && ret2 != -ENOENT) {
3008                 ret = ret2;
3009                 goto out;
3010         }
3011
3012         if (ret != -ENOENT || ret2 != -ENOENT)
3013                 ret = 0;
3014 out:
3015         btrfs_free_path(path);
3016         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3017                 WARN_ON(ret > 0);
3018         return ret;
3019 }
3020
3021 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3022                            struct btrfs_root *root,
3023                            struct extent_buffer *buf,
3024                            int full_backref, int inc, int for_cow)
3025 {
3026         u64 bytenr;
3027         u64 num_bytes;
3028         u64 parent;
3029         u64 ref_root;
3030         u32 nritems;
3031         struct btrfs_key key;
3032         struct btrfs_file_extent_item *fi;
3033         int i;
3034         int level;
3035         int ret = 0;
3036         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3037                             u64, u64, u64, u64, u64, u64, int);
3038
3039         ref_root = btrfs_header_owner(buf);
3040         nritems = btrfs_header_nritems(buf);
3041         level = btrfs_header_level(buf);
3042
3043         if (!root->ref_cows && level == 0)
3044                 return 0;
3045
3046         if (inc)
3047                 process_func = btrfs_inc_extent_ref;
3048         else
3049                 process_func = btrfs_free_extent;
3050
3051         if (full_backref)
3052                 parent = buf->start;
3053         else
3054                 parent = 0;
3055
3056         for (i = 0; i < nritems; i++) {
3057                 if (level == 0) {
3058                         btrfs_item_key_to_cpu(buf, &key, i);
3059                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3060                                 continue;
3061                         fi = btrfs_item_ptr(buf, i,
3062                                             struct btrfs_file_extent_item);
3063                         if (btrfs_file_extent_type(buf, fi) ==
3064                             BTRFS_FILE_EXTENT_INLINE)
3065                                 continue;
3066                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3067                         if (bytenr == 0)
3068                                 continue;
3069
3070                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3071                         key.offset -= btrfs_file_extent_offset(buf, fi);
3072                         ret = process_func(trans, root, bytenr, num_bytes,
3073                                            parent, ref_root, key.objectid,
3074                                            key.offset, for_cow);
3075                         if (ret)
3076                                 goto fail;
3077                 } else {
3078                         bytenr = btrfs_node_blockptr(buf, i);
3079                         num_bytes = btrfs_level_size(root, level - 1);
3080                         ret = process_func(trans, root, bytenr, num_bytes,
3081                                            parent, ref_root, level - 1, 0,
3082                                            for_cow);
3083                         if (ret)
3084                                 goto fail;
3085                 }
3086         }
3087         return 0;
3088 fail:
3089         return ret;
3090 }
3091
3092 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3093                   struct extent_buffer *buf, int full_backref, int for_cow)
3094 {
3095         return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3096 }
3097
3098 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3099                   struct extent_buffer *buf, int full_backref, int for_cow)
3100 {
3101         return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3102 }
3103
3104 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3105                                  struct btrfs_root *root,
3106                                  struct btrfs_path *path,
3107                                  struct btrfs_block_group_cache *cache)
3108 {
3109         int ret;
3110         struct btrfs_root *extent_root = root->fs_info->extent_root;
3111         unsigned long bi;
3112         struct extent_buffer *leaf;
3113
3114         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3115         if (ret < 0)
3116                 goto fail;
3117         BUG_ON(ret); /* Corruption */
3118
3119         leaf = path->nodes[0];
3120         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3121         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3122         btrfs_mark_buffer_dirty(leaf);
3123         btrfs_release_path(path);
3124 fail:
3125         if (ret) {
3126                 btrfs_abort_transaction(trans, root, ret);
3127                 return ret;
3128         }
3129         return 0;
3130
3131 }
3132
3133 static struct btrfs_block_group_cache *
3134 next_block_group(struct btrfs_root *root,
3135                  struct btrfs_block_group_cache *cache)
3136 {
3137         struct rb_node *node;
3138         spin_lock(&root->fs_info->block_group_cache_lock);
3139         node = rb_next(&cache->cache_node);
3140         btrfs_put_block_group(cache);
3141         if (node) {
3142                 cache = rb_entry(node, struct btrfs_block_group_cache,
3143                                  cache_node);
3144                 btrfs_get_block_group(cache);
3145         } else
3146                 cache = NULL;
3147         spin_unlock(&root->fs_info->block_group_cache_lock);
3148         return cache;
3149 }
3150
3151 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3152                             struct btrfs_trans_handle *trans,
3153                             struct btrfs_path *path)
3154 {
3155         struct btrfs_root *root = block_group->fs_info->tree_root;
3156         struct inode *inode = NULL;
3157         u64 alloc_hint = 0;
3158         int dcs = BTRFS_DC_ERROR;
3159         int num_pages = 0;
3160         int retries = 0;
3161         int ret = 0;
3162
3163         /*
3164          * If this block group is smaller than 100 megs don't bother caching the
3165          * block group.
3166          */
3167         if (block_group->key.offset < (100 * 1024 * 1024)) {
3168                 spin_lock(&block_group->lock);
3169                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3170                 spin_unlock(&block_group->lock);
3171                 return 0;
3172         }
3173
3174 again:
3175         inode = lookup_free_space_inode(root, block_group, path);
3176         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3177                 ret = PTR_ERR(inode);
3178                 btrfs_release_path(path);
3179                 goto out;
3180         }
3181
3182         if (IS_ERR(inode)) {
3183                 BUG_ON(retries);
3184                 retries++;
3185
3186                 if (block_group->ro)
3187                         goto out_free;
3188
3189                 ret = create_free_space_inode(root, trans, block_group, path);
3190                 if (ret)
3191                         goto out_free;
3192                 goto again;
3193         }
3194
3195         /* We've already setup this transaction, go ahead and exit */
3196         if (block_group->cache_generation == trans->transid &&
3197             i_size_read(inode)) {
3198                 dcs = BTRFS_DC_SETUP;
3199                 goto out_put;
3200         }
3201
3202         /*
3203          * We want to set the generation to 0, that way if anything goes wrong
3204          * from here on out we know not to trust this cache when we load up next
3205          * time.
3206          */
3207         BTRFS_I(inode)->generation = 0;
3208         ret = btrfs_update_inode(trans, root, inode);
3209         WARN_ON(ret);
3210
3211         if (i_size_read(inode) > 0) {
3212                 ret = btrfs_check_trunc_cache_free_space(root,
3213                                         &root->fs_info->global_block_rsv);
3214                 if (ret)
3215                         goto out_put;
3216
3217                 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3218                 if (ret)
3219                         goto out_put;
3220         }
3221
3222         spin_lock(&block_group->lock);
3223         if (block_group->cached != BTRFS_CACHE_FINISHED ||
3224             !btrfs_test_opt(root, SPACE_CACHE)) {
3225                 /*
3226                  * don't bother trying to write stuff out _if_
3227                  * a) we're not cached,
3228                  * b) we're with nospace_cache mount option.
3229                  */
3230                 dcs = BTRFS_DC_WRITTEN;
3231                 spin_unlock(&block_group->lock);
3232                 goto out_put;
3233         }
3234         spin_unlock(&block_group->lock);
3235
3236         /*
3237          * Try to preallocate enough space based on how big the block group is.
3238          * Keep in mind this has to include any pinned space which could end up
3239          * taking up quite a bit since it's not folded into the other space
3240          * cache.
3241          */
3242         num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3243         if (!num_pages)
3244                 num_pages = 1;
3245
3246         num_pages *= 16;
3247         num_pages *= PAGE_CACHE_SIZE;
3248
3249         ret = btrfs_check_data_free_space(inode, num_pages);
3250         if (ret)
3251                 goto out_put;
3252
3253         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3254                                               num_pages, num_pages,
3255                                               &alloc_hint);
3256         if (!ret)
3257                 dcs = BTRFS_DC_SETUP;
3258         btrfs_free_reserved_data_space(inode, num_pages);
3259
3260 out_put:
3261         iput(inode);
3262 out_free:
3263         btrfs_release_path(path);
3264 out:
3265         spin_lock(&block_group->lock);
3266         if (!ret && dcs == BTRFS_DC_SETUP)
3267                 block_group->cache_generation = trans->transid;
3268         block_group->disk_cache_state = dcs;
3269         spin_unlock(&block_group->lock);
3270
3271         return ret;
3272 }
3273
3274 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3275                                    struct btrfs_root *root)
3276 {
3277         struct btrfs_block_group_cache *cache;
3278         int err = 0;
3279         struct btrfs_path *path;
3280         u64 last = 0;
3281
3282         path = btrfs_alloc_path();
3283         if (!path)
3284                 return -ENOMEM;
3285
3286 again:
3287         while (1) {
3288                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3289                 while (cache) {
3290                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3291                                 break;
3292                         cache = next_block_group(root, cache);
3293                 }
3294                 if (!cache) {
3295                         if (last == 0)
3296                                 break;
3297                         last = 0;
3298                         continue;
3299                 }
3300                 err = cache_save_setup(cache, trans, path);
3301                 last = cache->key.objectid + cache->key.offset;
3302                 btrfs_put_block_group(cache);
3303         }
3304
3305         while (1) {
3306                 if (last == 0) {
3307                         err = btrfs_run_delayed_refs(trans, root,
3308                                                      (unsigned long)-1);
3309                         if (err) /* File system offline */
3310                                 goto out;
3311                 }
3312
3313                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3314                 while (cache) {
3315                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3316                                 btrfs_put_block_group(cache);
3317                                 goto again;
3318                         }
3319
3320                         if (cache->dirty)
3321                                 break;
3322                         cache = next_block_group(root, cache);
3323                 }
3324                 if (!cache) {
3325                         if (last == 0)
3326                                 break;
3327                         last = 0;
3328                         continue;
3329                 }
3330
3331                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3332                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3333                 cache->dirty = 0;
3334                 last = cache->key.objectid + cache->key.offset;
3335
3336                 err = write_one_cache_group(trans, root, path, cache);
3337                 btrfs_put_block_group(cache);
3338                 if (err) /* File system offline */
3339                         goto out;
3340         }
3341
3342         while (1) {
3343                 /*
3344                  * I don't think this is needed since we're just marking our
3345                  * preallocated extent as written, but just in case it can't
3346                  * hurt.
3347                  */
3348                 if (last == 0) {
3349                         err = btrfs_run_delayed_refs(trans, root,
3350                                                      (unsigned long)-1);
3351                         if (err) /* File system offline */
3352                                 goto out;
3353                 }
3354
3355                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3356                 while (cache) {
3357                         /*
3358                          * Really this shouldn't happen, but it could if we
3359                          * couldn't write the entire preallocated extent and
3360                          * splitting the extent resulted in a new block.
3361                          */
3362                         if (cache->dirty) {
3363                                 btrfs_put_block_group(cache);
3364                                 goto again;
3365                         }
3366                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3367                                 break;
3368                         cache = next_block_group(root, cache);
3369                 }
3370                 if (!cache) {
3371                         if (last == 0)
3372                                 break;
3373                         last = 0;
3374                         continue;
3375                 }
3376
3377                 err = btrfs_write_out_cache(root, trans, cache, path);
3378
3379                 /*
3380                  * If we didn't have an error then the cache state is still
3381                  * NEED_WRITE, so we can set it to WRITTEN.
3382                  */
3383                 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3384                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
3385                 last = cache->key.objectid + cache->key.offset;
3386                 btrfs_put_block_group(cache);
3387         }
3388 out:
3389
3390         btrfs_free_path(path);
3391         return err;
3392 }
3393
3394 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3395 {
3396         struct btrfs_block_group_cache *block_group;
3397         int readonly = 0;
3398
3399         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3400         if (!block_group || block_group->ro)
3401                 readonly = 1;
3402         if (block_group)
3403                 btrfs_put_block_group(block_group);
3404         return readonly;
3405 }
3406
3407 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3408                              u64 total_bytes, u64 bytes_used,
3409                              struct btrfs_space_info **space_info)
3410 {
3411         struct btrfs_space_info *found;
3412         int i;
3413         int factor;
3414         int ret;
3415
3416         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3417                      BTRFS_BLOCK_GROUP_RAID10))
3418                 factor = 2;
3419         else
3420                 factor = 1;
3421
3422         found = __find_space_info(info, flags);
3423         if (found) {
3424                 spin_lock(&found->lock);
3425                 found->total_bytes += total_bytes;
3426                 found->disk_total += total_bytes * factor;
3427                 found->bytes_used += bytes_used;
3428                 found->disk_used += bytes_used * factor;
3429                 found->full = 0;
3430                 spin_unlock(&found->lock);
3431                 *space_info = found;
3432                 return 0;
3433         }
3434         found = kzalloc(sizeof(*found), GFP_NOFS);
3435         if (!found)
3436                 return -ENOMEM;
3437
3438         ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3439         if (ret) {
3440                 kfree(found);
3441                 return ret;
3442         }
3443
3444         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3445                 INIT_LIST_HEAD(&found->block_groups[i]);
3446         init_rwsem(&found->groups_sem);
3447         spin_lock_init(&found->lock);
3448         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3449         found->total_bytes = total_bytes;
3450         found->disk_total = total_bytes * factor;
3451         found->bytes_used = bytes_used;
3452         found->disk_used = bytes_used * factor;
3453         found->bytes_pinned = 0;
3454         found->bytes_reserved = 0;
3455         found->bytes_readonly = 0;
3456         found->bytes_may_use = 0;
3457         found->full = 0;
3458         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3459         found->chunk_alloc = 0;
3460         found->flush = 0;
3461         init_waitqueue_head(&found->wait);
3462         *space_info = found;
3463         list_add_rcu(&found->list, &info->space_info);
3464         if (flags & BTRFS_BLOCK_GROUP_DATA)
3465                 info->data_sinfo = found;
3466         return 0;
3467 }
3468
3469 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3470 {
3471         u64 extra_flags = chunk_to_extended(flags) &
3472                                 BTRFS_EXTENDED_PROFILE_MASK;
3473
3474         write_seqlock(&fs_info->profiles_lock);
3475         if (flags & BTRFS_BLOCK_GROUP_DATA)
3476                 fs_info->avail_data_alloc_bits |= extra_flags;
3477         if (flags & BTRFS_BLOCK_GROUP_METADATA)
3478                 fs_info->avail_metadata_alloc_bits |= extra_flags;
3479         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3480                 fs_info->avail_system_alloc_bits |= extra_flags;
3481         write_sequnlock(&fs_info->profiles_lock);
3482 }
3483
3484 /*
3485  * returns target flags in extended format or 0 if restripe for this
3486  * chunk_type is not in progress
3487  *
3488  * should be called with either volume_mutex or balance_lock held
3489  */
3490 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3491 {
3492         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3493         u64 target = 0;
3494
3495         if (!bctl)
3496                 return 0;
3497
3498         if (flags & BTRFS_BLOCK_GROUP_DATA &&
3499             bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3500                 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3501         } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3502                    bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3503                 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3504         } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3505                    bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3506                 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3507         }
3508
3509         return target;
3510 }
3511
3512 /*
3513  * @flags: available profiles in extended format (see ctree.h)
3514  *
3515  * Returns reduced profile in chunk format.  If profile changing is in
3516  * progress (either running or paused) picks the target profile (if it's
3517  * already available), otherwise falls back to plain reducing.
3518  */
3519 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3520 {
3521         /*
3522          * we add in the count of missing devices because we want
3523          * to make sure that any RAID levels on a degraded FS
3524          * continue to be honored.
3525          */
3526         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3527                 root->fs_info->fs_devices->missing_devices;
3528         u64 target;
3529         u64 tmp;
3530
3531         /*
3532          * see if restripe for this chunk_type is in progress, if so
3533          * try to reduce to the target profile
3534          */
3535         spin_lock(&root->fs_info->balance_lock);
3536         target = get_restripe_target(root->fs_info, flags);
3537         if (target) {
3538                 /* pick target profile only if it's already available */
3539                 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3540                         spin_unlock(&root->fs_info->balance_lock);
3541                         return extended_to_chunk(target);
3542                 }
3543         }
3544         spin_unlock(&root->fs_info->balance_lock);
3545
3546         /* First, mask out the RAID levels which aren't possible */
3547         if (num_devices == 1)
3548                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3549                            BTRFS_BLOCK_GROUP_RAID5);
3550         if (num_devices < 3)
3551                 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3552         if (num_devices < 4)
3553                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3554
3555         tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3556                        BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3557                        BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3558         flags &= ~tmp;
3559
3560         if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3561                 tmp = BTRFS_BLOCK_GROUP_RAID6;
3562         else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3563                 tmp = BTRFS_BLOCK_GROUP_RAID5;
3564         else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3565                 tmp = BTRFS_BLOCK_GROUP_RAID10;
3566         else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3567                 tmp = BTRFS_BLOCK_GROUP_RAID1;
3568         else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3569                 tmp = BTRFS_BLOCK_GROUP_RAID0;
3570
3571         return extended_to_chunk(flags | tmp);
3572 }
3573
3574 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3575 {
3576         unsigned seq;
3577
3578         do {
3579                 seq = read_seqbegin(&root->fs_info->profiles_lock);
3580
3581                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3582                         flags |= root->fs_info->avail_data_alloc_bits;
3583                 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3584                         flags |= root->fs_info->avail_system_alloc_bits;
3585                 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3586                         flags |= root->fs_info->avail_metadata_alloc_bits;
3587         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3588
3589         return btrfs_reduce_alloc_profile(root, flags);
3590 }
3591
3592 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3593 {
3594         u64 flags;
3595         u64 ret;
3596
3597         if (data)
3598                 flags = BTRFS_BLOCK_GROUP_DATA;
3599         else if (root == root->fs_info->chunk_root)
3600                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3601         else
3602                 flags = BTRFS_BLOCK_GROUP_METADATA;
3603
3604         ret = get_alloc_profile(root, flags);
3605         return ret;
3606 }
3607
3608 /*
3609  * This will check the space that the inode allocates from to make sure we have
3610  * enough space for bytes.
3611  */
3612 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3613 {
3614         struct btrfs_space_info *data_sinfo;
3615         struct btrfs_root *root = BTRFS_I(inode)->root;
3616         struct btrfs_fs_info *fs_info = root->fs_info;
3617         u64 used;
3618         int ret = 0, committed = 0, alloc_chunk = 1;
3619
3620         /* make sure bytes are sectorsize aligned */
3621         bytes = ALIGN(bytes, root->sectorsize);
3622
3623         if (btrfs_is_free_space_inode(inode)) {
3624                 committed = 1;
3625                 ASSERT(current->journal_info);
3626         }
3627
3628         data_sinfo = fs_info->data_sinfo;
3629         if (!data_sinfo)
3630                 goto alloc;
3631
3632 again:
3633         /* make sure we have enough space to handle the data first */
3634         spin_lock(&data_sinfo->lock);
3635         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3636                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3637                 data_sinfo->bytes_may_use;
3638
3639         if (used + bytes > data_sinfo->total_bytes) {
3640                 struct btrfs_trans_handle *trans;
3641
3642                 /*
3643                  * if we don't have enough free bytes in this space then we need
3644                  * to alloc a new chunk.
3645                  */
3646                 if (!data_sinfo->full && alloc_chunk) {
3647                         u64 alloc_target;
3648
3649                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3650                         spin_unlock(&data_sinfo->lock);
3651 alloc:
3652                         alloc_target = btrfs_get_alloc_profile(root, 1);
3653                         /*
3654                          * It is ugly that we don't call nolock join
3655                          * transaction for the free space inode case here.
3656                          * But it is safe because we only do the data space
3657                          * reservation for the free space cache in the
3658                          * transaction context, the common join transaction
3659                          * just increase the counter of the current transaction
3660                          * handler, doesn't try to acquire the trans_lock of
3661                          * the fs.
3662                          */
3663                         trans = btrfs_join_transaction(root);
3664                         if (IS_ERR(trans))
3665                                 return PTR_ERR(trans);
3666
3667                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3668                                              alloc_target,
3669                                              CHUNK_ALLOC_NO_FORCE);
3670                         btrfs_end_transaction(trans, root);
3671                         if (ret < 0) {
3672                                 if (ret != -ENOSPC)
3673                                         return ret;
3674                                 else
3675                                         goto commit_trans;
3676                         }
3677
3678                         if (!data_sinfo)
3679                                 data_sinfo = fs_info->data_sinfo;
3680
3681                         goto again;
3682                 }
3683
3684                 /*
3685                  * If we don't have enough pinned space to deal with this
3686                  * allocation don't bother committing the transaction.
3687                  */
3688                 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3689                                            bytes) < 0)
3690                         committed = 1;
3691                 spin_unlock(&data_sinfo->lock);
3692
3693                 /* commit the current transaction and try again */
3694 commit_trans:
3695                 if (!committed &&
3696                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3697                         committed = 1;
3698
3699                         trans = btrfs_join_transaction(root);
3700                         if (IS_ERR(trans))
3701                                 return PTR_ERR(trans);
3702                         ret = btrfs_commit_transaction(trans, root);
3703                         if (ret)
3704                                 return ret;
3705                         goto again;
3706                 }
3707
3708                 trace_btrfs_space_reservation(root->fs_info,
3709                                               "space_info:enospc",
3710                                               data_sinfo->flags, bytes, 1);
3711                 return -ENOSPC;
3712         }
3713         data_sinfo->bytes_may_use += bytes;
3714         trace_btrfs_space_reservation(root->fs_info, "space_info",
3715                                       data_sinfo->flags, bytes, 1);
3716         spin_unlock(&data_sinfo->lock);
3717
3718         return 0;
3719 }
3720
3721 /*
3722  * Called if we need to clear a data reservation for this inode.
3723  */
3724 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3725 {
3726         struct btrfs_root *root = BTRFS_I(inode)->root;
3727         struct btrfs_space_info *data_sinfo;
3728
3729         /* make sure bytes are sectorsize aligned */
3730         bytes = ALIGN(bytes, root->sectorsize);
3731
3732         data_sinfo = root->fs_info->data_sinfo;
3733         spin_lock(&data_sinfo->lock);
3734         WARN_ON(data_sinfo->bytes_may_use < bytes);
3735         data_sinfo->bytes_may_use -= bytes;
3736         trace_btrfs_space_reservation(root->fs_info, "space_info",
3737                                       data_sinfo->flags, bytes, 0);
3738         spin_unlock(&data_sinfo->lock);
3739 }
3740
3741 static void force_metadata_allocation(struct btrfs_fs_info *info)
3742 {
3743         struct list_head *head = &info->space_info;
3744         struct btrfs_space_info *found;
3745
3746         rcu_read_lock();
3747         list_for_each_entry_rcu(found, head, list) {
3748                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3749                         found->force_alloc = CHUNK_ALLOC_FORCE;
3750         }
3751         rcu_read_unlock();
3752 }
3753
3754 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3755 {
3756         return (global->size << 1);
3757 }
3758
3759 static int should_alloc_chunk(struct btrfs_root *root,
3760                               struct btrfs_space_info *sinfo, int force)
3761 {
3762         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3763         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3764         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3765         u64 thresh;
3766
3767         if (force == CHUNK_ALLOC_FORCE)
3768                 return 1;
3769
3770         /*
3771          * We need to take into account the global rsv because for all intents
3772          * and purposes it's used space.  Don't worry about locking the
3773          * global_rsv, it doesn't change except when the transaction commits.
3774          */
3775         if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3776                 num_allocated += calc_global_rsv_need_space(global_rsv);
3777
3778         /*
3779          * in limited mode, we want to have some free space up to
3780          * about 1% of the FS size.
3781          */
3782         if (force == CHUNK_ALLOC_LIMITED) {
3783                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3784                 thresh = max_t(u64, 64 * 1024 * 1024,
3785                                div_factor_fine(thresh, 1));
3786
3787                 if (num_bytes - num_allocated < thresh)
3788                         return 1;
3789         }
3790
3791         if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3792                 return 0;
3793         return 1;
3794 }
3795
3796 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3797 {
3798         u64 num_dev;
3799
3800         if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3801                     BTRFS_BLOCK_GROUP_RAID0 |
3802                     BTRFS_BLOCK_GROUP_RAID5 |
3803                     BTRFS_BLOCK_GROUP_RAID6))
3804                 num_dev = root->fs_info->fs_devices->rw_devices;
3805         else if (type & BTRFS_BLOCK_GROUP_RAID1)
3806                 num_dev = 2;
3807         else
3808                 num_dev = 1;    /* DUP or single */
3809
3810         /* metadata for updaing devices and chunk tree */
3811         return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3812 }
3813
3814 static void check_system_chunk(struct btrfs_trans_handle *trans,
3815                                struct btrfs_root *root, u64 type)
3816 {
3817         struct btrfs_space_info *info;
3818         u64 left;
3819         u64 thresh;
3820
3821         info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3822         spin_lock(&info->lock);
3823         left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3824                 info->bytes_reserved - info->bytes_readonly;
3825         spin_unlock(&info->lock);
3826
3827         thresh = get_system_chunk_thresh(root, type);
3828         if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3829                 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3830                         left, thresh, type);
3831                 dump_space_info(info, 0, 0);
3832         }
3833
3834         if (left < thresh) {
3835                 u64 flags;
3836
3837                 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3838                 btrfs_alloc_chunk(trans, root, flags);
3839         }
3840 }
3841
3842 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3843                           struct btrfs_root *extent_root, u64 flags, int force)
3844 {
3845         struct btrfs_space_info *space_info;
3846         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3847         int wait_for_alloc = 0;
3848         int ret = 0;
3849
3850         /* Don't re-enter if we're already allocating a chunk */
3851         if (trans->allocating_chunk)
3852                 return -ENOSPC;
3853
3854         space_info = __find_space_info(extent_root->fs_info, flags);
3855         if (!space_info) {
3856                 ret = update_space_info(extent_root->fs_info, flags,
3857                                         0, 0, &space_info);
3858                 BUG_ON(ret); /* -ENOMEM */
3859         }
3860         BUG_ON(!space_info); /* Logic error */
3861
3862 again:
3863         spin_lock(&space_info->lock);
3864         if (force < space_info->force_alloc)
3865                 force = space_info->force_alloc;
3866         if (space_info->full) {
3867                 if (should_alloc_chunk(extent_root, space_info, force))
3868                         ret = -ENOSPC;
3869                 else
3870                         ret = 0;
3871                 spin_unlock(&space_info->lock);
3872                 return ret;
3873         }
3874
3875         if (!should_alloc_chunk(extent_root, space_info, force)) {
3876                 spin_unlock(&space_info->lock);
3877                 return 0;
3878         } else if (space_info->chunk_alloc) {
3879                 wait_for_alloc = 1;
3880         } else {
3881                 space_info->chunk_alloc = 1;
3882         }
3883
3884         spin_unlock(&space_info->lock);
3885
3886         mutex_lock(&fs_info->chunk_mutex);
3887
3888         /*
3889          * The chunk_mutex is held throughout the entirety of a chunk
3890          * allocation, so once we've acquired the chunk_mutex we know that the
3891          * other guy is done and we need to recheck and see if we should
3892          * allocate.
3893          */
3894         if (wait_for_alloc) {
3895                 mutex_unlock(&fs_info->chunk_mutex);
3896                 wait_for_alloc = 0;
3897                 goto again;
3898         }
3899
3900         trans->allocating_chunk = true;
3901
3902         /*
3903          * If we have mixed data/metadata chunks we want to make sure we keep
3904          * allocating mixed chunks instead of individual chunks.
3905          */
3906         if (btrfs_mixed_space_info(space_info))
3907                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3908
3909         /*
3910          * if we're doing a data chunk, go ahead and make sure that
3911          * we keep a reasonable number of metadata chunks allocated in the
3912          * FS as well.
3913          */
3914         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3915                 fs_info->data_chunk_allocations++;
3916                 if (!(fs_info->data_chunk_allocations %
3917                       fs_info->metadata_ratio))
3918                         force_metadata_allocation(fs_info);
3919         }
3920
3921         /*
3922          * Check if we have enough space in SYSTEM chunk because we may need
3923          * to update devices.
3924          */
3925         check_system_chunk(trans, extent_root, flags);
3926
3927         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3928         trans->allocating_chunk = false;
3929
3930         spin_lock(&space_info->lock);
3931         if (ret < 0 && ret != -ENOSPC)
3932                 goto out;
3933         if (ret)
3934                 space_info->full = 1;
3935         else
3936                 ret = 1;
3937
3938         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3939 out:
3940         space_info->chunk_alloc = 0;
3941         spin_unlock(&space_info->lock);
3942         mutex_unlock(&fs_info->chunk_mutex);
3943         return ret;
3944 }
3945
3946 static int can_overcommit(struct btrfs_root *root,
3947                           struct btrfs_space_info *space_info, u64 bytes,
3948                           enum btrfs_reserve_flush_enum flush)
3949 {
3950         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3951         u64 profile = btrfs_get_alloc_profile(root, 0);
3952         u64 space_size;
3953         u64 avail;
3954         u64 used;
3955
3956         used = space_info->bytes_used + space_info->bytes_reserved +
3957                 space_info->bytes_pinned + space_info->bytes_readonly;
3958
3959         /*
3960          * We only want to allow over committing if we have lots of actual space
3961          * free, but if we don't have enough space to handle the global reserve
3962          * space then we could end up having a real enospc problem when trying
3963          * to allocate a chunk or some other such important allocation.
3964          */
3965         spin_lock(&global_rsv->lock);
3966         space_size = calc_global_rsv_need_space(global_rsv);
3967         spin_unlock(&global_rsv->lock);
3968         if (used + space_size >= space_info->total_bytes)
3969                 return 0;
3970
3971         used += space_info->bytes_may_use;
3972
3973         spin_lock(&root->fs_info->free_chunk_lock);
3974         avail = root->fs_info->free_chunk_space;
3975         spin_unlock(&root->fs_info->free_chunk_lock);
3976
3977         /*
3978          * If we have dup, raid1 or raid10 then only half of the free
3979          * space is actually useable.  For raid56, the space info used
3980          * doesn't include the parity drive, so we don't have to
3981          * change the math
3982          */
3983         if (profile & (BTRFS_BLOCK_GROUP_DUP |
3984                        BTRFS_BLOCK_GROUP_RAID1 |
3985                        BTRFS_BLOCK_GROUP_RAID10))
3986                 avail >>= 1;
3987
3988         /*
3989          * If we aren't flushing all things, let us overcommit up to
3990          * 1/2th of the space. If we can flush, don't let us overcommit
3991          * too much, let it overcommit up to 1/8 of the space.
3992          */
3993         if (flush == BTRFS_RESERVE_FLUSH_ALL)
3994                 avail >>= 3;
3995         else
3996                 avail >>= 1;
3997
3998         if (used + bytes < space_info->total_bytes + avail)
3999                 return 1;
4000         return 0;
4001 }
4002
4003 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4004                                          unsigned long nr_pages)
4005 {
4006         struct super_block *sb = root->fs_info->sb;
4007
4008         if (down_read_trylock(&sb->s_umount)) {
4009                 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4010                 up_read(&sb->s_umount);
4011         } else {
4012                 /*
4013                  * We needn't worry the filesystem going from r/w to r/o though
4014                  * we don't acquire ->s_umount mutex, because the filesystem
4015                  * should guarantee the delalloc inodes list be empty after
4016                  * the filesystem is readonly(all dirty pages are written to
4017                  * the disk).
4018                  */
4019                 btrfs_start_all_delalloc_inodes(root->fs_info, 0);
4020                 if (!current->journal_info)
4021                         btrfs_wait_all_ordered_extents(root->fs_info);
4022         }
4023 }
4024
4025 /*
4026  * shrink metadata reservation for delalloc
4027  */
4028 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4029                             bool wait_ordered)
4030 {
4031         struct btrfs_block_rsv *block_rsv;
4032         struct btrfs_space_info *space_info;
4033         struct btrfs_trans_handle *trans;
4034         u64 delalloc_bytes;
4035         u64 max_reclaim;
4036         long time_left;
4037         unsigned long nr_pages;
4038         int loops;
4039         enum btrfs_reserve_flush_enum flush;
4040
4041         trans = (struct btrfs_trans_handle *)current->journal_info;
4042         block_rsv = &root->fs_info->delalloc_block_rsv;
4043         space_info = block_rsv->space_info;
4044
4045         delalloc_bytes = percpu_counter_sum_positive(
4046                                                 &root->fs_info->delalloc_bytes);
4047         if (delalloc_bytes == 0) {
4048                 if (trans)
4049                         return;
4050                 if (wait_ordered)
4051                         btrfs_wait_all_ordered_extents(root->fs_info);
4052                 return;
4053         }
4054
4055         loops = 0;
4056         while (delalloc_bytes && loops < 3) {
4057                 max_reclaim = min(delalloc_bytes, to_reclaim);
4058                 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4059                 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4060                 /*
4061                  * We need to wait for the async pages to actually start before
4062                  * we do anything.
4063                  */
4064                 wait_event(root->fs_info->async_submit_wait,
4065                            !atomic_read(&root->fs_info->async_delalloc_pages));
4066
4067                 if (!trans)
4068                         flush = BTRFS_RESERVE_FLUSH_ALL;
4069                 else
4070                         flush = BTRFS_RESERVE_NO_FLUSH;
4071                 spin_lock(&space_info->lock);
4072                 if (can_overcommit(root, space_info, orig, flush)) {
4073                         spin_unlock(&space_info->lock);
4074                         break;
4075                 }
4076                 spin_unlock(&space_info->lock);
4077
4078                 loops++;
4079                 if (wait_ordered && !trans) {
4080                         btrfs_wait_all_ordered_extents(root->fs_info);
4081                 } else {
4082                         time_left = schedule_timeout_killable(1);
4083                         if (time_left)
4084                                 break;
4085                 }
4086                 delalloc_bytes = percpu_counter_sum_positive(
4087                                                 &root->fs_info->delalloc_bytes);
4088         }
4089 }
4090
4091 /**
4092  * maybe_commit_transaction - possibly commit the transaction if its ok to
4093  * @root - the root we're allocating for
4094  * @bytes - the number of bytes we want to reserve
4095  * @force - force the commit
4096  *
4097  * This will check to make sure that committing the transaction will actually
4098  * get us somewhere and then commit the transaction if it does.  Otherwise it
4099  * will return -ENOSPC.
4100  */
4101 static int may_commit_transaction(struct btrfs_root *root,
4102                                   struct btrfs_space_info *space_info,
4103                                   u64 bytes, int force)
4104 {
4105         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4106         struct btrfs_trans_handle *trans;
4107
4108         trans = (struct btrfs_trans_handle *)current->journal_info;
4109         if (trans)
4110                 return -EAGAIN;
4111
4112         if (force)
4113                 goto commit;
4114
4115         /* See if there is enough pinned space to make this reservation */
4116         spin_lock(&space_info->lock);
4117         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4118                                    bytes) >= 0) {
4119                 spin_unlock(&space_info->lock);
4120                 goto commit;
4121         }
4122         spin_unlock(&space_info->lock);
4123
4124         /*
4125          * See if there is some space in the delayed insertion reservation for
4126          * this reservation.
4127          */
4128         if (space_info != delayed_rsv->space_info)
4129                 return -ENOSPC;
4130
4131         spin_lock(&space_info->lock);
4132         spin_lock(&delayed_rsv->lock);
4133         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4134                                    bytes - delayed_rsv->size) >= 0) {
4135                 spin_unlock(&delayed_rsv->lock);
4136                 spin_unlock(&space_info->lock);
4137                 return -ENOSPC;
4138         }
4139         spin_unlock(&delayed_rsv->lock);
4140         spin_unlock(&space_info->lock);
4141
4142 commit:
4143         trans = btrfs_join_transaction(root);
4144         if (IS_ERR(trans))
4145                 return -ENOSPC;
4146
4147         return btrfs_commit_transaction(trans, root);
4148 }
4149
4150 enum flush_state {
4151         FLUSH_DELAYED_ITEMS_NR  =       1,
4152         FLUSH_DELAYED_ITEMS     =       2,
4153         FLUSH_DELALLOC          =       3,
4154         FLUSH_DELALLOC_WAIT     =       4,
4155         ALLOC_CHUNK             =       5,
4156         COMMIT_TRANS            =       6,
4157 };
4158
4159 static int flush_space(struct btrfs_root *root,
4160                        struct btrfs_space_info *space_info, u64 num_bytes,
4161                        u64 orig_bytes, int state)
4162 {
4163         struct btrfs_trans_handle *trans;
4164         int nr;
4165         int ret = 0;
4166
4167         switch (state) {
4168         case FLUSH_DELAYED_ITEMS_NR:
4169         case FLUSH_DELAYED_ITEMS:
4170                 if (state == FLUSH_DELAYED_ITEMS_NR) {
4171                         u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4172
4173                         nr = (int)div64_u64(num_bytes, bytes);
4174                         if (!nr)
4175                                 nr = 1;
4176                         nr *= 2;
4177                 } else {
4178                         nr = -1;
4179                 }
4180                 trans = btrfs_join_transaction(root);
4181                 if (IS_ERR(trans)) {
4182                         ret = PTR_ERR(trans);
4183                         break;
4184                 }
4185                 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4186                 btrfs_end_transaction(trans, root);
4187                 break;
4188         case FLUSH_DELALLOC:
4189         case FLUSH_DELALLOC_WAIT:
4190                 shrink_delalloc(root, num_bytes, orig_bytes,
4191                                 state == FLUSH_DELALLOC_WAIT);
4192                 break;
4193         case ALLOC_CHUNK:
4194                 trans = btrfs_join_transaction(root);
4195                 if (IS_ERR(trans)) {
4196                         ret = PTR_ERR(trans);
4197                         break;
4198                 }
4199                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4200                                      btrfs_get_alloc_profile(root, 0),
4201                                      CHUNK_ALLOC_NO_FORCE);
4202                 btrfs_end_transaction(trans, root);
4203                 if (ret == -ENOSPC)
4204                         ret = 0;
4205                 break;
4206         case COMMIT_TRANS:
4207                 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4208                 break;
4209         default:
4210                 ret = -ENOSPC;
4211                 break;
4212         }
4213
4214         return ret;
4215 }
4216 /**
4217  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4218  * @root - the root we're allocating for
4219  * @block_rsv - the block_rsv we're allocating for
4220  * @orig_bytes - the number of bytes we want
4221  * @flush - whether or not we can flush to make our reservation
4222  *
4223  * This will reserve orgi_bytes number of bytes from the space info associated
4224  * with the block_rsv.  If there is not enough space it will make an attempt to
4225  * flush out space to make room.  It will do this by flushing delalloc if
4226  * possible or committing the transaction.  If flush is 0 then no attempts to
4227  * regain reservations will be made and this will fail if there is not enough
4228  * space already.
4229  */
4230 static int reserve_metadata_bytes(struct btrfs_root *root,
4231                                   struct btrfs_block_rsv *block_rsv,
4232                                   u64 orig_bytes,
4233                                   enum btrfs_reserve_flush_enum flush)
4234 {
4235         struct btrfs_space_info *space_info = block_rsv->space_info;
4236         u64 used;
4237         u64 num_bytes = orig_bytes;
4238         int flush_state = FLUSH_DELAYED_ITEMS_NR;
4239         int ret = 0;
4240         bool flushing = false;
4241
4242 again:
4243         ret = 0;
4244         spin_lock(&space_info->lock);
4245         /*
4246          * We only want to wait if somebody other than us is flushing and we
4247          * are actually allowed to flush all things.
4248          */
4249         while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4250                space_info->flush) {
4251                 spin_unlock(&space_info->lock);
4252                 /*
4253                  * If we have a trans handle we can't wait because the flusher
4254                  * may have to commit the transaction, which would mean we would
4255                  * deadlock since we are waiting for the flusher to finish, but
4256                  * hold the current transaction open.
4257                  */
4258                 if (current->journal_info)
4259                         return -EAGAIN;
4260                 ret = wait_event_killable(space_info->wait, !space_info->flush);
4261                 /* Must have been killed, return */
4262                 if (ret)
4263                         return -EINTR;
4264
4265                 spin_lock(&space_info->lock);
4266         }
4267
4268         ret = -ENOSPC;
4269         used = space_info->bytes_used + space_info->bytes_reserved +
4270                 space_info->bytes_pinned + space_info->bytes_readonly +
4271                 space_info->bytes_may_use;
4272
4273         /*
4274          * The idea here is that we've not already over-reserved the block group
4275          * then we can go ahead and save our reservation first and then start
4276          * flushing if we need to.  Otherwise if we've already overcommitted
4277          * lets start flushing stuff first and then come back and try to make
4278          * our reservation.
4279          */
4280         if (used <= space_info->total_bytes) {
4281                 if (used + orig_bytes <= space_info->total_bytes) {
4282                         space_info->bytes_may_use += orig_bytes;
4283                         trace_btrfs_space_reservation(root->fs_info,
4284                                 "space_info", space_info->flags, orig_bytes, 1);
4285                         ret = 0;
4286                 } else {
4287                         /*
4288                          * Ok set num_bytes to orig_bytes since we aren't
4289                          * overocmmitted, this way we only try and reclaim what
4290                          * we need.
4291                          */
4292                         num_bytes = orig_bytes;
4293                 }
4294         } else {
4295                 /*
4296                  * Ok we're over committed, set num_bytes to the overcommitted
4297                  * amount plus the amount of bytes that we need for this
4298                  * reservation.
4299                  */
4300                 num_bytes = used - space_info->total_bytes +
4301                         (orig_bytes * 2);
4302         }
4303
4304         if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4305                 space_info->bytes_may_use += orig_bytes;
4306                 trace_btrfs_space_reservation(root->fs_info, "space_info",
4307                                               space_info->flags, orig_bytes,
4308                                               1);
4309                 ret = 0;
4310         }
4311
4312         /*
4313          * Couldn't make our reservation, save our place so while we're trying
4314          * to reclaim space we can actually use it instead of somebody else
4315          * stealing it from us.
4316          *
4317          * We make the other tasks wait for the flush only when we can flush
4318          * all things.
4319          */
4320         if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4321                 flushing = true;
4322                 space_info->flush = 1;
4323         }
4324
4325         spin_unlock(&space_info->lock);
4326
4327         if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4328                 goto out;
4329
4330         ret = flush_space(root, space_info, num_bytes, orig_bytes,
4331                           flush_state);
4332         flush_state++;
4333
4334         /*
4335          * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4336          * would happen. So skip delalloc flush.
4337          */
4338         if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4339             (flush_state == FLUSH_DELALLOC ||
4340              flush_state == FLUSH_DELALLOC_WAIT))
4341                 flush_state = ALLOC_CHUNK;
4342
4343         if (!ret)
4344                 goto again;
4345         else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4346                  flush_state < COMMIT_TRANS)
4347                 goto again;
4348         else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4349                  flush_state <= COMMIT_TRANS)
4350                 goto again;
4351
4352 out:
4353         if (ret == -ENOSPC &&
4354             unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4355                 struct btrfs_block_rsv *global_rsv =
4356                         &root->fs_info->global_block_rsv;
4357
4358                 if (block_rsv != global_rsv &&
4359                     !block_rsv_use_bytes(global_rsv, orig_bytes))
4360                         ret = 0;
4361         }
4362         if (ret == -ENOSPC)
4363                 trace_btrfs_space_reservation(root->fs_info,
4364                                               "space_info:enospc",
4365                                               space_info->flags, orig_bytes, 1);
4366         if (flushing) {
4367                 spin_lock(&space_info->lock);
4368                 space_info->flush = 0;
4369                 wake_up_all(&space_info->wait);
4370                 spin_unlock(&space_info->lock);
4371         }
4372         return ret;
4373 }
4374
4375 static struct btrfs_block_rsv *get_block_rsv(
4376                                         const struct btrfs_trans_handle *trans,
4377                                         const struct btrfs_root *root)
4378 {
4379         struct btrfs_block_rsv *block_rsv = NULL;
4380
4381         if (root->ref_cows)
4382                 block_rsv = trans->block_rsv;
4383
4384         if (root == root->fs_info->csum_root && trans->adding_csums)
4385                 block_rsv = trans->block_rsv;
4386
4387         if (root == root->fs_info->uuid_root)
4388                 block_rsv = trans->block_rsv;
4389
4390         if (!block_rsv)
4391                 block_rsv = root->block_rsv;
4392
4393         if (!block_rsv)
4394                 block_rsv = &root->fs_info->empty_block_rsv;
4395
4396         return block_rsv;
4397 }
4398
4399 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4400                                u64 num_bytes)
4401 {
4402         int ret = -ENOSPC;
4403         spin_lock(&block_rsv->lock);
4404         if (block_rsv->reserved >= num_bytes) {
4405                 block_rsv->reserved -= num_bytes;
4406                 if (block_rsv->reserved < block_rsv->size)
4407                         block_rsv->full = 0;
4408                 ret = 0;
4409         }
4410         spin_unlock(&block_rsv->lock);
4411         return ret;
4412 }
4413
4414 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4415                                 u64 num_bytes, int update_size)
4416 {
4417         spin_lock(&block_rsv->lock);
4418         block_rsv->reserved += num_bytes;
4419         if (update_size)
4420                 block_rsv->size += num_bytes;
4421         else if (block_rsv->reserved >= block_rsv->size)
4422                 block_rsv->full = 1;
4423         spin_unlock(&block_rsv->lock);
4424 }
4425
4426 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4427                              struct btrfs_block_rsv *dest, u64 num_bytes,
4428                              int min_factor)
4429 {
4430         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4431         u64 min_bytes;
4432
4433         if (global_rsv->space_info != dest->space_info)
4434                 return -ENOSPC;
4435
4436         spin_lock(&global_rsv->lock);
4437         min_bytes = div_factor(global_rsv->size, min_factor);
4438         if (global_rsv->reserved < min_bytes + num_bytes) {
4439                 spin_unlock(&global_rsv->lock);
4440                 return -ENOSPC;
4441         }
4442         global_rsv->reserved -= num_bytes;
4443         if (global_rsv->reserved < global_rsv->size)
4444                 global_rsv->full = 0;
4445         spin_unlock(&global_rsv->lock);
4446
4447         block_rsv_add_bytes(dest, num_bytes, 1);
4448         return 0;
4449 }
4450
4451 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4452                                     struct btrfs_block_rsv *block_rsv,
4453                                     struct btrfs_block_rsv *dest, u64 num_bytes)
4454 {
4455         struct btrfs_space_info *space_info = block_rsv->space_info;
4456
4457         spin_lock(&block_rsv->lock);
4458         if (num_bytes == (u64)-1)
4459                 num_bytes = block_rsv->size;
4460         block_rsv->size -= num_bytes;
4461         if (block_rsv->reserved >= block_rsv->size) {
4462                 num_bytes = block_rsv->reserved - block_rsv->size;
4463                 block_rsv->reserved = block_rsv->size;
4464                 block_rsv->full = 1;
4465         } else {
4466                 num_bytes = 0;
4467         }
4468         spin_unlock(&block_rsv->lock);
4469
4470         if (num_bytes > 0) {
4471                 if (dest) {
4472                         spin_lock(&dest->lock);
4473                         if (!dest->full) {
4474                                 u64 bytes_to_add;
4475
4476                                 bytes_to_add = dest->size - dest->reserved;
4477                                 bytes_to_add = min(num_bytes, bytes_to_add);
4478                                 dest->reserved += bytes_to_add;
4479                                 if (dest->reserved >= dest->size)
4480                                         dest->full = 1;
4481                                 num_bytes -= bytes_to_add;
4482                         }
4483                         spin_unlock(&dest->lock);
4484                 }
4485                 if (num_bytes) {
4486                         spin_lock(&space_info->lock);
4487                         space_info->bytes_may_use -= num_bytes;
4488                         trace_btrfs_space_reservation(fs_info, "space_info",
4489                                         space_info->flags, num_bytes, 0);
4490                         spin_unlock(&space_info->lock);
4491                 }
4492         }
4493 }
4494
4495 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4496                                    struct btrfs_block_rsv *dst, u64 num_bytes)
4497 {
4498         int ret;
4499
4500         ret = block_rsv_use_bytes(src, num_bytes);
4501         if (ret)
4502                 return ret;
4503
4504         block_rsv_add_bytes(dst, num_bytes, 1);
4505         return 0;
4506 }
4507
4508 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4509 {
4510         memset(rsv, 0, sizeof(*rsv));
4511         spin_lock_init(&rsv->lock);
4512         rsv->type = type;
4513 }
4514
4515 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4516                                               unsigned short type)
4517 {
4518         struct btrfs_block_rsv *block_rsv;
4519         struct btrfs_fs_info *fs_info = root->fs_info;
4520
4521         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4522         if (!block_rsv)
4523                 return NULL;
4524
4525         btrfs_init_block_rsv(block_rsv, type);
4526         block_rsv->space_info = __find_space_info(fs_info,
4527                                                   BTRFS_BLOCK_GROUP_METADATA);
4528         return block_rsv;
4529 }
4530
4531 void btrfs_free_block_rsv(struct btrfs_root *root,
4532                           struct btrfs_block_rsv *rsv)
4533 {
4534         if (!rsv)
4535                 return;
4536         btrfs_block_rsv_release(root, rsv, (u64)-1);
4537         kfree(rsv);
4538 }
4539
4540 int btrfs_block_rsv_add(struct btrfs_root *root,
4541                         struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4542                         enum btrfs_reserve_flush_enum flush)
4543 {
4544         int ret;
4545
4546         if (num_bytes == 0)
4547                 return 0;
4548
4549         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4550         if (!ret) {
4551                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4552                 return 0;
4553         }
4554
4555         return ret;
4556 }
4557
4558 int btrfs_block_rsv_check(struct btrfs_root *root,
4559                           struct btrfs_block_rsv *block_rsv, int min_factor)
4560 {
4561         u64 num_bytes = 0;
4562         int ret = -ENOSPC;
4563
4564         if (!block_rsv)
4565                 return 0;
4566
4567         spin_lock(&block_rsv->lock);
4568         num_bytes = div_factor(block_rsv->size, min_factor);
4569         if (block_rsv->reserved >= num_bytes)
4570                 ret = 0;
4571         spin_unlock(&block_rsv->lock);
4572
4573         return ret;
4574 }
4575
4576 int btrfs_block_rsv_refill(struct btrfs_root *root,
4577                            struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4578                            enum btrfs_reserve_flush_enum flush)
4579 {
4580         u64 num_bytes = 0;
4581         int ret = -ENOSPC;
4582
4583         if (!block_rsv)
4584                 return 0;
4585
4586         spin_lock(&block_rsv->lock);
4587         num_bytes = min_reserved;
4588         if (block_rsv->reserved >= num_bytes)
4589                 ret = 0;
4590         else
4591                 num_bytes -= block_rsv->reserved;
4592         spin_unlock(&block_rsv->lock);
4593
4594         if (!ret)
4595                 return 0;
4596
4597         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4598         if (!ret) {
4599                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4600                 return 0;
4601         }
4602
4603         return ret;
4604 }
4605
4606 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4607                             struct btrfs_block_rsv *dst_rsv,
4608                             u64 num_bytes)
4609 {
4610         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4611 }
4612
4613 void btrfs_block_rsv_release(struct btrfs_root *root,
4614                              struct btrfs_block_rsv *block_rsv,
4615                              u64 num_bytes)
4616 {
4617         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4618         if (global_rsv->full || global_rsv == block_rsv ||
4619             block_rsv->space_info != global_rsv->space_info)
4620                 global_rsv = NULL;
4621         block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4622                                 num_bytes);
4623 }
4624
4625 /*
4626  * helper to calculate size of global block reservation.
4627  * the desired value is sum of space used by extent tree,
4628  * checksum tree and root tree
4629  */
4630 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4631 {
4632         struct btrfs_space_info *sinfo;
4633         u64 num_bytes;
4634         u64 meta_used;
4635         u64 data_used;
4636         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4637
4638         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4639         spin_lock(&sinfo->lock);
4640         data_used = sinfo->bytes_used;
4641         spin_unlock(&sinfo->lock);
4642
4643         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4644         spin_lock(&sinfo->lock);
4645         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4646                 data_used = 0;
4647         meta_used = sinfo->bytes_used;
4648         spin_unlock(&sinfo->lock);
4649
4650         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4651                     csum_size * 2;
4652         num_bytes += div64_u64(data_used + meta_used, 50);
4653
4654         if (num_bytes * 3 > meta_used)
4655                 num_bytes = div64_u64(meta_used, 3);
4656
4657         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4658 }
4659
4660 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4661 {
4662         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4663         struct btrfs_space_info *sinfo = block_rsv->space_info;
4664         u64 num_bytes;
4665
4666         num_bytes = calc_global_metadata_size(fs_info);
4667
4668         spin_lock(&sinfo->lock);
4669         spin_lock(&block_rsv->lock);
4670
4671         block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4672
4673         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4674                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4675                     sinfo->bytes_may_use;
4676
4677         if (sinfo->total_bytes > num_bytes) {
4678                 num_bytes = sinfo->total_bytes - num_bytes;
4679                 block_rsv->reserved += num_bytes;
4680                 sinfo->bytes_may_use += num_bytes;
4681                 trace_btrfs_space_reservation(fs_info, "space_info",
4682                                       sinfo->flags, num_bytes, 1);
4683         }
4684
4685         if (block_rsv->reserved >= block_rsv->size) {
4686                 num_bytes = block_rsv->reserved - block_rsv->size;
4687                 sinfo->bytes_may_use -= num_bytes;
4688                 trace_btrfs_space_reservation(fs_info, "space_info",
4689                                       sinfo->flags, num_bytes, 0);
4690                 block_rsv->reserved = block_rsv->size;
4691                 block_rsv->full = 1;
4692         }
4693
4694         spin_unlock(&block_rsv->lock);
4695         spin_unlock(&sinfo->lock);
4696 }
4697
4698 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4699 {
4700         struct btrfs_space_info *space_info;
4701
4702         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4703         fs_info->chunk_block_rsv.space_info = space_info;
4704
4705         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4706         fs_info->global_block_rsv.space_info = space_info;
4707         fs_info->delalloc_block_rsv.space_info = space_info;
4708         fs_info->trans_block_rsv.space_info = space_info;
4709         fs_info->empty_block_rsv.space_info = space_info;
4710         fs_info->delayed_block_rsv.space_info = space_info;
4711
4712         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4713         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4714         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4715         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4716         if (fs_info->quota_root)
4717                 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4718         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4719
4720         update_global_block_rsv(fs_info);
4721 }
4722
4723 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4724 {
4725         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4726                                 (u64)-1);
4727         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4728         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4729         WARN_ON(fs_info->trans_block_rsv.size > 0);
4730         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4731         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4732         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4733         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4734         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4735 }
4736
4737 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4738                                   struct btrfs_root *root)
4739 {
4740         if (!trans->block_rsv)
4741                 return;
4742
4743         if (!trans->bytes_reserved)
4744                 return;
4745
4746         trace_btrfs_space_reservation(root->fs_info, "transaction",
4747                                       trans->transid, trans->bytes_reserved, 0);
4748         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4749         trans->bytes_reserved = 0;
4750 }
4751
4752 /* Can only return 0 or -ENOSPC */
4753 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4754                                   struct inode *inode)
4755 {
4756         struct btrfs_root *root = BTRFS_I(inode)->root;
4757         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4758         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4759
4760         /*
4761          * We need to hold space in order to delete our orphan item once we've
4762          * added it, so this takes the reservation so we can release it later
4763          * when we are truly done with the orphan item.
4764          */
4765         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4766         trace_btrfs_space_reservation(root->fs_info, "orphan",
4767                                       btrfs_ino(inode), num_bytes, 1);
4768         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4769 }
4770
4771 void btrfs_orphan_release_metadata(struct inode *inode)
4772 {
4773         struct btrfs_root *root = BTRFS_I(inode)->root;
4774         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4775         trace_btrfs_space_reservation(root->fs_info, "orphan",
4776                                       btrfs_ino(inode), num_bytes, 0);
4777         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4778 }
4779
4780 /*
4781  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4782  * root: the root of the parent directory
4783  * rsv: block reservation
4784  * items: the number of items that we need do reservation
4785  * qgroup_reserved: used to return the reserved size in qgroup
4786  *
4787  * This function is used to reserve the space for snapshot/subvolume
4788  * creation and deletion. Those operations are different with the
4789  * common file/directory operations, they change two fs/file trees
4790  * and root tree, the number of items that the qgroup reserves is
4791  * different with the free space reservation. So we can not use
4792  * the space reseravtion mechanism in start_transaction().
4793  */
4794 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4795                                      struct btrfs_block_rsv *rsv,
4796                                      int items,
4797                                      u64 *qgroup_reserved,
4798                                      bool use_global_rsv)
4799 {
4800         u64 num_bytes;
4801         int ret;
4802         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4803
4804         if (root->fs_info->quota_enabled) {
4805                 /* One for parent inode, two for dir entries */
4806                 num_bytes = 3 * root->leafsize;
4807                 ret = btrfs_qgroup_reserve(root, num_bytes);
4808                 if (ret)
4809                         return ret;
4810         } else {
4811                 num_bytes = 0;
4812         }
4813
4814         *qgroup_reserved = num_bytes;
4815
4816         num_bytes = btrfs_calc_trans_metadata_size(root, items);
4817         rsv->space_info = __find_space_info(root->fs_info,
4818                                             BTRFS_BLOCK_GROUP_METADATA);
4819         ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4820                                   BTRFS_RESERVE_FLUSH_ALL);
4821
4822         if (ret == -ENOSPC && use_global_rsv)
4823                 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4824
4825         if (ret) {
4826                 if (*qgroup_reserved)
4827                         btrfs_qgroup_free(root, *qgroup_reserved);
4828         }
4829
4830         return ret;
4831 }
4832
4833 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4834                                       struct btrfs_block_rsv *rsv,
4835                                       u64 qgroup_reserved)
4836 {
4837         btrfs_block_rsv_release(root, rsv, (u64)-1);
4838         if (qgroup_reserved)
4839                 btrfs_qgroup_free(root, qgroup_reserved);
4840 }
4841
4842 /**
4843  * drop_outstanding_extent - drop an outstanding extent
4844  * @inode: the inode we're dropping the extent for
4845  *
4846  * This is called when we are freeing up an outstanding extent, either called
4847  * after an error or after an extent is written.  This will return the number of
4848  * reserved extents that need to be freed.  This must be called with
4849  * BTRFS_I(inode)->lock held.
4850  */
4851 static unsigned drop_outstanding_extent(struct inode *inode)
4852 {
4853         unsigned drop_inode_space = 0;
4854         unsigned dropped_extents = 0;
4855
4856         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4857         BTRFS_I(inode)->outstanding_extents--;
4858
4859         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4860             test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4861                                &BTRFS_I(inode)->runtime_flags))
4862                 drop_inode_space = 1;
4863
4864         /*
4865          * If we have more or the same amount of outsanding extents than we have
4866          * reserved then we need to leave the reserved extents count alone.
4867          */
4868         if (BTRFS_I(inode)->outstanding_extents >=
4869             BTRFS_I(inode)->reserved_extents)
4870                 return drop_inode_space;
4871
4872         dropped_extents = BTRFS_I(inode)->reserved_extents -
4873                 BTRFS_I(inode)->outstanding_extents;
4874         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4875         return dropped_extents + drop_inode_space;
4876 }
4877
4878 /**
4879  * calc_csum_metadata_size - return the amount of metada space that must be
4880  *      reserved/free'd for the given bytes.
4881  * @inode: the inode we're manipulating
4882  * @num_bytes: the number of bytes in question
4883  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4884  *
4885  * This adjusts the number of csum_bytes in the inode and then returns the
4886  * correct amount of metadata that must either be reserved or freed.  We
4887  * calculate how many checksums we can fit into one leaf and then divide the
4888  * number of bytes that will need to be checksumed by this value to figure out
4889  * how many checksums will be required.  If we are adding bytes then the number
4890  * may go up and we will return the number of additional bytes that must be
4891  * reserved.  If it is going down we will return the number of bytes that must
4892  * be freed.
4893  *
4894  * This must be called with BTRFS_I(inode)->lock held.
4895  */
4896 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4897                                    int reserve)
4898 {
4899         struct btrfs_root *root = BTRFS_I(inode)->root;
4900         u64 csum_size;
4901         int num_csums_per_leaf;
4902         int num_csums;
4903         int old_csums;
4904
4905         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4906             BTRFS_I(inode)->csum_bytes == 0)
4907                 return 0;
4908
4909         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4910         if (reserve)
4911                 BTRFS_I(inode)->csum_bytes += num_bytes;
4912         else
4913                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4914         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4915         num_csums_per_leaf = (int)div64_u64(csum_size,
4916                                             sizeof(struct btrfs_csum_item) +
4917                                             sizeof(struct btrfs_disk_key));
4918         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4919         num_csums = num_csums + num_csums_per_leaf - 1;
4920         num_csums = num_csums / num_csums_per_leaf;
4921
4922         old_csums = old_csums + num_csums_per_leaf - 1;
4923         old_csums = old_csums / num_csums_per_leaf;
4924
4925         /* No change, no need to reserve more */
4926         if (old_csums == num_csums)
4927                 return 0;
4928
4929         if (reserve)
4930                 return btrfs_calc_trans_metadata_size(root,
4931                                                       num_csums - old_csums);
4932
4933         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4934 }
4935
4936 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4937 {
4938         struct btrfs_root *root = BTRFS_I(inode)->root;
4939         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4940         u64 to_reserve = 0;
4941         u64 csum_bytes;
4942         unsigned nr_extents = 0;
4943         int extra_reserve = 0;
4944         enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4945         int ret = 0;
4946         bool delalloc_lock = true;
4947         u64 to_free = 0;
4948         unsigned dropped;
4949
4950         /* If we are a free space inode we need to not flush since we will be in
4951          * the middle of a transaction commit.  We also don't need the delalloc
4952          * mutex since we won't race with anybody.  We need this mostly to make
4953          * lockdep shut its filthy mouth.
4954          */
4955         if (btrfs_is_free_space_inode(inode)) {
4956                 flush = BTRFS_RESERVE_NO_FLUSH;
4957                 delalloc_lock = false;
4958         }
4959
4960         if (flush != BTRFS_RESERVE_NO_FLUSH &&
4961             btrfs_transaction_in_commit(root->fs_info))
4962                 schedule_timeout(1);
4963
4964         if (delalloc_lock)
4965                 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4966
4967         num_bytes = ALIGN(num_bytes, root->sectorsize);
4968
4969         spin_lock(&BTRFS_I(inode)->lock);
4970         BTRFS_I(inode)->outstanding_extents++;
4971
4972         if (BTRFS_I(inode)->outstanding_extents >
4973             BTRFS_I(inode)->reserved_extents)
4974                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4975                         BTRFS_I(inode)->reserved_extents;
4976
4977         /*
4978          * Add an item to reserve for updating the inode when we complete the
4979          * delalloc io.
4980          */
4981         if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4982                       &BTRFS_I(inode)->runtime_flags)) {
4983                 nr_extents++;
4984                 extra_reserve = 1;
4985         }
4986
4987         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4988         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4989         csum_bytes = BTRFS_I(inode)->csum_bytes;
4990         spin_unlock(&BTRFS_I(inode)->lock);
4991
4992         if (root->fs_info->quota_enabled) {
4993                 ret = btrfs_qgroup_reserve(root, num_bytes +
4994                                            nr_extents * root->leafsize);
4995                 if (ret)
4996                         goto out_fail;
4997         }
4998
4999         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5000         if (unlikely(ret)) {
5001                 if (root->fs_info->quota_enabled)
5002                         btrfs_qgroup_free(root, num_bytes +
5003                                                 nr_extents * root->leafsize);
5004                 goto out_fail;
5005         }
5006
5007         spin_lock(&BTRFS_I(inode)->lock);
5008         if (extra_reserve) {
5009                 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5010                         &BTRFS_I(inode)->runtime_flags);
5011                 nr_extents--;
5012         }
5013         BTRFS_I(inode)->reserved_extents += nr_extents;
5014         spin_unlock(&BTRFS_I(inode)->lock);
5015
5016         if (delalloc_lock)
5017                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5018
5019         if (to_reserve)
5020                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5021                                               btrfs_ino(inode), to_reserve, 1);
5022         block_rsv_add_bytes(block_rsv, to_reserve, 1);
5023
5024         return 0;
5025
5026 out_fail:
5027         spin_lock(&BTRFS_I(inode)->lock);
5028         dropped = drop_outstanding_extent(inode);
5029         /*
5030          * If the inodes csum_bytes is the same as the original
5031          * csum_bytes then we know we haven't raced with any free()ers
5032          * so we can just reduce our inodes csum bytes and carry on.
5033          */
5034         if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5035                 calc_csum_metadata_size(inode, num_bytes, 0);
5036         } else {
5037                 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5038                 u64 bytes;
5039
5040                 /*
5041                  * This is tricky, but first we need to figure out how much we
5042                  * free'd from any free-ers that occured during this
5043                  * reservation, so we reset ->csum_bytes to the csum_bytes
5044                  * before we dropped our lock, and then call the free for the
5045                  * number of bytes that were freed while we were trying our
5046                  * reservation.
5047                  */
5048                 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5049                 BTRFS_I(inode)->csum_bytes = csum_bytes;
5050                 to_free = calc_csum_metadata_size(inode, bytes, 0);
5051
5052
5053                 /*
5054                  * Now we need to see how much we would have freed had we not
5055                  * been making this reservation and our ->csum_bytes were not
5056                  * artificially inflated.
5057                  */
5058                 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5059                 bytes = csum_bytes - orig_csum_bytes;
5060                 bytes = calc_csum_metadata_size(inode, bytes, 0);
5061
5062                 /*
5063                  * Now reset ->csum_bytes to what it should be.  If bytes is
5064                  * more than to_free then we would have free'd more space had we
5065                  * not had an artificially high ->csum_bytes, so we need to free
5066                  * the remainder.  If bytes is the same or less then we don't
5067                  * need to do anything, the other free-ers did the correct
5068                  * thing.
5069                  */
5070                 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5071                 if (bytes > to_free)
5072                         to_free = bytes - to_free;
5073                 else
5074                         to_free = 0;
5075         }
5076         spin_unlock(&BTRFS_I(inode)->lock);
5077         if (dropped)
5078                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5079
5080         if (to_free) {
5081                 btrfs_block_rsv_release(root, block_rsv, to_free);
5082                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5083                                               btrfs_ino(inode), to_free, 0);
5084         }
5085         if (delalloc_lock)
5086                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5087         return ret;
5088 }
5089
5090 /**
5091  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5092  * @inode: the inode to release the reservation for
5093  * @num_bytes: the number of bytes we're releasing
5094  *
5095  * This will release the metadata reservation for an inode.  This can be called
5096  * once we complete IO for a given set of bytes to release their metadata
5097  * reservations.
5098  */
5099 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5100 {
5101         struct btrfs_root *root = BTRFS_I(inode)->root;
5102         u64 to_free = 0;
5103         unsigned dropped;
5104
5105         num_bytes = ALIGN(num_bytes, root->sectorsize);
5106         spin_lock(&BTRFS_I(inode)->lock);
5107         dropped = drop_outstanding_extent(inode);
5108
5109         if (num_bytes)
5110                 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5111         spin_unlock(&BTRFS_I(inode)->lock);
5112         if (dropped > 0)
5113                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5114
5115         trace_btrfs_space_reservation(root->fs_info, "delalloc",
5116                                       btrfs_ino(inode), to_free, 0);
5117         if (root->fs_info->quota_enabled) {
5118                 btrfs_qgroup_free(root, num_bytes +
5119                                         dropped * root->leafsize);
5120         }
5121
5122         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5123                                 to_free);
5124 }
5125
5126 /**
5127  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5128  * @inode: inode we're writing to
5129  * @num_bytes: the number of bytes we want to allocate
5130  *
5131  * This will do the following things
5132  *
5133  * o reserve space in the data space info for num_bytes
5134  * o reserve space in the metadata space info based on number of outstanding
5135  *   extents and how much csums will be needed
5136  * o add to the inodes ->delalloc_bytes
5137  * o add it to the fs_info's delalloc inodes list.
5138  *
5139  * This will return 0 for success and -ENOSPC if there is no space left.
5140  */
5141 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5142 {
5143         int ret;
5144
5145         ret = btrfs_check_data_free_space(inode, num_bytes);
5146         if (ret)
5147                 return ret;
5148
5149         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5150         if (ret) {
5151                 btrfs_free_reserved_data_space(inode, num_bytes);
5152                 return ret;
5153         }
5154
5155         return 0;
5156 }
5157
5158 /**
5159  * btrfs_delalloc_release_space - release data and metadata space for delalloc
5160  * @inode: inode we're releasing space for
5161  * @num_bytes: the number of bytes we want to free up
5162  *
5163  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
5164  * called in the case that we don't need the metadata AND data reservations
5165  * anymore.  So if there is an error or we insert an inline extent.
5166  *
5167  * This function will release the metadata space that was not used and will
5168  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5169  * list if there are no delalloc bytes left.
5170  */
5171 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5172 {
5173         btrfs_delalloc_release_metadata(inode, num_bytes);
5174         btrfs_free_reserved_data_space(inode, num_bytes);
5175 }
5176
5177 static int update_block_group(struct btrfs_root *root,
5178                               u64 bytenr, u64 num_bytes, int alloc)
5179 {
5180         struct btrfs_block_group_cache *cache = NULL;
5181         struct btrfs_fs_info *info = root->fs_info;
5182         u64 total = num_bytes;
5183         u64 old_val;
5184         u64 byte_in_group;
5185         int factor;
5186
5187         /* block accounting for super block */
5188         spin_lock(&info->delalloc_root_lock);
5189         old_val = btrfs_super_bytes_used(info->super_copy);
5190         if (alloc)
5191                 old_val += num_bytes;
5192         else
5193                 old_val -= num_bytes;
5194         btrfs_set_super_bytes_used(info->super_copy, old_val);
5195         spin_unlock(&info->delalloc_root_lock);
5196
5197         while (total) {
5198                 cache = btrfs_lookup_block_group(info, bytenr);
5199                 if (!cache)
5200                         return -ENOENT;
5201                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5202                                     BTRFS_BLOCK_GROUP_RAID1 |
5203                                     BTRFS_BLOCK_GROUP_RAID10))
5204                         factor = 2;
5205                 else
5206                         factor = 1;
5207                 /*
5208                  * If this block group has free space cache written out, we
5209                  * need to make sure to load it if we are removing space.  This
5210                  * is because we need the unpinning stage to actually add the
5211                  * space back to the block group, otherwise we will leak space.
5212                  */
5213                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5214                         cache_block_group(cache, 1);
5215
5216                 byte_in_group = bytenr - cache->key.objectid;
5217                 WARN_ON(byte_in_group > cache->key.offset);
5218
5219                 spin_lock(&cache->space_info->lock);
5220                 spin_lock(&cache->lock);
5221
5222                 if (btrfs_test_opt(root, SPACE_CACHE) &&
5223                     cache->disk_cache_state < BTRFS_DC_CLEAR)
5224                         cache->disk_cache_state = BTRFS_DC_CLEAR;
5225
5226                 cache->dirty = 1;
5227                 old_val = btrfs_block_group_used(&cache->item);
5228                 num_bytes = min(total, cache->key.offset - byte_in_group);
5229                 if (alloc) {
5230                         old_val += num_bytes;
5231                         btrfs_set_block_group_used(&cache->item, old_val);
5232                         cache->reserved -= num_bytes;
5233                         cache->space_info->bytes_reserved -= num_bytes;
5234                         cache->space_info->bytes_used += num_bytes;
5235                         cache->space_info->disk_used += num_bytes * factor;
5236                         spin_unlock(&cache->lock);
5237                         spin_unlock(&cache->space_info->lock);
5238                 } else {
5239                         old_val -= num_bytes;
5240                         btrfs_set_block_group_used(&cache->item, old_val);
5241                         cache->pinned += num_bytes;
5242                         cache->space_info->bytes_pinned += num_bytes;
5243                         cache->space_info->bytes_used -= num_bytes;
5244                         cache->space_info->disk_used -= num_bytes * factor;
5245                         spin_unlock(&cache->lock);
5246                         spin_unlock(&cache->space_info->lock);
5247
5248                         set_extent_dirty(info->pinned_extents,
5249                                          bytenr, bytenr + num_bytes - 1,
5250                                          GFP_NOFS | __GFP_NOFAIL);
5251                 }
5252                 btrfs_put_block_group(cache);
5253                 total -= num_bytes;
5254                 bytenr += num_bytes;
5255         }
5256         return 0;
5257 }
5258
5259 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5260 {
5261         struct btrfs_block_group_cache *cache;
5262         u64 bytenr;
5263
5264         spin_lock(&root->fs_info->block_group_cache_lock);
5265         bytenr = root->fs_info->first_logical_byte;
5266         spin_unlock(&root->fs_info->block_group_cache_lock);
5267
5268         if (bytenr < (u64)-1)
5269                 return bytenr;
5270
5271         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5272         if (!cache)
5273                 return 0;
5274
5275         bytenr = cache->key.objectid;
5276         btrfs_put_block_group(cache);
5277
5278         return bytenr;
5279 }
5280
5281 static int pin_down_extent(struct btrfs_root *root,
5282                            struct btrfs_block_group_cache *cache,
5283                            u64 bytenr, u64 num_bytes, int reserved)
5284 {
5285         spin_lock(&cache->space_info->lock);
5286         spin_lock(&cache->lock);
5287         cache->pinned += num_bytes;
5288         cache->space_info->bytes_pinned += num_bytes;
5289         if (reserved) {
5290                 cache->reserved -= num_bytes;
5291                 cache->space_info->bytes_reserved -= num_bytes;
5292         }
5293         spin_unlock(&cache->lock);
5294         spin_unlock(&cache->space_info->lock);
5295
5296         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5297                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5298         if (reserved)
5299                 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5300         return 0;
5301 }
5302
5303 /*
5304  * this function must be called within transaction
5305  */
5306 int btrfs_pin_extent(struct btrfs_root *root,
5307                      u64 bytenr, u64 num_bytes, int reserved)
5308 {
5309         struct btrfs_block_group_cache *cache;
5310
5311         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5312         BUG_ON(!cache); /* Logic error */
5313
5314         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5315
5316         btrfs_put_block_group(cache);
5317         return 0;
5318 }
5319
5320 /*
5321  * this function must be called within transaction
5322  */
5323 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5324                                     u64 bytenr, u64 num_bytes)
5325 {
5326         struct btrfs_block_group_cache *cache;
5327         int ret;
5328
5329         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5330         if (!cache)
5331                 return -EINVAL;
5332
5333         /*
5334          * pull in the free space cache (if any) so that our pin
5335          * removes the free space from the cache.  We have load_only set
5336          * to one because the slow code to read in the free extents does check
5337          * the pinned extents.
5338          */
5339         cache_block_group(cache, 1);
5340
5341         pin_down_extent(root, cache, bytenr, num_bytes, 0);
5342
5343         /* remove us from the free space cache (if we're there at all) */
5344         ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5345         btrfs_put_block_group(cache);
5346         return ret;
5347 }
5348
5349 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5350 {
5351         int ret;
5352         struct btrfs_block_group_cache *block_group;
5353         struct btrfs_caching_control *caching_ctl;
5354
5355         block_group = btrfs_lookup_block_group(root->fs_info, start);
5356         if (!block_group)
5357                 return -EINVAL;
5358
5359         cache_block_group(block_group, 0);
5360         caching_ctl = get_caching_control(block_group);
5361
5362         if (!caching_ctl) {
5363                 /* Logic error */
5364                 BUG_ON(!block_group_cache_done(block_group));
5365                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5366         } else {
5367                 mutex_lock(&caching_ctl->mutex);
5368
5369                 if (start >= caching_ctl->progress) {
5370                         ret = add_excluded_extent(root, start, num_bytes);
5371                 } else if (start + num_bytes <= caching_ctl->progress) {
5372                         ret = btrfs_remove_free_space(block_group,
5373                                                       start, num_bytes);
5374                 } else {
5375                         num_bytes = caching_ctl->progress - start;
5376                         ret = btrfs_remove_free_space(block_group,
5377                                                       start, num_bytes);
5378                         if (ret)
5379                                 goto out_lock;
5380
5381                         num_bytes = (start + num_bytes) -
5382                                 caching_ctl->progress;
5383                         start = caching_ctl->progress;
5384                         ret = add_excluded_extent(root, start, num_bytes);
5385                 }
5386 out_lock:
5387                 mutex_unlock(&caching_ctl->mutex);
5388                 put_caching_control(caching_ctl);
5389         }
5390         btrfs_put_block_group(block_group);
5391         return ret;
5392 }
5393
5394 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5395                                  struct extent_buffer *eb)
5396 {
5397         struct btrfs_file_extent_item *item;
5398         struct btrfs_key key;
5399         int found_type;
5400         int i;
5401
5402         if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5403                 return 0;
5404
5405         for (i = 0; i < btrfs_header_nritems(eb); i++) {
5406                 btrfs_item_key_to_cpu(eb, &key, i);
5407                 if (key.type != BTRFS_EXTENT_DATA_KEY)
5408                         continue;
5409                 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5410                 found_type = btrfs_file_extent_type(eb, item);
5411                 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5412                         continue;
5413                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5414                         continue;
5415                 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5416                 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5417                 __exclude_logged_extent(log, key.objectid, key.offset);
5418         }
5419
5420         return 0;
5421 }
5422
5423 /**
5424  * btrfs_update_reserved_bytes - update the block_group and space info counters
5425  * @cache:      The cache we are manipulating
5426  * @num_bytes:  The number of bytes in question
5427  * @reserve:    One of the reservation enums
5428  *
5429  * This is called by the allocator when it reserves space, or by somebody who is
5430  * freeing space that was never actually used on disk.  For example if you
5431  * reserve some space for a new leaf in transaction A and before transaction A
5432  * commits you free that leaf, you call this with reserve set to 0 in order to
5433  * clear the reservation.
5434  *
5435  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5436  * ENOSPC accounting.  For data we handle the reservation through clearing the
5437  * delalloc bits in the io_tree.  We have to do this since we could end up
5438  * allocating less disk space for the amount of data we have reserved in the
5439  * case of compression.
5440  *
5441  * If this is a reservation and the block group has become read only we cannot
5442  * make the reservation and return -EAGAIN, otherwise this function always
5443  * succeeds.
5444  */
5445 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5446                                        u64 num_bytes, int reserve)
5447 {
5448         struct btrfs_space_info *space_info = cache->space_info;
5449         int ret = 0;
5450
5451         spin_lock(&space_info->lock);
5452         spin_lock(&cache->lock);
5453         if (reserve != RESERVE_FREE) {
5454                 if (cache->ro) {
5455                         ret = -EAGAIN;
5456                 } else {
5457                         cache->reserved += num_bytes;
5458                         space_info->bytes_reserved += num_bytes;
5459                         if (reserve == RESERVE_ALLOC) {
5460                                 trace_btrfs_space_reservation(cache->fs_info,
5461                                                 "space_info", space_info->flags,
5462                                                 num_bytes, 0);
5463                                 space_info->bytes_may_use -= num_bytes;
5464                         }
5465                 }
5466         } else {
5467                 if (cache->ro)
5468                         space_info->bytes_readonly += num_bytes;
5469                 cache->reserved -= num_bytes;
5470                 space_info->bytes_reserved -= num_bytes;
5471         }
5472         spin_unlock(&cache->lock);
5473         spin_unlock(&space_info->lock);
5474         return ret;
5475 }
5476
5477 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5478                                 struct btrfs_root *root)
5479 {
5480         struct btrfs_fs_info *fs_info = root->fs_info;
5481         struct btrfs_caching_control *next;
5482         struct btrfs_caching_control *caching_ctl;
5483         struct btrfs_block_group_cache *cache;
5484         struct btrfs_space_info *space_info;
5485
5486         down_write(&fs_info->extent_commit_sem);
5487
5488         list_for_each_entry_safe(caching_ctl, next,
5489                                  &fs_info->caching_block_groups, list) {
5490                 cache = caching_ctl->block_group;
5491                 if (block_group_cache_done(cache)) {
5492                         cache->last_byte_to_unpin = (u64)-1;
5493                         list_del_init(&caching_ctl->list);
5494                         put_caching_control(caching_ctl);
5495                 } else {
5496                         cache->last_byte_to_unpin = caching_ctl->progress;
5497                 }
5498         }
5499
5500         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5501                 fs_info->pinned_extents = &fs_info->freed_extents[1];
5502         else
5503                 fs_info->pinned_extents = &fs_info->freed_extents[0];
5504
5505         up_write(&fs_info->extent_commit_sem);
5506
5507         list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5508                 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5509
5510         update_global_block_rsv(fs_info);
5511 }
5512
5513 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5514 {
5515         struct btrfs_fs_info *fs_info = root->fs_info;
5516         struct btrfs_block_group_cache *cache = NULL;
5517         struct btrfs_space_info *space_info;
5518         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5519         u64 len;
5520         bool readonly;
5521
5522         while (start <= end) {
5523                 readonly = false;
5524                 if (!cache ||
5525                     start >= cache->key.objectid + cache->key.offset) {
5526                         if (cache)
5527                                 btrfs_put_block_group(cache);
5528                         cache = btrfs_lookup_block_group(fs_info, start);
5529                         BUG_ON(!cache); /* Logic error */
5530                 }
5531
5532                 len = cache->key.objectid + cache->key.offset - start;
5533                 len = min(len, end + 1 - start);
5534
5535                 if (start < cache->last_byte_to_unpin) {
5536                         len = min(len, cache->last_byte_to_unpin - start);
5537                         btrfs_add_free_space(cache, start, len);
5538                 }
5539
5540                 start += len;
5541                 space_info = cache->space_info;
5542
5543                 spin_lock(&space_info->lock);
5544                 spin_lock(&cache->lock);
5545                 cache->pinned -= len;
5546                 space_info->bytes_pinned -= len;
5547                 if (cache->ro) {
5548                         space_info->bytes_readonly += len;
5549                         readonly = true;
5550                 }
5551                 spin_unlock(&cache->lock);
5552                 if (!readonly && global_rsv->space_info == space_info) {
5553                         spin_lock(&global_rsv->lock);
5554                         if (!global_rsv->full) {
5555                                 len = min(len, global_rsv->size -
5556                                           global_rsv->reserved);
5557                                 global_rsv->reserved += len;
5558                                 space_info->bytes_may_use += len;
5559                                 if (global_rsv->reserved >= global_rsv->size)
5560                                         global_rsv->full = 1;
5561                         }
5562                         spin_unlock(&global_rsv->lock);
5563                 }
5564                 spin_unlock(&space_info->lock);
5565         }
5566
5567         if (cache)
5568                 btrfs_put_block_group(cache);
5569         return 0;
5570 }
5571
5572 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5573                                struct btrfs_root *root)
5574 {
5575         struct btrfs_fs_info *fs_info = root->fs_info;
5576         struct extent_io_tree *unpin;
5577         u64 start;
5578         u64 end;
5579         int ret;
5580
5581         if (trans->aborted)
5582                 return 0;
5583
5584         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5585                 unpin = &fs_info->freed_extents[1];
5586         else
5587                 unpin = &fs_info->freed_extents[0];
5588
5589         while (1) {
5590                 ret = find_first_extent_bit(unpin, 0, &start, &end,
5591                                             EXTENT_DIRTY, NULL);
5592                 if (ret)
5593                         break;
5594
5595                 if (btrfs_test_opt(root, DISCARD))
5596                         ret = btrfs_discard_extent(root, start,
5597                                                    end + 1 - start, NULL);
5598
5599                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5600                 unpin_extent_range(root, start, end);
5601                 cond_resched();
5602         }
5603
5604         return 0;
5605 }
5606
5607 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5608                              u64 owner, u64 root_objectid)
5609 {
5610         struct btrfs_space_info *space_info;
5611         u64 flags;
5612
5613         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5614                 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5615                         flags = BTRFS_BLOCK_GROUP_SYSTEM;
5616                 else
5617                         flags = BTRFS_BLOCK_GROUP_METADATA;
5618         } else {
5619                 flags = BTRFS_BLOCK_GROUP_DATA;
5620         }
5621
5622         space_info = __find_space_info(fs_info, flags);
5623         BUG_ON(!space_info); /* Logic bug */
5624         percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5625 }
5626
5627
5628 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5629                                 struct btrfs_root *root,
5630                                 u64 bytenr, u64 num_bytes, u64 parent,
5631                                 u64 root_objectid, u64 owner_objectid,
5632                                 u64 owner_offset, int refs_to_drop,
5633                                 struct btrfs_delayed_extent_op *extent_op)
5634 {
5635         struct btrfs_key key;
5636         struct btrfs_path *path;
5637         struct btrfs_fs_info *info = root->fs_info;
5638         struct btrfs_root *extent_root = info->extent_root;
5639         struct extent_buffer *leaf;
5640         struct btrfs_extent_item *ei;
5641         struct btrfs_extent_inline_ref *iref;
5642         int ret;
5643         int is_data;
5644         int extent_slot = 0;
5645         int found_extent = 0;
5646         int num_to_del = 1;
5647         u32 item_size;
5648         u64 refs;
5649         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5650                                                  SKINNY_METADATA);
5651
5652         path = btrfs_alloc_path();
5653         if (!path)
5654                 return -ENOMEM;
5655
5656         path->reada = 1;
5657         path->leave_spinning = 1;
5658
5659         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5660         BUG_ON(!is_data && refs_to_drop != 1);
5661
5662         if (is_data)
5663                 skinny_metadata = 0;
5664
5665         ret = lookup_extent_backref(trans, extent_root, path, &iref,
5666                                     bytenr, num_bytes, parent,
5667                                     root_objectid, owner_objectid,
5668                                     owner_offset);
5669         if (ret == 0) {
5670                 extent_slot = path->slots[0];
5671                 while (extent_slot >= 0) {
5672                         btrfs_item_key_to_cpu(path->nodes[0], &key,
5673                                               extent_slot);
5674                         if (key.objectid != bytenr)
5675                                 break;
5676                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5677                             key.offset == num_bytes) {
5678                                 found_extent = 1;
5679                                 break;
5680                         }
5681                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
5682                             key.offset == owner_objectid) {
5683                                 found_extent = 1;
5684                                 break;
5685                         }
5686                         if (path->slots[0] - extent_slot > 5)
5687                                 break;
5688                         extent_slot--;
5689                 }
5690 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5691                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5692                 if (found_extent && item_size < sizeof(*ei))
5693                         found_extent = 0;
5694 #endif
5695                 if (!found_extent) {
5696                         BUG_ON(iref);
5697                         ret = remove_extent_backref(trans, extent_root, path,
5698                                                     NULL, refs_to_drop,
5699                                                     is_data);
5700                         if (ret) {
5701                                 btrfs_abort_transaction(trans, extent_root, ret);
5702                                 goto out;
5703                         }
5704                         btrfs_release_path(path);
5705                         path->leave_spinning = 1;
5706
5707                         key.objectid = bytenr;
5708                         key.type = BTRFS_EXTENT_ITEM_KEY;
5709                         key.offset = num_bytes;
5710
5711                         if (!is_data && skinny_metadata) {
5712                                 key.type = BTRFS_METADATA_ITEM_KEY;
5713                                 key.offset = owner_objectid;
5714                         }
5715
5716                         ret = btrfs_search_slot(trans, extent_root,
5717                                                 &key, path, -1, 1);
5718                         if (ret > 0 && skinny_metadata && path->slots[0]) {
5719                                 /*
5720                                  * Couldn't find our skinny metadata item,
5721                                  * see if we have ye olde extent item.
5722                                  */
5723                                 path->slots[0]--;
5724                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
5725                                                       path->slots[0]);
5726                                 if (key.objectid == bytenr &&
5727                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
5728                                     key.offset == num_bytes)
5729                                         ret = 0;
5730                         }
5731
5732                         if (ret > 0 && skinny_metadata) {
5733                                 skinny_metadata = false;
5734                                 key.type = BTRFS_EXTENT_ITEM_KEY;
5735                                 key.offset = num_bytes;
5736                                 btrfs_release_path(path);
5737                                 ret = btrfs_search_slot(trans, extent_root,
5738                                                         &key, path, -1, 1);
5739                         }
5740
5741                         if (ret) {
5742                                 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5743                                         ret, bytenr);
5744                                 if (ret > 0)
5745                                         btrfs_print_leaf(extent_root,
5746                                                          path->nodes[0]);
5747                         }
5748                         if (ret < 0) {
5749                                 btrfs_abort_transaction(trans, extent_root, ret);
5750                                 goto out;
5751                         }
5752                         extent_slot = path->slots[0];
5753                 }
5754         } else if (WARN_ON(ret == -ENOENT)) {
5755                 btrfs_print_leaf(extent_root, path->nodes[0]);
5756                 btrfs_err(info,
5757                         "unable to find ref byte nr %llu parent %llu root %llu  owner %llu offset %llu",
5758                         bytenr, parent, root_objectid, owner_objectid,
5759                         owner_offset);
5760         } else {
5761                 btrfs_abort_transaction(trans, extent_root, ret);
5762                 goto out;
5763         }
5764
5765         leaf = path->nodes[0];
5766         item_size = btrfs_item_size_nr(leaf, extent_slot);
5767 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5768         if (item_size < sizeof(*ei)) {
5769                 BUG_ON(found_extent || extent_slot != path->slots[0]);
5770                 ret = convert_extent_item_v0(trans, extent_root, path,
5771                                              owner_objectid, 0);
5772                 if (ret < 0) {
5773                         btrfs_abort_transaction(trans, extent_root, ret);
5774                         goto out;
5775                 }
5776
5777                 btrfs_release_path(path);
5778                 path->leave_spinning = 1;
5779
5780                 key.objectid = bytenr;
5781                 key.type = BTRFS_EXTENT_ITEM_KEY;
5782                 key.offset = num_bytes;
5783
5784                 ret = btrfs_search_slot(trans, extent_root, &key, path,
5785                                         -1, 1);
5786                 if (ret) {
5787                         btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5788                                 ret, bytenr);
5789                         btrfs_print_leaf(extent_root, path->nodes[0]);
5790                 }
5791                 if (ret < 0) {
5792                         btrfs_abort_transaction(trans, extent_root, ret);
5793                         goto out;
5794                 }
5795
5796                 extent_slot = path->slots[0];
5797                 leaf = path->nodes[0];
5798                 item_size = btrfs_item_size_nr(leaf, extent_slot);
5799         }
5800 #endif
5801         BUG_ON(item_size < sizeof(*ei));
5802         ei = btrfs_item_ptr(leaf, extent_slot,
5803                             struct btrfs_extent_item);
5804         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5805             key.type == BTRFS_EXTENT_ITEM_KEY) {
5806                 struct btrfs_tree_block_info *bi;
5807                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5808                 bi = (struct btrfs_tree_block_info *)(ei + 1);
5809                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5810         }
5811
5812         refs = btrfs_extent_refs(leaf, ei);
5813         if (refs < refs_to_drop) {
5814                 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5815                           "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5816                 ret = -EINVAL;
5817                 btrfs_abort_transaction(trans, extent_root, ret);
5818                 goto out;
5819         }
5820         refs -= refs_to_drop;
5821
5822         if (refs > 0) {
5823                 if (extent_op)
5824                         __run_delayed_extent_op(extent_op, leaf, ei);
5825                 /*
5826                  * In the case of inline back ref, reference count will
5827                  * be updated by remove_extent_backref
5828                  */
5829                 if (iref) {
5830                         BUG_ON(!found_extent);
5831                 } else {
5832                         btrfs_set_extent_refs(leaf, ei, refs);
5833                         btrfs_mark_buffer_dirty(leaf);
5834                 }
5835                 if (found_extent) {
5836                         ret = remove_extent_backref(trans, extent_root, path,
5837                                                     iref, refs_to_drop,
5838                                                     is_data);
5839                         if (ret) {
5840                                 btrfs_abort_transaction(trans, extent_root, ret);
5841                                 goto out;
5842                         }
5843                 }
5844                 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5845                                  root_objectid);
5846         } else {
5847                 if (found_extent) {
5848                         BUG_ON(is_data && refs_to_drop !=
5849                                extent_data_ref_count(root, path, iref));
5850                         if (iref) {
5851                                 BUG_ON(path->slots[0] != extent_slot);
5852                         } else {
5853                                 BUG_ON(path->slots[0] != extent_slot + 1);
5854                                 path->slots[0] = extent_slot;
5855                                 num_to_del = 2;
5856                         }
5857                 }
5858
5859                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5860                                       num_to_del);
5861                 if (ret) {
5862                         btrfs_abort_transaction(trans, extent_root, ret);
5863                         goto out;
5864                 }
5865                 btrfs_release_path(path);
5866
5867                 if (is_data) {
5868                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5869                         if (ret) {
5870                                 btrfs_abort_transaction(trans, extent_root, ret);
5871                                 goto out;
5872                         }
5873                 }
5874
5875                 ret = update_block_group(root, bytenr, num_bytes, 0);
5876                 if (ret) {
5877                         btrfs_abort_transaction(trans, extent_root, ret);
5878                         goto out;
5879                 }
5880         }
5881 out:
5882         btrfs_free_path(path);
5883         return ret;
5884 }
5885
5886 /*
5887  * when we free an block, it is possible (and likely) that we free the last
5888  * delayed ref for that extent as well.  This searches the delayed ref tree for
5889  * a given extent, and if there are no other delayed refs to be processed, it
5890  * removes it from the tree.
5891  */
5892 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5893                                       struct btrfs_root *root, u64 bytenr)
5894 {
5895         struct btrfs_delayed_ref_head *head;
5896         struct btrfs_delayed_ref_root *delayed_refs;
5897         struct btrfs_delayed_ref_node *ref;
5898         struct rb_node *node;
5899         int ret = 0;
5900
5901         delayed_refs = &trans->transaction->delayed_refs;
5902         spin_lock(&delayed_refs->lock);
5903         head = btrfs_find_delayed_ref_head(trans, bytenr);
5904         if (!head)
5905                 goto out;
5906
5907         node = rb_prev(&head->node.rb_node);
5908         if (!node)
5909                 goto out;
5910
5911         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5912
5913         /* there are still entries for this ref, we can't drop it */
5914         if (ref->bytenr == bytenr)
5915                 goto out;
5916
5917         if (head->extent_op) {
5918                 if (!head->must_insert_reserved)
5919                         goto out;
5920                 btrfs_free_delayed_extent_op(head->extent_op);
5921                 head->extent_op = NULL;
5922         }
5923
5924         /*
5925          * waiting for the lock here would deadlock.  If someone else has it
5926          * locked they are already in the process of dropping it anyway
5927          */
5928         if (!mutex_trylock(&head->mutex))
5929                 goto out;
5930
5931         /*
5932          * at this point we have a head with no other entries.  Go
5933          * ahead and process it.
5934          */
5935         head->node.in_tree = 0;
5936         rb_erase(&head->node.rb_node, &delayed_refs->root);
5937
5938         delayed_refs->num_entries--;
5939
5940         /*
5941          * we don't take a ref on the node because we're removing it from the
5942          * tree, so we just steal the ref the tree was holding.
5943          */
5944         delayed_refs->num_heads--;
5945         if (list_empty(&head->cluster))
5946                 delayed_refs->num_heads_ready--;
5947
5948         list_del_init(&head->cluster);
5949         spin_unlock(&delayed_refs->lock);
5950
5951         BUG_ON(head->extent_op);
5952         if (head->must_insert_reserved)
5953                 ret = 1;
5954
5955         mutex_unlock(&head->mutex);
5956         btrfs_put_delayed_ref(&head->node);
5957         return ret;
5958 out:
5959         spin_unlock(&delayed_refs->lock);
5960         return 0;
5961 }
5962
5963 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5964                            struct btrfs_root *root,
5965                            struct extent_buffer *buf,
5966                            u64 parent, int last_ref)
5967 {
5968         struct btrfs_block_group_cache *cache = NULL;
5969         int pin = 1;
5970         int ret;
5971
5972         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5973                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5974                                         buf->start, buf->len,
5975                                         parent, root->root_key.objectid,
5976                                         btrfs_header_level(buf),
5977                                         BTRFS_DROP_DELAYED_REF, NULL, 0);
5978                 BUG_ON(ret); /* -ENOMEM */
5979         }
5980
5981         if (!last_ref)
5982                 return;
5983
5984         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5985
5986         if (btrfs_header_generation(buf) == trans->transid) {
5987                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5988                         ret = check_ref_cleanup(trans, root, buf->start);
5989                         if (!ret)
5990                                 goto out;
5991                 }
5992
5993                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5994                         pin_down_extent(root, cache, buf->start, buf->len, 1);
5995                         goto out;
5996                 }
5997
5998                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5999
6000                 btrfs_add_free_space(cache, buf->start, buf->len);
6001                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
6002                 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6003                 pin = 0;
6004         }
6005 out:
6006         if (pin)
6007                 add_pinned_bytes(root->fs_info, buf->len,
6008                                  btrfs_header_level(buf),
6009                                  root->root_key.objectid);
6010
6011         /*
6012          * Deleting the buffer, clear the corrupt flag since it doesn't matter
6013          * anymore.
6014          */
6015         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6016         btrfs_put_block_group(cache);
6017 }
6018
6019 /* Can return -ENOMEM */
6020 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6021                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6022                       u64 owner, u64 offset, int for_cow)
6023 {
6024         int ret;
6025         struct btrfs_fs_info *fs_info = root->fs_info;
6026
6027         add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6028
6029         /*
6030          * tree log blocks never actually go into the extent allocation
6031          * tree, just update pinning info and exit early.
6032          */
6033         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6034                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6035                 /* unlocks the pinned mutex */
6036                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6037                 ret = 0;
6038         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6039                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6040                                         num_bytes,
6041                                         parent, root_objectid, (int)owner,
6042                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6043         } else {
6044                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6045                                                 num_bytes,
6046                                                 parent, root_objectid, owner,
6047                                                 offset, BTRFS_DROP_DELAYED_REF,
6048                                                 NULL, for_cow);
6049         }
6050         return ret;
6051 }
6052
6053 static u64 stripe_align(struct btrfs_root *root,
6054                         struct btrfs_block_group_cache *cache,
6055                         u64 val, u64 num_bytes)
6056 {
6057         u64 ret = ALIGN(val, root->stripesize);
6058         return ret;
6059 }
6060
6061 /*
6062  * when we wait for progress in the block group caching, its because
6063  * our allocation attempt failed at least once.  So, we must sleep
6064  * and let some progress happen before we try again.
6065  *
6066  * This function will sleep at least once waiting for new free space to
6067  * show up, and then it will check the block group free space numbers
6068  * for our min num_bytes.  Another option is to have it go ahead
6069  * and look in the rbtree for a free extent of a given size, but this
6070  * is a good start.
6071  *
6072  * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6073  * any of the information in this block group.
6074  */
6075 static noinline void
6076 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6077                                 u64 num_bytes)
6078 {
6079         struct btrfs_caching_control *caching_ctl;
6080
6081         caching_ctl = get_caching_control(cache);
6082         if (!caching_ctl)
6083                 return;
6084
6085         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6086                    (cache->free_space_ctl->free_space >= num_bytes));
6087
6088         put_caching_control(caching_ctl);
6089 }
6090
6091 static noinline int
6092 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6093 {
6094         struct btrfs_caching_control *caching_ctl;
6095         int ret = 0;
6096
6097         caching_ctl = get_caching_control(cache);
6098         if (!caching_ctl)
6099                 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6100
6101         wait_event(caching_ctl->wait, block_group_cache_done(cache));
6102         if (cache->cached == BTRFS_CACHE_ERROR)
6103                 ret = -EIO;
6104         put_caching_control(caching_ctl);
6105         return ret;
6106 }
6107
6108 int __get_raid_index(u64 flags)
6109 {
6110         if (flags & BTRFS_BLOCK_GROUP_RAID10)
6111                 return BTRFS_RAID_RAID10;
6112         else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6113                 return BTRFS_RAID_RAID1;
6114         else if (flags & BTRFS_BLOCK_GROUP_DUP)
6115                 return BTRFS_RAID_DUP;
6116         else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6117                 return BTRFS_RAID_RAID0;
6118         else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6119                 return BTRFS_RAID_RAID5;
6120         else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6121                 return BTRFS_RAID_RAID6;
6122
6123         return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6124 }
6125
6126 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6127 {
6128         return __get_raid_index(cache->flags);
6129 }
6130
6131 enum btrfs_loop_type {
6132         LOOP_CACHING_NOWAIT = 0,
6133         LOOP_CACHING_WAIT = 1,
6134         LOOP_ALLOC_CHUNK = 2,
6135         LOOP_NO_EMPTY_SIZE = 3,
6136 };
6137
6138 /*
6139  * walks the btree of allocated extents and find a hole of a given size.
6140  * The key ins is changed to record the hole:
6141  * ins->objectid == start position
6142  * ins->flags = BTRFS_EXTENT_ITEM_KEY
6143  * ins->offset == the size of the hole.
6144  * Any available blocks before search_start are skipped.
6145  *
6146  * If there is no suitable free space, we will record the max size of
6147  * the free space extent currently.
6148  */
6149 static noinline int find_free_extent(struct btrfs_root *orig_root,
6150                                      u64 num_bytes, u64 empty_size,
6151                                      u64 hint_byte, struct btrfs_key *ins,
6152                                      u64 flags)
6153 {
6154         int ret = 0;
6155         struct btrfs_root *root = orig_root->fs_info->extent_root;
6156         struct btrfs_free_cluster *last_ptr = NULL;
6157         struct btrfs_block_group_cache *block_group = NULL;
6158         struct btrfs_block_group_cache *used_block_group;
6159         u64 search_start = 0;
6160         u64 max_extent_size = 0;
6161         int empty_cluster = 2 * 1024 * 1024;
6162         struct btrfs_space_info *space_info;
6163         int loop = 0;
6164         int index = __get_raid_index(flags);
6165         int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6166                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6167         bool found_uncached_bg = false;
6168         bool failed_cluster_refill = false;
6169         bool failed_alloc = false;
6170         bool use_cluster = true;
6171         bool have_caching_bg = false;
6172
6173         WARN_ON(num_bytes < root->sectorsize);
6174         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6175         ins->objectid = 0;
6176         ins->offset = 0;
6177
6178         trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6179
6180         space_info = __find_space_info(root->fs_info, flags);
6181         if (!space_info) {
6182                 btrfs_err(root->fs_info, "No space info for %llu", flags);
6183                 return -ENOSPC;
6184         }
6185
6186         /*
6187          * If the space info is for both data and metadata it means we have a
6188          * small filesystem and we can't use the clustering stuff.
6189          */
6190         if (btrfs_mixed_space_info(space_info))
6191                 use_cluster = false;
6192
6193         if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6194                 last_ptr = &root->fs_info->meta_alloc_cluster;
6195                 if (!btrfs_test_opt(root, SSD))
6196                         empty_cluster = 64 * 1024;
6197         }
6198
6199         if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6200             btrfs_test_opt(root, SSD)) {
6201                 last_ptr = &root->fs_info->data_alloc_cluster;
6202         }
6203
6204         if (last_ptr) {
6205                 spin_lock(&last_ptr->lock);
6206                 if (last_ptr->block_group)
6207                         hint_byte = last_ptr->window_start;
6208                 spin_unlock(&last_ptr->lock);
6209         }
6210
6211         search_start = max(search_start, first_logical_byte(root, 0));
6212         search_start = max(search_start, hint_byte);
6213
6214         if (!last_ptr)
6215                 empty_cluster = 0;
6216
6217         if (search_start == hint_byte) {
6218                 block_group = btrfs_lookup_block_group(root->fs_info,
6219                                                        search_start);
6220                 used_block_group = block_group;
6221                 /*
6222                  * we don't want to use the block group if it doesn't match our
6223                  * allocation bits, or if its not cached.
6224                  *
6225                  * However if we are re-searching with an ideal block group
6226                  * picked out then we don't care that the block group is cached.
6227                  */
6228                 if (block_group && block_group_bits(block_group, flags) &&
6229                     block_group->cached != BTRFS_CACHE_NO) {
6230                         down_read(&space_info->groups_sem);
6231                         if (list_empty(&block_group->list) ||
6232                             block_group->ro) {
6233                                 /*
6234                                  * someone is removing this block group,
6235                                  * we can't jump into the have_block_group
6236                                  * target because our list pointers are not
6237                                  * valid
6238                                  */
6239                                 btrfs_put_block_group(block_group);
6240                                 up_read(&space_info->groups_sem);
6241                         } else {
6242                                 index = get_block_group_index(block_group);
6243                                 goto have_block_group;
6244                         }
6245                 } else if (block_group) {
6246                         btrfs_put_block_group(block_group);
6247                 }
6248         }
6249 search:
6250         have_caching_bg = false;
6251         down_read(&space_info->groups_sem);
6252         list_for_each_entry(block_group, &space_info->block_groups[index],
6253                             list) {
6254                 u64 offset;
6255                 int cached;
6256
6257                 used_block_group = block_group;
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                         found_uncached_bg = true;
6286                         ret = cache_block_group(block_group, 0);
6287                         BUG_ON(ret < 0);
6288                         ret = 0;
6289                 }
6290
6291                 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6292                         goto loop;
6293                 if (unlikely(block_group->ro))
6294                         goto loop;
6295
6296                 /*
6297                  * Ok we want to try and use the cluster allocator, so
6298                  * lets look there
6299                  */
6300                 if (last_ptr) {
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                                 used_block_group = block_group;
6313                                 goto refill_cluster;
6314                         }
6315
6316                         if (used_block_group != block_group)
6317                                 btrfs_get_block_group(used_block_group);
6318
6319                         offset = btrfs_alloc_from_cluster(used_block_group,
6320                                                 last_ptr,
6321                                                 num_bytes,
6322                                                 used_block_group->key.objectid,
6323                                                 &max_extent_size);
6324                         if (offset) {
6325                                 /* we have a block, we're done */
6326                                 spin_unlock(&last_ptr->refill_lock);
6327                                 trace_btrfs_reserve_extent_cluster(root,
6328                                         block_group, search_start, num_bytes);
6329                                 goto checks;
6330                         }
6331
6332                         WARN_ON(last_ptr->block_group != used_block_group);
6333                         if (used_block_group != block_group) {
6334                                 btrfs_put_block_group(used_block_group);
6335                                 used_block_group = block_group;
6336                         }
6337 refill_cluster:
6338                         BUG_ON(used_block_group != block_group);
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, used_block_group,
6458                                             offset, num_bytes);
6459
6460                 /* move on to the next group */
6461                 if (search_start + num_bytes >
6462                     used_block_group->key.objectid + used_block_group->key.offset) {
6463                         btrfs_add_free_space(used_block_group, offset, num_bytes);
6464                         goto loop;
6465                 }
6466
6467                 if (offset < search_start)
6468                         btrfs_add_free_space(used_block_group, offset,
6469                                              search_start - offset);
6470                 BUG_ON(offset > search_start);
6471
6472                 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6473                                                   alloc_type);
6474                 if (ret == -EAGAIN) {
6475                         btrfs_add_free_space(used_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                 if (used_block_group != block_group)
6486                         btrfs_put_block_group(used_block_group);
6487                 btrfs_put_block_group(block_group);
6488                 break;
6489 loop:
6490                 failed_cluster_refill = false;
6491                 failed_alloc = false;
6492                 BUG_ON(index != get_block_group_index(block_group));
6493                 if (used_block_group != block_group)
6494                         btrfs_put_block_group(used_block_group);
6495                 btrfs_put_block_group(block_group);
6496         }
6497         up_read(&space_info->groups_sem);
6498
6499         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6500                 goto search;
6501
6502         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6503                 goto search;
6504
6505         /*
6506          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6507          *                      caching kthreads as we move along
6508          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6509          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6510          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6511          *                      again
6512          */
6513         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6514                 index = 0;
6515                 loop++;
6516                 if (loop == LOOP_ALLOC_CHUNK) {
6517                         struct btrfs_trans_handle *trans;
6518
6519                         trans = btrfs_join_transaction(root);
6520                         if (IS_ERR(trans)) {
6521                                 ret = PTR_ERR(trans);
6522                                 goto out;
6523                         }
6524
6525                         ret = do_chunk_alloc(trans, root, flags,
6526                                              CHUNK_ALLOC_FORCE);
6527                         /*
6528                          * Do not bail out on ENOSPC since we
6529                          * can do more things.
6530                          */
6531                         if (ret < 0 && ret != -ENOSPC)
6532                                 btrfs_abort_transaction(trans,
6533                                                         root, ret);
6534                         else
6535                                 ret = 0;
6536                         btrfs_end_transaction(trans, root);
6537                         if (ret)
6538                                 goto out;
6539                 }
6540
6541                 if (loop == LOOP_NO_EMPTY_SIZE) {
6542                         empty_size = 0;
6543                         empty_cluster = 0;
6544                 }
6545
6546                 goto search;
6547         } else if (!ins->objectid) {
6548                 ret = -ENOSPC;
6549         } else if (ins->objectid) {
6550                 ret = 0;
6551         }
6552 out:
6553         if (ret == -ENOSPC)
6554                 ins->offset = max_extent_size;
6555         return ret;
6556 }
6557
6558 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6559                             int dump_block_groups)
6560 {
6561         struct btrfs_block_group_cache *cache;
6562         int index = 0;
6563
6564         spin_lock(&info->lock);
6565         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6566                info->flags,
6567                info->total_bytes - info->bytes_used - info->bytes_pinned -
6568                info->bytes_reserved - info->bytes_readonly,
6569                (info->full) ? "" : "not ");
6570         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6571                "reserved=%llu, may_use=%llu, readonly=%llu\n",
6572                info->total_bytes, info->bytes_used, info->bytes_pinned,
6573                info->bytes_reserved, info->bytes_may_use,
6574                info->bytes_readonly);
6575         spin_unlock(&info->lock);
6576
6577         if (!dump_block_groups)
6578                 return;
6579
6580         down_read(&info->groups_sem);
6581 again:
6582         list_for_each_entry(cache, &info->block_groups[index], list) {
6583                 spin_lock(&cache->lock);
6584                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6585                        cache->key.objectid, cache->key.offset,
6586                        btrfs_block_group_used(&cache->item), cache->pinned,
6587                        cache->reserved, cache->ro ? "[readonly]" : "");
6588                 btrfs_dump_free_space(cache, bytes);
6589                 spin_unlock(&cache->lock);
6590         }
6591         if (++index < BTRFS_NR_RAID_TYPES)
6592                 goto again;
6593         up_read(&info->groups_sem);
6594 }
6595
6596 int btrfs_reserve_extent(struct btrfs_root *root,
6597                          u64 num_bytes, u64 min_alloc_size,
6598                          u64 empty_size, u64 hint_byte,
6599                          struct btrfs_key *ins, int is_data)
6600 {
6601         bool final_tried = false;
6602         u64 flags;
6603         int ret;
6604
6605         flags = btrfs_get_alloc_profile(root, is_data);
6606 again:
6607         WARN_ON(num_bytes < root->sectorsize);
6608         ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6609                                flags);
6610
6611         if (ret == -ENOSPC) {
6612                 if (!final_tried && ins->offset) {
6613                         num_bytes = min(num_bytes >> 1, ins->offset);
6614                         num_bytes = round_down(num_bytes, root->sectorsize);
6615                         num_bytes = max(num_bytes, min_alloc_size);
6616                         if (num_bytes == min_alloc_size)
6617                                 final_tried = true;
6618                         goto again;
6619                 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6620                         struct btrfs_space_info *sinfo;
6621
6622                         sinfo = __find_space_info(root->fs_info, flags);
6623                         btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6624                                 flags, num_bytes);
6625                         if (sinfo)
6626                                 dump_space_info(sinfo, num_bytes, 1);
6627                 }
6628         }
6629
6630         return ret;
6631 }
6632
6633 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6634                                         u64 start, u64 len, int pin)
6635 {
6636         struct btrfs_block_group_cache *cache;
6637         int ret = 0;
6638
6639         cache = btrfs_lookup_block_group(root->fs_info, start);
6640         if (!cache) {
6641                 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6642                         start);
6643                 return -ENOSPC;
6644         }
6645
6646         if (btrfs_test_opt(root, DISCARD))
6647                 ret = btrfs_discard_extent(root, start, len, NULL);
6648
6649         if (pin)
6650                 pin_down_extent(root, cache, start, len, 1);
6651         else {
6652                 btrfs_add_free_space(cache, start, len);
6653                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6654         }
6655         btrfs_put_block_group(cache);
6656
6657         trace_btrfs_reserved_extent_free(root, start, len);
6658
6659         return ret;
6660 }
6661
6662 int btrfs_free_reserved_extent(struct btrfs_root *root,
6663                                         u64 start, u64 len)
6664 {
6665         return __btrfs_free_reserved_extent(root, start, len, 0);
6666 }
6667
6668 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6669                                        u64 start, u64 len)
6670 {
6671         return __btrfs_free_reserved_extent(root, start, len, 1);
6672 }
6673
6674 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6675                                       struct btrfs_root *root,
6676                                       u64 parent, u64 root_objectid,
6677                                       u64 flags, u64 owner, u64 offset,
6678                                       struct btrfs_key *ins, int ref_mod)
6679 {
6680         int ret;
6681         struct btrfs_fs_info *fs_info = root->fs_info;
6682         struct btrfs_extent_item *extent_item;
6683         struct btrfs_extent_inline_ref *iref;
6684         struct btrfs_path *path;
6685         struct extent_buffer *leaf;
6686         int type;
6687         u32 size;
6688
6689         if (parent > 0)
6690                 type = BTRFS_SHARED_DATA_REF_KEY;
6691         else
6692                 type = BTRFS_EXTENT_DATA_REF_KEY;
6693
6694         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6695
6696         path = btrfs_alloc_path();
6697         if (!path)
6698                 return -ENOMEM;
6699
6700         path->leave_spinning = 1;
6701         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6702                                       ins, size);
6703         if (ret) {
6704                 btrfs_free_path(path);
6705                 return ret;
6706         }
6707
6708         leaf = path->nodes[0];
6709         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6710                                      struct btrfs_extent_item);
6711         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6712         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6713         btrfs_set_extent_flags(leaf, extent_item,
6714                                flags | BTRFS_EXTENT_FLAG_DATA);
6715
6716         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6717         btrfs_set_extent_inline_ref_type(leaf, iref, type);
6718         if (parent > 0) {
6719                 struct btrfs_shared_data_ref *ref;
6720                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6721                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6722                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6723         } else {
6724                 struct btrfs_extent_data_ref *ref;
6725                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6726                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6727                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6728                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6729                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6730         }
6731
6732         btrfs_mark_buffer_dirty(path->nodes[0]);
6733         btrfs_free_path(path);
6734
6735         ret = update_block_group(root, ins->objectid, ins->offset, 1);
6736         if (ret) { /* -ENOENT, logic error */
6737                 btrfs_err(fs_info, "update block group failed for %llu %llu",
6738                         ins->objectid, ins->offset);
6739                 BUG();
6740         }
6741         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6742         return ret;
6743 }
6744
6745 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6746                                      struct btrfs_root *root,
6747                                      u64 parent, u64 root_objectid,
6748                                      u64 flags, struct btrfs_disk_key *key,
6749                                      int level, struct btrfs_key *ins)
6750 {
6751         int ret;
6752         struct btrfs_fs_info *fs_info = root->fs_info;
6753         struct btrfs_extent_item *extent_item;
6754         struct btrfs_tree_block_info *block_info;
6755         struct btrfs_extent_inline_ref *iref;
6756         struct btrfs_path *path;
6757         struct extent_buffer *leaf;
6758         u32 size = sizeof(*extent_item) + sizeof(*iref);
6759         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6760                                                  SKINNY_METADATA);
6761
6762         if (!skinny_metadata)
6763                 size += sizeof(*block_info);
6764
6765         path = btrfs_alloc_path();
6766         if (!path) {
6767                 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6768                                                    root->leafsize);
6769                 return -ENOMEM;
6770         }
6771
6772         path->leave_spinning = 1;
6773         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6774                                       ins, size);
6775         if (ret) {
6776                 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6777                                                    root->leafsize);
6778                 btrfs_free_path(path);
6779                 return ret;
6780         }
6781
6782         leaf = path->nodes[0];
6783         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6784                                      struct btrfs_extent_item);
6785         btrfs_set_extent_refs(leaf, extent_item, 1);
6786         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6787         btrfs_set_extent_flags(leaf, extent_item,
6788                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6789
6790         if (skinny_metadata) {
6791                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6792         } else {
6793                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6794                 btrfs_set_tree_block_key(leaf, block_info, key);
6795                 btrfs_set_tree_block_level(leaf, block_info, level);
6796                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6797         }
6798
6799         if (parent > 0) {
6800                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6801                 btrfs_set_extent_inline_ref_type(leaf, iref,
6802                                                  BTRFS_SHARED_BLOCK_REF_KEY);
6803                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6804         } else {
6805                 btrfs_set_extent_inline_ref_type(leaf, iref,
6806                                                  BTRFS_TREE_BLOCK_REF_KEY);
6807                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6808         }
6809
6810         btrfs_mark_buffer_dirty(leaf);
6811         btrfs_free_path(path);
6812
6813         ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6814         if (ret) { /* -ENOENT, logic error */
6815                 btrfs_err(fs_info, "update block group failed for %llu %llu",
6816                         ins->objectid, ins->offset);
6817                 BUG();
6818         }
6819
6820         trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6821         return ret;
6822 }
6823
6824 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6825                                      struct btrfs_root *root,
6826                                      u64 root_objectid, u64 owner,
6827                                      u64 offset, struct btrfs_key *ins)
6828 {
6829         int ret;
6830
6831         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6832
6833         ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6834                                          ins->offset, 0,
6835                                          root_objectid, owner, offset,
6836                                          BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6837         return ret;
6838 }
6839
6840 /*
6841  * this is used by the tree logging recovery code.  It records that
6842  * an extent has been allocated and makes sure to clear the free
6843  * space cache bits as well
6844  */
6845 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6846                                    struct btrfs_root *root,
6847                                    u64 root_objectid, u64 owner, u64 offset,
6848                                    struct btrfs_key *ins)
6849 {
6850         int ret;
6851         struct btrfs_block_group_cache *block_group;
6852
6853         /*
6854          * Mixed block groups will exclude before processing the log so we only
6855          * need to do the exlude dance if this fs isn't mixed.
6856          */
6857         if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6858                 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6859                 if (ret)
6860                         return ret;
6861         }
6862
6863         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6864         if (!block_group)
6865                 return -EINVAL;
6866
6867         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6868                                           RESERVE_ALLOC_NO_ACCOUNT);
6869         BUG_ON(ret); /* logic error */
6870         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6871                                          0, owner, offset, ins, 1);
6872         btrfs_put_block_group(block_group);
6873         return ret;
6874 }
6875
6876 static struct extent_buffer *
6877 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6878                       u64 bytenr, u32 blocksize, int level)
6879 {
6880         struct extent_buffer *buf;
6881
6882         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6883         if (!buf)
6884                 return ERR_PTR(-ENOMEM);
6885         btrfs_set_header_generation(buf, trans->transid);
6886         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6887         btrfs_tree_lock(buf);
6888         clean_tree_block(trans, root, buf);
6889         clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6890
6891         btrfs_set_lock_blocking(buf);
6892         btrfs_set_buffer_uptodate(buf);
6893
6894         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6895                 /*
6896                  * we allow two log transactions at a time, use different
6897                  * EXENT bit to differentiate dirty pages.
6898                  */
6899                 if (root->log_transid % 2 == 0)
6900                         set_extent_dirty(&root->dirty_log_pages, buf->start,
6901                                         buf->start + buf->len - 1, GFP_NOFS);
6902                 else
6903                         set_extent_new(&root->dirty_log_pages, buf->start,
6904                                         buf->start + buf->len - 1, GFP_NOFS);
6905         } else {
6906                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6907                          buf->start + buf->len - 1, GFP_NOFS);
6908         }
6909         trans->blocks_used++;
6910         /* this returns a buffer locked for blocking */
6911         return buf;
6912 }
6913
6914 static struct btrfs_block_rsv *
6915 use_block_rsv(struct btrfs_trans_handle *trans,
6916               struct btrfs_root *root, u32 blocksize)
6917 {
6918         struct btrfs_block_rsv *block_rsv;
6919         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6920         int ret;
6921         bool global_updated = false;
6922
6923         block_rsv = get_block_rsv(trans, root);
6924
6925         if (unlikely(block_rsv->size == 0))
6926                 goto try_reserve;
6927 again:
6928         ret = block_rsv_use_bytes(block_rsv, blocksize);
6929         if (!ret)
6930                 return block_rsv;
6931
6932         if (block_rsv->failfast)
6933                 return ERR_PTR(ret);
6934
6935         if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6936                 global_updated = true;
6937                 update_global_block_rsv(root->fs_info);
6938                 goto again;
6939         }
6940
6941         if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6942                 static DEFINE_RATELIMIT_STATE(_rs,
6943                                 DEFAULT_RATELIMIT_INTERVAL * 10,
6944                                 /*DEFAULT_RATELIMIT_BURST*/ 1);
6945                 if (__ratelimit(&_rs))
6946                         WARN(1, KERN_DEBUG
6947                                 "btrfs: block rsv returned %d\n", ret);
6948         }
6949 try_reserve:
6950         ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6951                                      BTRFS_RESERVE_NO_FLUSH);
6952         if (!ret)
6953                 return block_rsv;
6954         /*
6955          * If we couldn't reserve metadata bytes try and use some from
6956          * the global reserve if its space type is the same as the global
6957          * reservation.
6958          */
6959         if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6960             block_rsv->space_info == global_rsv->space_info) {
6961                 ret = block_rsv_use_bytes(global_rsv, blocksize);
6962                 if (!ret)
6963                         return global_rsv;
6964         }
6965         return ERR_PTR(ret);
6966 }
6967
6968 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6969                             struct btrfs_block_rsv *block_rsv, u32 blocksize)
6970 {
6971         block_rsv_add_bytes(block_rsv, blocksize, 0);
6972         block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6973 }
6974
6975 /*
6976  * finds a free extent and does all the dirty work required for allocation
6977  * returns the key for the extent through ins, and a tree buffer for
6978  * the first block of the extent through buf.
6979  *
6980  * returns the tree buffer or NULL.
6981  */
6982 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6983                                         struct btrfs_root *root, u32 blocksize,
6984                                         u64 parent, u64 root_objectid,
6985                                         struct btrfs_disk_key *key, int level,
6986                                         u64 hint, u64 empty_size)
6987 {
6988         struct btrfs_key ins;
6989         struct btrfs_block_rsv *block_rsv;
6990         struct extent_buffer *buf;
6991         u64 flags = 0;
6992         int ret;
6993         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6994                                                  SKINNY_METADATA);
6995
6996         block_rsv = use_block_rsv(trans, root, blocksize);
6997         if (IS_ERR(block_rsv))
6998                 return ERR_CAST(block_rsv);
6999
7000         ret = btrfs_reserve_extent(root, blocksize, blocksize,
7001                                    empty_size, hint, &ins, 0);
7002         if (ret) {
7003                 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7004                 return ERR_PTR(ret);
7005         }
7006
7007         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7008                                     blocksize, level);
7009         BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7010
7011         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7012                 if (parent == 0)
7013                         parent = ins.objectid;
7014                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7015         } else
7016                 BUG_ON(parent > 0);
7017
7018         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7019                 struct btrfs_delayed_extent_op *extent_op;
7020                 extent_op = btrfs_alloc_delayed_extent_op();
7021                 BUG_ON(!extent_op); /* -ENOMEM */
7022                 if (key)
7023                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
7024                 else
7025                         memset(&extent_op->key, 0, sizeof(extent_op->key));
7026                 extent_op->flags_to_set = flags;
7027                 if (skinny_metadata)
7028                         extent_op->update_key = 0;
7029                 else
7030                         extent_op->update_key = 1;
7031                 extent_op->update_flags = 1;
7032                 extent_op->is_data = 0;
7033                 extent_op->level = level;
7034
7035                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7036                                         ins.objectid,
7037                                         ins.offset, parent, root_objectid,
7038                                         level, BTRFS_ADD_DELAYED_EXTENT,
7039                                         extent_op, 0);
7040                 BUG_ON(ret); /* -ENOMEM */
7041         }
7042         return buf;
7043 }
7044
7045 struct walk_control {
7046         u64 refs[BTRFS_MAX_LEVEL];
7047         u64 flags[BTRFS_MAX_LEVEL];
7048         struct btrfs_key update_progress;
7049         int stage;
7050         int level;
7051         int shared_level;
7052         int update_ref;
7053         int keep_locks;
7054         int reada_slot;
7055         int reada_count;
7056         int for_reloc;
7057 };
7058
7059 #define DROP_REFERENCE  1
7060 #define UPDATE_BACKREF  2
7061
7062 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7063                                      struct btrfs_root *root,
7064                                      struct walk_control *wc,
7065                                      struct btrfs_path *path)
7066 {
7067         u64 bytenr;
7068         u64 generation;
7069         u64 refs;
7070         u64 flags;
7071         u32 nritems;
7072         u32 blocksize;
7073         struct btrfs_key key;
7074         struct extent_buffer *eb;
7075         int ret;
7076         int slot;
7077         int nread = 0;
7078
7079         if (path->slots[wc->level] < wc->reada_slot) {
7080                 wc->reada_count = wc->reada_count * 2 / 3;
7081                 wc->reada_count = max(wc->reada_count, 2);
7082         } else {
7083                 wc->reada_count = wc->reada_count * 3 / 2;
7084                 wc->reada_count = min_t(int, wc->reada_count,
7085                                         BTRFS_NODEPTRS_PER_BLOCK(root));
7086         }
7087
7088         eb = path->nodes[wc->level];
7089         nritems = btrfs_header_nritems(eb);
7090         blocksize = btrfs_level_size(root, wc->level - 1);
7091
7092         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7093                 if (nread >= wc->reada_count)
7094                         break;
7095
7096                 cond_resched();
7097                 bytenr = btrfs_node_blockptr(eb, slot);
7098                 generation = btrfs_node_ptr_generation(eb, slot);
7099
7100                 if (slot == path->slots[wc->level])
7101                         goto reada;
7102
7103                 if (wc->stage == UPDATE_BACKREF &&
7104                     generation <= root->root_key.offset)
7105                         continue;
7106
7107                 /* We don't lock the tree block, it's OK to be racy here */
7108                 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7109                                                wc->level - 1, 1, &refs,
7110                                                &flags);
7111                 /* We don't care about errors in readahead. */
7112                 if (ret < 0)
7113                         continue;
7114                 BUG_ON(refs == 0);
7115
7116                 if (wc->stage == DROP_REFERENCE) {
7117                         if (refs == 1)
7118                                 goto reada;
7119
7120                         if (wc->level == 1 &&
7121                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7122                                 continue;
7123                         if (!wc->update_ref ||
7124                             generation <= root->root_key.offset)
7125                                 continue;
7126                         btrfs_node_key_to_cpu(eb, &key, slot);
7127                         ret = btrfs_comp_cpu_keys(&key,
7128                                                   &wc->update_progress);
7129                         if (ret < 0)
7130                                 continue;
7131                 } else {
7132                         if (wc->level == 1 &&
7133                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7134                                 continue;
7135                 }
7136 reada:
7137                 ret = readahead_tree_block(root, bytenr, blocksize,
7138                                            generation);
7139                 if (ret)
7140                         break;
7141                 nread++;
7142         }
7143         wc->reada_slot = slot;
7144 }
7145
7146 /*
7147  * helper to process tree block while walking down the tree.
7148  *
7149  * when wc->stage == UPDATE_BACKREF, this function updates
7150  * back refs for pointers in the block.
7151  *
7152  * NOTE: return value 1 means we should stop walking down.
7153  */
7154 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7155                                    struct btrfs_root *root,
7156                                    struct btrfs_path *path,
7157                                    struct walk_control *wc, int lookup_info)
7158 {
7159         int level = wc->level;
7160         struct extent_buffer *eb = path->nodes[level];
7161         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7162         int ret;
7163
7164         if (wc->stage == UPDATE_BACKREF &&
7165             btrfs_header_owner(eb) != root->root_key.objectid)
7166                 return 1;
7167
7168         /*
7169          * when reference count of tree block is 1, it won't increase
7170          * again. once full backref flag is set, we never clear it.
7171          */
7172         if (lookup_info &&
7173             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7174              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7175                 BUG_ON(!path->locks[level]);
7176                 ret = btrfs_lookup_extent_info(trans, root,
7177                                                eb->start, level, 1,
7178                                                &wc->refs[level],
7179                                                &wc->flags[level]);
7180                 BUG_ON(ret == -ENOMEM);
7181                 if (ret)
7182                         return ret;
7183                 BUG_ON(wc->refs[level] == 0);
7184         }
7185
7186         if (wc->stage == DROP_REFERENCE) {
7187                 if (wc->refs[level] > 1)
7188                         return 1;
7189
7190                 if (path->locks[level] && !wc->keep_locks) {
7191                         btrfs_tree_unlock_rw(eb, path->locks[level]);
7192                         path->locks[level] = 0;
7193                 }
7194                 return 0;
7195         }
7196
7197         /* wc->stage == UPDATE_BACKREF */
7198         if (!(wc->flags[level] & flag)) {
7199                 BUG_ON(!path->locks[level]);
7200                 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7201                 BUG_ON(ret); /* -ENOMEM */
7202                 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7203                 BUG_ON(ret); /* -ENOMEM */
7204                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7205                                                   eb->len, flag,
7206                                                   btrfs_header_level(eb), 0);
7207                 BUG_ON(ret); /* -ENOMEM */
7208                 wc->flags[level] |= flag;
7209         }
7210
7211         /*
7212          * the block is shared by multiple trees, so it's not good to
7213          * keep the tree lock
7214          */
7215         if (path->locks[level] && level > 0) {
7216                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7217                 path->locks[level] = 0;
7218         }
7219         return 0;
7220 }
7221
7222 /*
7223  * helper to process tree block pointer.
7224  *
7225  * when wc->stage == DROP_REFERENCE, this function checks
7226  * reference count of the block pointed to. if the block
7227  * is shared and we need update back refs for the subtree
7228  * rooted at the block, this function changes wc->stage to
7229  * UPDATE_BACKREF. if the block is shared and there is no
7230  * need to update back, this function drops the reference
7231  * to the block.
7232  *
7233  * NOTE: return value 1 means we should stop walking down.
7234  */
7235 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7236                                  struct btrfs_root *root,
7237                                  struct btrfs_path *path,
7238                                  struct walk_control *wc, int *lookup_info)
7239 {
7240         u64 bytenr;
7241         u64 generation;
7242         u64 parent;
7243         u32 blocksize;
7244         struct btrfs_key key;
7245         struct extent_buffer *next;
7246         int level = wc->level;
7247         int reada = 0;
7248         int ret = 0;
7249
7250         generation = btrfs_node_ptr_generation(path->nodes[level],
7251                                                path->slots[level]);
7252         /*
7253          * if the lower level block was created before the snapshot
7254          * was created, we know there is no need to update back refs
7255          * for the subtree
7256          */
7257         if (wc->stage == UPDATE_BACKREF &&
7258             generation <= root->root_key.offset) {
7259                 *lookup_info = 1;
7260                 return 1;
7261         }
7262
7263         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7264         blocksize = btrfs_level_size(root, level - 1);
7265
7266         next = btrfs_find_tree_block(root, bytenr, blocksize);
7267         if (!next) {
7268                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7269                 if (!next)
7270                         return -ENOMEM;
7271                 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7272                                                level - 1);
7273                 reada = 1;
7274         }
7275         btrfs_tree_lock(next);
7276         btrfs_set_lock_blocking(next);
7277
7278         ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7279                                        &wc->refs[level - 1],
7280                                        &wc->flags[level - 1]);
7281         if (ret < 0) {
7282                 btrfs_tree_unlock(next);
7283                 return ret;
7284         }
7285
7286         if (unlikely(wc->refs[level - 1] == 0)) {
7287                 btrfs_err(root->fs_info, "Missing references.");
7288                 BUG();
7289         }
7290         *lookup_info = 0;
7291
7292         if (wc->stage == DROP_REFERENCE) {
7293                 if (wc->refs[level - 1] > 1) {
7294                         if (level == 1 &&
7295                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7296                                 goto skip;
7297
7298                         if (!wc->update_ref ||
7299                             generation <= root->root_key.offset)
7300                                 goto skip;
7301
7302                         btrfs_node_key_to_cpu(path->nodes[level], &key,
7303                                               path->slots[level]);
7304                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7305                         if (ret < 0)
7306                                 goto skip;
7307
7308                         wc->stage = UPDATE_BACKREF;
7309                         wc->shared_level = level - 1;
7310                 }
7311         } else {
7312                 if (level == 1 &&
7313                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7314                         goto skip;
7315         }
7316
7317         if (!btrfs_buffer_uptodate(next, generation, 0)) {
7318                 btrfs_tree_unlock(next);
7319                 free_extent_buffer(next);
7320                 next = NULL;
7321                 *lookup_info = 1;
7322         }
7323
7324         if (!next) {
7325                 if (reada && level == 1)
7326                         reada_walk_down(trans, root, wc, path);
7327                 next = read_tree_block(root, bytenr, blocksize, generation);
7328                 if (!next || !extent_buffer_uptodate(next)) {
7329                         free_extent_buffer(next);
7330                         return -EIO;
7331                 }
7332                 btrfs_tree_lock(next);
7333                 btrfs_set_lock_blocking(next);
7334         }
7335
7336         level--;
7337         BUG_ON(level != btrfs_header_level(next));
7338         path->nodes[level] = next;
7339         path->slots[level] = 0;
7340         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7341         wc->level = level;
7342         if (wc->level == 1)
7343                 wc->reada_slot = 0;
7344         return 0;
7345 skip:
7346         wc->refs[level - 1] = 0;
7347         wc->flags[level - 1] = 0;
7348         if (wc->stage == DROP_REFERENCE) {
7349                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7350                         parent = path->nodes[level]->start;
7351                 } else {
7352                         BUG_ON(root->root_key.objectid !=
7353                                btrfs_header_owner(path->nodes[level]));
7354                         parent = 0;
7355                 }
7356
7357                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7358                                 root->root_key.objectid, level - 1, 0, 0);
7359                 BUG_ON(ret); /* -ENOMEM */
7360         }
7361         btrfs_tree_unlock(next);
7362         free_extent_buffer(next);
7363         *lookup_info = 1;
7364         return 1;
7365 }
7366
7367 /*
7368  * helper to process tree block while walking up the tree.
7369  *
7370  * when wc->stage == DROP_REFERENCE, this function drops
7371  * reference count on the block.
7372  *
7373  * when wc->stage == UPDATE_BACKREF, this function changes
7374  * wc->stage back to DROP_REFERENCE if we changed wc->stage
7375  * to UPDATE_BACKREF previously while processing the block.
7376  *
7377  * NOTE: return value 1 means we should stop walking up.
7378  */
7379 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7380                                  struct btrfs_root *root,
7381                                  struct btrfs_path *path,
7382                                  struct walk_control *wc)
7383 {
7384         int ret;
7385         int level = wc->level;
7386         struct extent_buffer *eb = path->nodes[level];
7387         u64 parent = 0;
7388
7389         if (wc->stage == UPDATE_BACKREF) {
7390                 BUG_ON(wc->shared_level < level);
7391                 if (level < wc->shared_level)
7392                         goto out;
7393
7394                 ret = find_next_key(path, level + 1, &wc->update_progress);
7395                 if (ret > 0)
7396                         wc->update_ref = 0;
7397
7398                 wc->stage = DROP_REFERENCE;
7399                 wc->shared_level = -1;
7400                 path->slots[level] = 0;
7401
7402                 /*
7403                  * check reference count again if the block isn't locked.
7404                  * we should start walking down the tree again if reference
7405                  * count is one.
7406                  */
7407                 if (!path->locks[level]) {
7408                         BUG_ON(level == 0);
7409                         btrfs_tree_lock(eb);
7410                         btrfs_set_lock_blocking(eb);
7411                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7412
7413                         ret = btrfs_lookup_extent_info(trans, root,
7414                                                        eb->start, level, 1,
7415                                                        &wc->refs[level],
7416                                                        &wc->flags[level]);
7417                         if (ret < 0) {
7418                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7419                                 path->locks[level] = 0;
7420                                 return ret;
7421                         }
7422                         BUG_ON(wc->refs[level] == 0);
7423                         if (wc->refs[level] == 1) {
7424                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7425                                 path->locks[level] = 0;
7426                                 return 1;
7427                         }
7428                 }
7429         }
7430
7431         /* wc->stage == DROP_REFERENCE */
7432         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7433
7434         if (wc->refs[level] == 1) {
7435                 if (level == 0) {
7436                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7437                                 ret = btrfs_dec_ref(trans, root, eb, 1,
7438                                                     wc->for_reloc);
7439                         else
7440                                 ret = btrfs_dec_ref(trans, root, eb, 0,
7441                                                     wc->for_reloc);
7442                         BUG_ON(ret); /* -ENOMEM */
7443                 }
7444                 /* make block locked assertion in clean_tree_block happy */
7445                 if (!path->locks[level] &&
7446                     btrfs_header_generation(eb) == trans->transid) {
7447                         btrfs_tree_lock(eb);
7448                         btrfs_set_lock_blocking(eb);
7449                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7450                 }
7451                 clean_tree_block(trans, root, eb);
7452         }
7453
7454         if (eb == root->node) {
7455                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7456                         parent = eb->start;
7457                 else
7458                         BUG_ON(root->root_key.objectid !=
7459                                btrfs_header_owner(eb));
7460         } else {
7461                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7462                         parent = path->nodes[level + 1]->start;
7463                 else
7464                         BUG_ON(root->root_key.objectid !=
7465                                btrfs_header_owner(path->nodes[level + 1]));
7466         }
7467
7468         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7469 out:
7470         wc->refs[level] = 0;
7471         wc->flags[level] = 0;
7472         return 0;
7473 }
7474
7475 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7476                                    struct btrfs_root *root,
7477                                    struct btrfs_path *path,
7478                                    struct walk_control *wc)
7479 {
7480         int level = wc->level;
7481         int lookup_info = 1;
7482         int ret;
7483
7484         while (level >= 0) {
7485                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7486                 if (ret > 0)
7487                         break;
7488
7489                 if (level == 0)
7490                         break;
7491
7492                 if (path->slots[level] >=
7493                     btrfs_header_nritems(path->nodes[level]))
7494                         break;
7495
7496                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7497                 if (ret > 0) {
7498                         path->slots[level]++;
7499                         continue;
7500                 } else if (ret < 0)
7501                         return ret;
7502                 level = wc->level;
7503         }
7504         return 0;
7505 }
7506
7507 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7508                                  struct btrfs_root *root,
7509                                  struct btrfs_path *path,
7510                                  struct walk_control *wc, int max_level)
7511 {
7512         int level = wc->level;
7513         int ret;
7514
7515         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7516         while (level < max_level && path->nodes[level]) {
7517                 wc->level = level;
7518                 if (path->slots[level] + 1 <
7519                     btrfs_header_nritems(path->nodes[level])) {
7520                         path->slots[level]++;
7521                         return 0;
7522                 } else {
7523                         ret = walk_up_proc(trans, root, path, wc);
7524                         if (ret > 0)
7525                                 return 0;
7526
7527                         if (path->locks[level]) {
7528                                 btrfs_tree_unlock_rw(path->nodes[level],
7529                                                      path->locks[level]);
7530                                 path->locks[level] = 0;
7531                         }
7532                         free_extent_buffer(path->nodes[level]);
7533                         path->nodes[level] = NULL;
7534                         level++;
7535                 }
7536         }
7537         return 1;
7538 }
7539
7540 /*
7541  * drop a subvolume tree.
7542  *
7543  * this function traverses the tree freeing any blocks that only
7544  * referenced by the tree.
7545  *
7546  * when a shared tree block is found. this function decreases its
7547  * reference count by one. if update_ref is true, this function
7548  * also make sure backrefs for the shared block and all lower level
7549  * blocks are properly updated.
7550  *
7551  * If called with for_reloc == 0, may exit early with -EAGAIN
7552  */
7553 int btrfs_drop_snapshot(struct btrfs_root *root,
7554                          struct btrfs_block_rsv *block_rsv, int update_ref,
7555                          int for_reloc)
7556 {
7557         struct btrfs_path *path;
7558         struct btrfs_trans_handle *trans;
7559         struct btrfs_root *tree_root = root->fs_info->tree_root;
7560         struct btrfs_root_item *root_item = &root->root_item;
7561         struct walk_control *wc;
7562         struct btrfs_key key;
7563         int err = 0;
7564         int ret;
7565         int level;
7566         bool root_dropped = false;
7567
7568         path = btrfs_alloc_path();
7569         if (!path) {
7570                 err = -ENOMEM;
7571                 goto out;
7572         }
7573
7574         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7575         if (!wc) {
7576                 btrfs_free_path(path);
7577                 err = -ENOMEM;
7578                 goto out;
7579         }
7580
7581         trans = btrfs_start_transaction(tree_root, 0);
7582         if (IS_ERR(trans)) {
7583                 err = PTR_ERR(trans);
7584                 goto out_free;
7585         }
7586
7587         if (block_rsv)
7588                 trans->block_rsv = block_rsv;
7589
7590         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7591                 level = btrfs_header_level(root->node);
7592                 path->nodes[level] = btrfs_lock_root_node(root);
7593                 btrfs_set_lock_blocking(path->nodes[level]);
7594                 path->slots[level] = 0;
7595                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7596                 memset(&wc->update_progress, 0,
7597                        sizeof(wc->update_progress));
7598         } else {
7599                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7600                 memcpy(&wc->update_progress, &key,
7601                        sizeof(wc->update_progress));
7602
7603                 level = root_item->drop_level;
7604                 BUG_ON(level == 0);
7605                 path->lowest_level = level;
7606                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7607                 path->lowest_level = 0;
7608                 if (ret < 0) {
7609                         err = ret;
7610                         goto out_end_trans;
7611                 }
7612                 WARN_ON(ret > 0);
7613
7614                 /*
7615                  * unlock our path, this is safe because only this
7616                  * function is allowed to delete this snapshot
7617                  */
7618                 btrfs_unlock_up_safe(path, 0);
7619
7620                 level = btrfs_header_level(root->node);
7621                 while (1) {
7622                         btrfs_tree_lock(path->nodes[level]);
7623                         btrfs_set_lock_blocking(path->nodes[level]);
7624                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7625
7626                         ret = btrfs_lookup_extent_info(trans, root,
7627                                                 path->nodes[level]->start,
7628                                                 level, 1, &wc->refs[level],
7629                                                 &wc->flags[level]);
7630                         if (ret < 0) {
7631                                 err = ret;
7632                                 goto out_end_trans;
7633                         }
7634                         BUG_ON(wc->refs[level] == 0);
7635
7636                         if (level == root_item->drop_level)
7637                                 break;
7638
7639                         btrfs_tree_unlock(path->nodes[level]);
7640                         path->locks[level] = 0;
7641                         WARN_ON(wc->refs[level] != 1);
7642                         level--;
7643                 }
7644         }
7645
7646         wc->level = level;
7647         wc->shared_level = -1;
7648         wc->stage = DROP_REFERENCE;
7649         wc->update_ref = update_ref;
7650         wc->keep_locks = 0;
7651         wc->for_reloc = for_reloc;
7652         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7653
7654         while (1) {
7655
7656                 ret = walk_down_tree(trans, root, path, wc);
7657                 if (ret < 0) {
7658                         err = ret;
7659                         break;
7660                 }
7661
7662                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7663                 if (ret < 0) {
7664                         err = ret;
7665                         break;
7666                 }
7667
7668                 if (ret > 0) {
7669                         BUG_ON(wc->stage != DROP_REFERENCE);
7670                         break;
7671                 }
7672
7673                 if (wc->stage == DROP_REFERENCE) {
7674                         level = wc->level;
7675                         btrfs_node_key(path->nodes[level],
7676                                        &root_item->drop_progress,
7677                                        path->slots[level]);
7678                         root_item->drop_level = level;
7679                 }
7680
7681                 BUG_ON(wc->level == 0);
7682                 if (btrfs_should_end_transaction(trans, tree_root) ||
7683                     (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7684                         ret = btrfs_update_root(trans, tree_root,
7685                                                 &root->root_key,
7686                                                 root_item);
7687                         if (ret) {
7688                                 btrfs_abort_transaction(trans, tree_root, ret);
7689                                 err = ret;
7690                                 goto out_end_trans;
7691                         }
7692
7693                         btrfs_end_transaction_throttle(trans, tree_root);
7694                         if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7695                                 pr_debug("btrfs: drop snapshot early exit\n");
7696                                 err = -EAGAIN;
7697                                 goto out_free;
7698                         }
7699
7700                         trans = btrfs_start_transaction(tree_root, 0);
7701                         if (IS_ERR(trans)) {
7702                                 err = PTR_ERR(trans);
7703                                 goto out_free;
7704                         }
7705                         if (block_rsv)
7706                                 trans->block_rsv = block_rsv;
7707                 }
7708         }
7709         btrfs_release_path(path);
7710         if (err)
7711                 goto out_end_trans;
7712
7713         ret = btrfs_del_root(trans, tree_root, &root->root_key);
7714         if (ret) {
7715                 btrfs_abort_transaction(trans, tree_root, ret);
7716                 goto out_end_trans;
7717         }
7718
7719         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7720                 ret = btrfs_find_root(tree_root, &root->root_key, path,
7721                                       NULL, NULL);
7722                 if (ret < 0) {
7723                         btrfs_abort_transaction(trans, tree_root, ret);
7724                         err = ret;
7725                         goto out_end_trans;
7726                 } else if (ret > 0) {
7727                         /* if we fail to delete the orphan item this time
7728                          * around, it'll get picked up the next time.
7729                          *
7730                          * The most common failure here is just -ENOENT.
7731                          */
7732                         btrfs_del_orphan_item(trans, tree_root,
7733                                               root->root_key.objectid);
7734                 }
7735         }
7736
7737         if (root->in_radix) {
7738                 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7739         } else {
7740                 free_extent_buffer(root->node);
7741                 free_extent_buffer(root->commit_root);
7742                 btrfs_put_fs_root(root);
7743         }
7744         root_dropped = true;
7745 out_end_trans:
7746         btrfs_end_transaction_throttle(trans, tree_root);
7747 out_free:
7748         kfree(wc);
7749         btrfs_free_path(path);
7750 out:
7751         /*
7752          * So if we need to stop dropping the snapshot for whatever reason we
7753          * need to make sure to add it back to the dead root list so that we
7754          * keep trying to do the work later.  This also cleans up roots if we
7755          * don't have it in the radix (like when we recover after a power fail
7756          * or unmount) so we don't leak memory.
7757          */
7758         if (!for_reloc && root_dropped == false)
7759                 btrfs_add_dead_root(root);
7760         if (err)
7761                 btrfs_std_error(root->fs_info, err);
7762         return err;
7763 }
7764
7765 /*
7766  * drop subtree rooted at tree block 'node'.
7767  *
7768  * NOTE: this function will unlock and release tree block 'node'
7769  * only used by relocation code
7770  */
7771 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7772                         struct btrfs_root *root,
7773                         struct extent_buffer *node,
7774                         struct extent_buffer *parent)
7775 {
7776         struct btrfs_path *path;
7777         struct walk_control *wc;
7778         int level;
7779         int parent_level;
7780         int ret = 0;
7781         int wret;
7782
7783         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7784
7785         path = btrfs_alloc_path();
7786         if (!path)
7787                 return -ENOMEM;
7788
7789         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7790         if (!wc) {
7791                 btrfs_free_path(path);
7792                 return -ENOMEM;
7793         }
7794
7795         btrfs_assert_tree_locked(parent);
7796         parent_level = btrfs_header_level(parent);
7797         extent_buffer_get(parent);
7798         path->nodes[parent_level] = parent;
7799         path->slots[parent_level] = btrfs_header_nritems(parent);
7800
7801         btrfs_assert_tree_locked(node);
7802         level = btrfs_header_level(node);
7803         path->nodes[level] = node;
7804         path->slots[level] = 0;
7805         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7806
7807         wc->refs[parent_level] = 1;
7808         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7809         wc->level = level;
7810         wc->shared_level = -1;
7811         wc->stage = DROP_REFERENCE;
7812         wc->update_ref = 0;
7813         wc->keep_locks = 1;
7814         wc->for_reloc = 1;
7815         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7816
7817         while (1) {
7818                 wret = walk_down_tree(trans, root, path, wc);
7819                 if (wret < 0) {
7820                         ret = wret;
7821                         break;
7822                 }
7823
7824                 wret = walk_up_tree(trans, root, path, wc, parent_level);
7825                 if (wret < 0)
7826                         ret = wret;
7827                 if (wret != 0)
7828                         break;
7829         }
7830
7831         kfree(wc);
7832         btrfs_free_path(path);
7833         return ret;
7834 }
7835
7836 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7837 {
7838         u64 num_devices;
7839         u64 stripped;
7840
7841         /*
7842          * if restripe for this chunk_type is on pick target profile and
7843          * return, otherwise do the usual balance
7844          */
7845         stripped = get_restripe_target(root->fs_info, flags);
7846         if (stripped)
7847                 return extended_to_chunk(stripped);
7848
7849         /*
7850          * we add in the count of missing devices because we want
7851          * to make sure that any RAID levels on a degraded FS
7852          * continue to be honored.
7853          */
7854         num_devices = root->fs_info->fs_devices->rw_devices +
7855                 root->fs_info->fs_devices->missing_devices;
7856
7857         stripped = BTRFS_BLOCK_GROUP_RAID0 |
7858                 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7859                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7860
7861         if (num_devices == 1) {
7862                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7863                 stripped = flags & ~stripped;
7864
7865                 /* turn raid0 into single device chunks */
7866                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7867                         return stripped;
7868
7869                 /* turn mirroring into duplication */
7870                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7871                              BTRFS_BLOCK_GROUP_RAID10))
7872                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7873         } else {
7874                 /* they already had raid on here, just return */
7875                 if (flags & stripped)
7876                         return flags;
7877
7878                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7879                 stripped = flags & ~stripped;
7880
7881                 /* switch duplicated blocks with raid1 */
7882                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7883                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7884
7885                 /* this is drive concat, leave it alone */
7886         }
7887
7888         return flags;
7889 }
7890
7891 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7892 {
7893         struct btrfs_space_info *sinfo = cache->space_info;
7894         u64 num_bytes;
7895         u64 min_allocable_bytes;
7896         int ret = -ENOSPC;
7897
7898
7899         /*
7900          * We need some metadata space and system metadata space for
7901          * allocating chunks in some corner cases until we force to set
7902          * it to be readonly.
7903          */
7904         if ((sinfo->flags &
7905              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7906             !force)
7907                 min_allocable_bytes = 1 * 1024 * 1024;
7908         else
7909                 min_allocable_bytes = 0;
7910
7911         spin_lock(&sinfo->lock);
7912         spin_lock(&cache->lock);
7913
7914         if (cache->ro) {
7915                 ret = 0;
7916                 goto out;
7917         }
7918
7919         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7920                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7921
7922         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7923             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7924             min_allocable_bytes <= sinfo->total_bytes) {
7925                 sinfo->bytes_readonly += num_bytes;
7926                 cache->ro = 1;
7927                 ret = 0;
7928         }
7929 out:
7930         spin_unlock(&cache->lock);
7931         spin_unlock(&sinfo->lock);
7932         return ret;
7933 }
7934
7935 int btrfs_set_block_group_ro(struct btrfs_root *root,
7936                              struct btrfs_block_group_cache *cache)
7937
7938 {
7939         struct btrfs_trans_handle *trans;
7940         u64 alloc_flags;
7941         int ret;
7942
7943         BUG_ON(cache->ro);
7944
7945         trans = btrfs_join_transaction(root);
7946         if (IS_ERR(trans))
7947                 return PTR_ERR(trans);
7948
7949         alloc_flags = update_block_group_flags(root, cache->flags);
7950         if (alloc_flags != cache->flags) {
7951                 ret = do_chunk_alloc(trans, root, alloc_flags,
7952                                      CHUNK_ALLOC_FORCE);
7953                 if (ret < 0)
7954                         goto out;
7955         }
7956
7957         ret = set_block_group_ro(cache, 0);
7958         if (!ret)
7959                 goto out;
7960         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7961         ret = do_chunk_alloc(trans, root, alloc_flags,
7962                              CHUNK_ALLOC_FORCE);
7963         if (ret < 0)
7964                 goto out;
7965         ret = set_block_group_ro(cache, 0);
7966 out:
7967         btrfs_end_transaction(trans, root);
7968         return ret;
7969 }
7970
7971 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7972                             struct btrfs_root *root, u64 type)
7973 {
7974         u64 alloc_flags = get_alloc_profile(root, type);
7975         return do_chunk_alloc(trans, root, alloc_flags,
7976                               CHUNK_ALLOC_FORCE);
7977 }
7978
7979 /*
7980  * helper to account the unused space of all the readonly block group in the
7981  * list. takes mirrors into account.
7982  */
7983 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7984 {
7985         struct btrfs_block_group_cache *block_group;
7986         u64 free_bytes = 0;
7987         int factor;
7988
7989         list_for_each_entry(block_group, groups_list, list) {
7990                 spin_lock(&block_group->lock);
7991
7992                 if (!block_group->ro) {
7993                         spin_unlock(&block_group->lock);
7994                         continue;
7995                 }
7996
7997                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7998                                           BTRFS_BLOCK_GROUP_RAID10 |
7999                                           BTRFS_BLOCK_GROUP_DUP))
8000                         factor = 2;
8001                 else
8002                         factor = 1;
8003
8004                 free_bytes += (block_group->key.offset -
8005                                btrfs_block_group_used(&block_group->item)) *
8006                                factor;
8007
8008                 spin_unlock(&block_group->lock);
8009         }
8010
8011         return free_bytes;
8012 }
8013
8014 /*
8015  * helper to account the unused space of all the readonly block group in the
8016  * space_info. takes mirrors into account.
8017  */
8018 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8019 {
8020         int i;
8021         u64 free_bytes = 0;
8022
8023         spin_lock(&sinfo->lock);
8024
8025         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8026                 if (!list_empty(&sinfo->block_groups[i]))
8027                         free_bytes += __btrfs_get_ro_block_group_free_space(
8028                                                 &sinfo->block_groups[i]);
8029
8030         spin_unlock(&sinfo->lock);
8031
8032         return free_bytes;
8033 }
8034
8035 void btrfs_set_block_group_rw(struct btrfs_root *root,
8036                               struct btrfs_block_group_cache *cache)
8037 {
8038         struct btrfs_space_info *sinfo = cache->space_info;
8039         u64 num_bytes;
8040
8041         BUG_ON(!cache->ro);
8042
8043         spin_lock(&sinfo->lock);
8044         spin_lock(&cache->lock);
8045         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8046                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8047         sinfo->bytes_readonly -= num_bytes;
8048         cache->ro = 0;
8049         spin_unlock(&cache->lock);
8050         spin_unlock(&sinfo->lock);
8051 }
8052
8053 /*
8054  * checks to see if its even possible to relocate this block group.
8055  *
8056  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8057  * ok to go ahead and try.
8058  */
8059 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8060 {
8061         struct btrfs_block_group_cache *block_group;
8062         struct btrfs_space_info *space_info;
8063         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8064         struct btrfs_device *device;
8065         struct btrfs_trans_handle *trans;
8066         u64 min_free;
8067         u64 dev_min = 1;
8068         u64 dev_nr = 0;
8069         u64 target;
8070         int index;
8071         int full = 0;
8072         int ret = 0;
8073
8074         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8075
8076         /* odd, couldn't find the block group, leave it alone */
8077         if (!block_group)
8078                 return -1;
8079
8080         min_free = btrfs_block_group_used(&block_group->item);
8081
8082         /* no bytes used, we're good */
8083         if (!min_free)
8084                 goto out;
8085
8086         space_info = block_group->space_info;
8087         spin_lock(&space_info->lock);
8088
8089         full = space_info->full;
8090
8091         /*
8092          * if this is the last block group we have in this space, we can't
8093          * relocate it unless we're able to allocate a new chunk below.
8094          *
8095          * Otherwise, we need to make sure we have room in the space to handle
8096          * all of the extents from this block group.  If we can, we're good
8097          */
8098         if ((space_info->total_bytes != block_group->key.offset) &&
8099             (space_info->bytes_used + space_info->bytes_reserved +
8100              space_info->bytes_pinned + space_info->bytes_readonly +
8101              min_free < space_info->total_bytes)) {
8102                 spin_unlock(&space_info->lock);
8103                 goto out;
8104         }
8105         spin_unlock(&space_info->lock);
8106
8107         /*
8108          * ok we don't have enough space, but maybe we have free space on our
8109          * devices to allocate new chunks for relocation, so loop through our
8110          * alloc devices and guess if we have enough space.  if this block
8111          * group is going to be restriped, run checks against the target
8112          * profile instead of the current one.
8113          */
8114         ret = -1;
8115
8116         /*
8117          * index:
8118          *      0: raid10
8119          *      1: raid1
8120          *      2: dup
8121          *      3: raid0
8122          *      4: single
8123          */
8124         target = get_restripe_target(root->fs_info, block_group->flags);
8125         if (target) {
8126                 index = __get_raid_index(extended_to_chunk(target));
8127         } else {
8128                 /*
8129                  * this is just a balance, so if we were marked as full
8130                  * we know there is no space for a new chunk
8131                  */
8132                 if (full)
8133                         goto out;
8134
8135                 index = get_block_group_index(block_group);
8136         }
8137
8138         if (index == BTRFS_RAID_RAID10) {
8139                 dev_min = 4;
8140                 /* Divide by 2 */
8141                 min_free >>= 1;
8142         } else if (index == BTRFS_RAID_RAID1) {
8143                 dev_min = 2;
8144         } else if (index == BTRFS_RAID_DUP) {
8145                 /* Multiply by 2 */
8146                 min_free <<= 1;
8147         } else if (index == BTRFS_RAID_RAID0) {
8148                 dev_min = fs_devices->rw_devices;
8149                 do_div(min_free, dev_min);
8150         }
8151
8152         /* We need to do this so that we can look at pending chunks */
8153         trans = btrfs_join_transaction(root);
8154         if (IS_ERR(trans)) {
8155                 ret = PTR_ERR(trans);
8156                 goto out;
8157         }
8158
8159         mutex_lock(&root->fs_info->chunk_mutex);
8160         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8161                 u64 dev_offset;
8162
8163                 /*
8164                  * check to make sure we can actually find a chunk with enough
8165                  * space to fit our block group in.
8166                  */
8167                 if (device->total_bytes > device->bytes_used + min_free &&
8168                     !device->is_tgtdev_for_dev_replace) {
8169                         ret = find_free_dev_extent(trans, device, min_free,
8170                                                    &dev_offset, NULL);
8171                         if (!ret)
8172                                 dev_nr++;
8173
8174                         if (dev_nr >= dev_min)
8175                                 break;
8176
8177                         ret = -1;
8178                 }
8179         }
8180         mutex_unlock(&root->fs_info->chunk_mutex);
8181         btrfs_end_transaction(trans, root);
8182 out:
8183         btrfs_put_block_group(block_group);
8184         return ret;
8185 }
8186
8187 static int find_first_block_group(struct btrfs_root *root,
8188                 struct btrfs_path *path, struct btrfs_key *key)
8189 {
8190         int ret = 0;
8191         struct btrfs_key found_key;
8192         struct extent_buffer *leaf;
8193         int slot;
8194
8195         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8196         if (ret < 0)
8197                 goto out;
8198
8199         while (1) {
8200                 slot = path->slots[0];
8201                 leaf = path->nodes[0];
8202                 if (slot >= btrfs_header_nritems(leaf)) {
8203                         ret = btrfs_next_leaf(root, path);
8204                         if (ret == 0)
8205                                 continue;
8206                         if (ret < 0)
8207                                 goto out;
8208                         break;
8209                 }
8210                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8211
8212                 if (found_key.objectid >= key->objectid &&
8213                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8214                         ret = 0;
8215                         goto out;
8216                 }
8217                 path->slots[0]++;
8218         }
8219 out:
8220         return ret;
8221 }
8222
8223 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8224 {
8225         struct btrfs_block_group_cache *block_group;
8226         u64 last = 0;
8227
8228         while (1) {
8229                 struct inode *inode;
8230
8231                 block_group = btrfs_lookup_first_block_group(info, last);
8232                 while (block_group) {
8233                         spin_lock(&block_group->lock);
8234                         if (block_group->iref)
8235                                 break;
8236                         spin_unlock(&block_group->lock);
8237                         block_group = next_block_group(info->tree_root,
8238                                                        block_group);
8239                 }
8240                 if (!block_group) {
8241                         if (last == 0)
8242                                 break;
8243                         last = 0;
8244                         continue;
8245                 }
8246
8247                 inode = block_group->inode;
8248                 block_group->iref = 0;
8249                 block_group->inode = NULL;
8250                 spin_unlock(&block_group->lock);
8251                 iput(inode);
8252                 last = block_group->key.objectid + block_group->key.offset;
8253                 btrfs_put_block_group(block_group);
8254         }
8255 }
8256
8257 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8258 {
8259         struct btrfs_block_group_cache *block_group;
8260         struct btrfs_space_info *space_info;
8261         struct btrfs_caching_control *caching_ctl;
8262         struct rb_node *n;
8263
8264         down_write(&info->extent_commit_sem);
8265         while (!list_empty(&info->caching_block_groups)) {
8266                 caching_ctl = list_entry(info->caching_block_groups.next,
8267                                          struct btrfs_caching_control, list);
8268                 list_del(&caching_ctl->list);
8269                 put_caching_control(caching_ctl);
8270         }
8271         up_write(&info->extent_commit_sem);
8272
8273         spin_lock(&info->block_group_cache_lock);
8274         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8275                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8276                                        cache_node);
8277                 rb_erase(&block_group->cache_node,
8278                          &info->block_group_cache_tree);
8279                 spin_unlock(&info->block_group_cache_lock);
8280
8281                 down_write(&block_group->space_info->groups_sem);
8282                 list_del(&block_group->list);
8283                 up_write(&block_group->space_info->groups_sem);
8284
8285                 if (block_group->cached == BTRFS_CACHE_STARTED)
8286                         wait_block_group_cache_done(block_group);
8287
8288                 /*
8289                  * We haven't cached this block group, which means we could
8290                  * possibly have excluded extents on this block group.
8291                  */
8292                 if (block_group->cached == BTRFS_CACHE_NO ||
8293                     block_group->cached == BTRFS_CACHE_ERROR)
8294                         free_excluded_extents(info->extent_root, block_group);
8295
8296                 btrfs_remove_free_space_cache(block_group);
8297                 btrfs_put_block_group(block_group);
8298
8299                 spin_lock(&info->block_group_cache_lock);
8300         }
8301         spin_unlock(&info->block_group_cache_lock);
8302
8303         /* now that all the block groups are freed, go through and
8304          * free all the space_info structs.  This is only called during
8305          * the final stages of unmount, and so we know nobody is
8306          * using them.  We call synchronize_rcu() once before we start,
8307          * just to be on the safe side.
8308          */
8309         synchronize_rcu();
8310
8311         release_global_block_rsv(info);
8312
8313         while (!list_empty(&info->space_info)) {
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                 percpu_counter_destroy(&space_info->total_bytes_pinned);
8325                 list_del(&space_info->list);
8326                 kfree(space_info);
8327         }
8328         return 0;
8329 }
8330
8331 static void __link_block_group(struct btrfs_space_info *space_info,
8332                                struct btrfs_block_group_cache *cache)
8333 {
8334         int index = get_block_group_index(cache);
8335
8336         down_write(&space_info->groups_sem);
8337         list_add_tail(&cache->list, &space_info->block_groups[index]);
8338         up_write(&space_info->groups_sem);
8339 }
8340
8341 int btrfs_read_block_groups(struct btrfs_root *root)
8342 {
8343         struct btrfs_path *path;
8344         int ret;
8345         struct btrfs_block_group_cache *cache;
8346         struct btrfs_fs_info *info = root->fs_info;
8347         struct btrfs_space_info *space_info;
8348         struct btrfs_key key;
8349         struct btrfs_key found_key;
8350         struct extent_buffer *leaf;
8351         int need_clear = 0;
8352         u64 cache_gen;
8353
8354         root = info->extent_root;
8355         key.objectid = 0;
8356         key.offset = 0;
8357         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8358         path = btrfs_alloc_path();
8359         if (!path)
8360                 return -ENOMEM;
8361         path->reada = 1;
8362
8363         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8364         if (btrfs_test_opt(root, SPACE_CACHE) &&
8365             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8366                 need_clear = 1;
8367         if (btrfs_test_opt(root, CLEAR_CACHE))
8368                 need_clear = 1;
8369
8370         while (1) {
8371                 ret = find_first_block_group(root, path, &key);
8372                 if (ret > 0)
8373                         break;
8374                 if (ret != 0)
8375                         goto error;
8376                 leaf = path->nodes[0];
8377                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8378                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8379                 if (!cache) {
8380                         ret = -ENOMEM;
8381                         goto error;
8382                 }
8383                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8384                                                 GFP_NOFS);
8385                 if (!cache->free_space_ctl) {
8386                         kfree(cache);
8387                         ret = -ENOMEM;
8388                         goto error;
8389                 }
8390
8391                 atomic_set(&cache->count, 1);
8392                 spin_lock_init(&cache->lock);
8393                 cache->fs_info = info;
8394                 INIT_LIST_HEAD(&cache->list);
8395                 INIT_LIST_HEAD(&cache->cluster_list);
8396
8397                 if (need_clear) {
8398                         /*
8399                          * When we mount with old space cache, we need to
8400                          * set BTRFS_DC_CLEAR and set dirty flag.
8401                          *
8402                          * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8403                          *    truncate the old free space cache inode and
8404                          *    setup a new one.
8405                          * b) Setting 'dirty flag' makes sure that we flush
8406                          *    the new space cache info onto disk.
8407                          */
8408                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8409                         if (btrfs_test_opt(root, SPACE_CACHE))
8410                                 cache->dirty = 1;
8411                 }
8412
8413                 read_extent_buffer(leaf, &cache->item,
8414                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8415                                    sizeof(cache->item));
8416                 memcpy(&cache->key, &found_key, sizeof(found_key));
8417
8418                 key.objectid = found_key.objectid + found_key.offset;
8419                 btrfs_release_path(path);
8420                 cache->flags = btrfs_block_group_flags(&cache->item);
8421                 cache->sectorsize = root->sectorsize;
8422                 cache->full_stripe_len = btrfs_full_stripe_len(root,
8423                                                &root->fs_info->mapping_tree,
8424                                                found_key.objectid);
8425                 btrfs_init_free_space_ctl(cache);
8426
8427                 /*
8428                  * We need to exclude the super stripes now so that the space
8429                  * info has super bytes accounted for, otherwise we'll think
8430                  * we have more space than we actually do.
8431                  */
8432                 ret = exclude_super_stripes(root, cache);
8433                 if (ret) {
8434                         /*
8435                          * We may have excluded something, so call this just in
8436                          * case.
8437                          */
8438                         free_excluded_extents(root, cache);
8439                         kfree(cache->free_space_ctl);
8440                         kfree(cache);
8441                         goto error;
8442                 }
8443
8444                 /*
8445                  * check for two cases, either we are full, and therefore
8446                  * don't need to bother with the caching work since we won't
8447                  * find any space, or we are empty, and we can just add all
8448                  * the space in and be done with it.  This saves us _alot_ of
8449                  * time, particularly in the full case.
8450                  */
8451                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8452                         cache->last_byte_to_unpin = (u64)-1;
8453                         cache->cached = BTRFS_CACHE_FINISHED;
8454                         free_excluded_extents(root, cache);
8455                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8456                         cache->last_byte_to_unpin = (u64)-1;
8457                         cache->cached = BTRFS_CACHE_FINISHED;
8458                         add_new_free_space(cache, root->fs_info,
8459                                            found_key.objectid,
8460                                            found_key.objectid +
8461                                            found_key.offset);
8462                         free_excluded_extents(root, cache);
8463                 }
8464
8465                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8466                 if (ret) {
8467                         btrfs_remove_free_space_cache(cache);
8468                         btrfs_put_block_group(cache);
8469                         goto error;
8470                 }
8471
8472                 ret = update_space_info(info, cache->flags, found_key.offset,
8473                                         btrfs_block_group_used(&cache->item),
8474                                         &space_info);
8475                 if (ret) {
8476                         btrfs_remove_free_space_cache(cache);
8477                         spin_lock(&info->block_group_cache_lock);
8478                         rb_erase(&cache->cache_node,
8479                                  &info->block_group_cache_tree);
8480                         spin_unlock(&info->block_group_cache_lock);
8481                         btrfs_put_block_group(cache);
8482                         goto error;
8483                 }
8484
8485                 cache->space_info = space_info;
8486                 spin_lock(&cache->space_info->lock);
8487                 cache->space_info->bytes_readonly += cache->bytes_super;
8488                 spin_unlock(&cache->space_info->lock);
8489
8490                 __link_block_group(space_info, cache);
8491
8492                 set_avail_alloc_bits(root->fs_info, cache->flags);
8493                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8494                         set_block_group_ro(cache, 1);
8495         }
8496
8497         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8498                 if (!(get_alloc_profile(root, space_info->flags) &
8499                       (BTRFS_BLOCK_GROUP_RAID10 |
8500                        BTRFS_BLOCK_GROUP_RAID1 |
8501                        BTRFS_BLOCK_GROUP_RAID5 |
8502                        BTRFS_BLOCK_GROUP_RAID6 |
8503                        BTRFS_BLOCK_GROUP_DUP)))
8504                         continue;
8505                 /*
8506                  * avoid allocating from un-mirrored block group if there are
8507                  * mirrored block groups.
8508                  */
8509                 list_for_each_entry(cache,
8510                                 &space_info->block_groups[BTRFS_RAID_RAID0],
8511                                 list)
8512                         set_block_group_ro(cache, 1);
8513                 list_for_each_entry(cache,
8514                                 &space_info->block_groups[BTRFS_RAID_SINGLE],
8515                                 list)
8516                         set_block_group_ro(cache, 1);
8517         }
8518
8519         init_global_block_rsv(info);
8520         ret = 0;
8521 error:
8522         btrfs_free_path(path);
8523         return ret;
8524 }
8525
8526 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8527                                        struct btrfs_root *root)
8528 {
8529         struct btrfs_block_group_cache *block_group, *tmp;
8530         struct btrfs_root *extent_root = root->fs_info->extent_root;
8531         struct btrfs_block_group_item item;
8532         struct btrfs_key key;
8533         int ret = 0;
8534
8535         list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8536                                  new_bg_list) {
8537                 list_del_init(&block_group->new_bg_list);
8538
8539                 if (ret)
8540                         continue;
8541
8542                 spin_lock(&block_group->lock);
8543                 memcpy(&item, &block_group->item, sizeof(item));
8544                 memcpy(&key, &block_group->key, sizeof(key));
8545                 spin_unlock(&block_group->lock);
8546
8547                 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8548                                         sizeof(item));
8549                 if (ret)
8550                         btrfs_abort_transaction(trans, extent_root, ret);
8551                 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8552                                                key.objectid, key.offset);
8553                 if (ret)
8554                         btrfs_abort_transaction(trans, extent_root, ret);
8555         }
8556 }
8557
8558 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8559                            struct btrfs_root *root, u64 bytes_used,
8560                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8561                            u64 size)
8562 {
8563         int ret;
8564         struct btrfs_root *extent_root;
8565         struct btrfs_block_group_cache *cache;
8566
8567         extent_root = root->fs_info->extent_root;
8568
8569         root->fs_info->last_trans_log_full_commit = trans->transid;
8570
8571         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8572         if (!cache)
8573                 return -ENOMEM;
8574         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8575                                         GFP_NOFS);
8576         if (!cache->free_space_ctl) {
8577                 kfree(cache);
8578                 return -ENOMEM;
8579         }
8580
8581         cache->key.objectid = chunk_offset;
8582         cache->key.offset = size;
8583         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8584         cache->sectorsize = root->sectorsize;
8585         cache->fs_info = root->fs_info;
8586         cache->full_stripe_len = btrfs_full_stripe_len(root,
8587                                                &root->fs_info->mapping_tree,
8588                                                chunk_offset);
8589
8590         atomic_set(&cache->count, 1);
8591         spin_lock_init(&cache->lock);
8592         INIT_LIST_HEAD(&cache->list);
8593         INIT_LIST_HEAD(&cache->cluster_list);
8594         INIT_LIST_HEAD(&cache->new_bg_list);
8595
8596         btrfs_init_free_space_ctl(cache);
8597
8598         btrfs_set_block_group_used(&cache->item, bytes_used);
8599         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8600         cache->flags = type;
8601         btrfs_set_block_group_flags(&cache->item, type);
8602
8603         cache->last_byte_to_unpin = (u64)-1;
8604         cache->cached = BTRFS_CACHE_FINISHED;
8605         ret = exclude_super_stripes(root, cache);
8606         if (ret) {
8607                 /*
8608                  * We may have excluded something, so call this just in
8609                  * case.
8610                  */
8611                 free_excluded_extents(root, cache);
8612                 kfree(cache->free_space_ctl);
8613                 kfree(cache);
8614                 return ret;
8615         }
8616
8617         add_new_free_space(cache, root->fs_info, chunk_offset,
8618                            chunk_offset + size);
8619
8620         free_excluded_extents(root, cache);
8621
8622         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8623         if (ret) {
8624                 btrfs_remove_free_space_cache(cache);
8625                 btrfs_put_block_group(cache);
8626                 return ret;
8627         }
8628
8629         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8630                                 &cache->space_info);
8631         if (ret) {
8632                 btrfs_remove_free_space_cache(cache);
8633                 spin_lock(&root->fs_info->block_group_cache_lock);
8634                 rb_erase(&cache->cache_node,
8635                          &root->fs_info->block_group_cache_tree);
8636                 spin_unlock(&root->fs_info->block_group_cache_lock);
8637                 btrfs_put_block_group(cache);
8638                 return ret;
8639         }
8640         update_global_block_rsv(root->fs_info);
8641
8642         spin_lock(&cache->space_info->lock);
8643         cache->space_info->bytes_readonly += cache->bytes_super;
8644         spin_unlock(&cache->space_info->lock);
8645
8646         __link_block_group(cache->space_info, cache);
8647
8648         list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8649
8650         set_avail_alloc_bits(extent_root->fs_info, type);
8651
8652         return 0;
8653 }
8654
8655 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8656 {
8657         u64 extra_flags = chunk_to_extended(flags) &
8658                                 BTRFS_EXTENDED_PROFILE_MASK;
8659
8660         write_seqlock(&fs_info->profiles_lock);
8661         if (flags & BTRFS_BLOCK_GROUP_DATA)
8662                 fs_info->avail_data_alloc_bits &= ~extra_flags;
8663         if (flags & BTRFS_BLOCK_GROUP_METADATA)
8664                 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8665         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8666                 fs_info->avail_system_alloc_bits &= ~extra_flags;
8667         write_sequnlock(&fs_info->profiles_lock);
8668 }
8669
8670 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8671                              struct btrfs_root *root, u64 group_start)
8672 {
8673         struct btrfs_path *path;
8674         struct btrfs_block_group_cache *block_group;
8675         struct btrfs_free_cluster *cluster;
8676         struct btrfs_root *tree_root = root->fs_info->tree_root;
8677         struct btrfs_key key;
8678         struct inode *inode;
8679         int ret;
8680         int index;
8681         int factor;
8682
8683         root = root->fs_info->extent_root;
8684
8685         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8686         BUG_ON(!block_group);
8687         BUG_ON(!block_group->ro);
8688
8689         /*
8690          * Free the reserved super bytes from this block group before
8691          * remove it.
8692          */
8693         free_excluded_extents(root, block_group);
8694
8695         memcpy(&key, &block_group->key, sizeof(key));
8696         index = get_block_group_index(block_group);
8697         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8698                                   BTRFS_BLOCK_GROUP_RAID1 |
8699                                   BTRFS_BLOCK_GROUP_RAID10))
8700                 factor = 2;
8701         else
8702                 factor = 1;
8703
8704         /* make sure this block group isn't part of an allocation cluster */
8705         cluster = &root->fs_info->data_alloc_cluster;
8706         spin_lock(&cluster->refill_lock);
8707         btrfs_return_cluster_to_free_space(block_group, cluster);
8708         spin_unlock(&cluster->refill_lock);
8709
8710         /*
8711          * make sure this block group isn't part of a metadata
8712          * allocation cluster
8713          */
8714         cluster = &root->fs_info->meta_alloc_cluster;
8715         spin_lock(&cluster->refill_lock);
8716         btrfs_return_cluster_to_free_space(block_group, cluster);
8717         spin_unlock(&cluster->refill_lock);
8718
8719         path = btrfs_alloc_path();
8720         if (!path) {
8721                 ret = -ENOMEM;
8722                 goto out;
8723         }
8724
8725         inode = lookup_free_space_inode(tree_root, block_group, path);
8726         if (!IS_ERR(inode)) {
8727                 ret = btrfs_orphan_add(trans, inode);
8728                 if (ret) {
8729                         btrfs_add_delayed_iput(inode);
8730                         goto out;
8731                 }
8732                 clear_nlink(inode);
8733                 /* One for the block groups ref */
8734                 spin_lock(&block_group->lock);
8735                 if (block_group->iref) {
8736                         block_group->iref = 0;
8737                         block_group->inode = NULL;
8738                         spin_unlock(&block_group->lock);
8739                         iput(inode);
8740                 } else {
8741                         spin_unlock(&block_group->lock);
8742                 }
8743                 /* One for our lookup ref */
8744                 btrfs_add_delayed_iput(inode);
8745         }
8746
8747         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8748         key.offset = block_group->key.objectid;
8749         key.type = 0;
8750
8751         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8752         if (ret < 0)
8753                 goto out;
8754         if (ret > 0)
8755                 btrfs_release_path(path);
8756         if (ret == 0) {
8757                 ret = btrfs_del_item(trans, tree_root, path);
8758                 if (ret)
8759                         goto out;
8760                 btrfs_release_path(path);
8761         }
8762
8763         spin_lock(&root->fs_info->block_group_cache_lock);
8764         rb_erase(&block_group->cache_node,
8765                  &root->fs_info->block_group_cache_tree);
8766
8767         if (root->fs_info->first_logical_byte == block_group->key.objectid)
8768                 root->fs_info->first_logical_byte = (u64)-1;
8769         spin_unlock(&root->fs_info->block_group_cache_lock);
8770
8771         down_write(&block_group->space_info->groups_sem);
8772         /*
8773          * we must use list_del_init so people can check to see if they
8774          * are still on the list after taking the semaphore
8775          */
8776         list_del_init(&block_group->list);
8777         if (list_empty(&block_group->space_info->block_groups[index]))
8778                 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8779         up_write(&block_group->space_info->groups_sem);
8780
8781         if (block_group->cached == BTRFS_CACHE_STARTED)
8782                 wait_block_group_cache_done(block_group);
8783
8784         btrfs_remove_free_space_cache(block_group);
8785
8786         spin_lock(&block_group->space_info->lock);
8787         block_group->space_info->total_bytes -= block_group->key.offset;
8788         block_group->space_info->bytes_readonly -= block_group->key.offset;
8789         block_group->space_info->disk_total -= block_group->key.offset * factor;
8790         spin_unlock(&block_group->space_info->lock);
8791
8792         memcpy(&key, &block_group->key, sizeof(key));
8793
8794         btrfs_clear_space_info_full(root->fs_info);
8795
8796         btrfs_put_block_group(block_group);
8797         btrfs_put_block_group(block_group);
8798
8799         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8800         if (ret > 0)
8801                 ret = -EIO;
8802         if (ret < 0)
8803                 goto out;
8804
8805         ret = btrfs_del_item(trans, root, path);
8806 out:
8807         btrfs_free_path(path);
8808         return ret;
8809 }
8810
8811 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8812 {
8813         struct btrfs_space_info *space_info;
8814         struct btrfs_super_block *disk_super;
8815         u64 features;
8816         u64 flags;
8817         int mixed = 0;
8818         int ret;
8819
8820         disk_super = fs_info->super_copy;
8821         if (!btrfs_super_root(disk_super))
8822                 return 1;
8823
8824         features = btrfs_super_incompat_flags(disk_super);
8825         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8826                 mixed = 1;
8827
8828         flags = BTRFS_BLOCK_GROUP_SYSTEM;
8829         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8830         if (ret)
8831                 goto out;
8832
8833         if (mixed) {
8834                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8835                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8836         } else {
8837                 flags = BTRFS_BLOCK_GROUP_METADATA;
8838                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8839                 if (ret)
8840                         goto out;
8841
8842                 flags = BTRFS_BLOCK_GROUP_DATA;
8843                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8844         }
8845 out:
8846         return ret;
8847 }
8848
8849 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8850 {
8851         return unpin_extent_range(root, start, end);
8852 }
8853
8854 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8855                                u64 num_bytes, u64 *actual_bytes)
8856 {
8857         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8858 }
8859
8860 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8861 {
8862         struct btrfs_fs_info *fs_info = root->fs_info;
8863         struct btrfs_block_group_cache *cache = NULL;
8864         u64 group_trimmed;
8865         u64 start;
8866         u64 end;
8867         u64 trimmed = 0;
8868         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8869         int ret = 0;
8870
8871         /*
8872          * try to trim all FS space, our block group may start from non-zero.
8873          */
8874         if (range->len == total_bytes)
8875                 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8876         else
8877                 cache = btrfs_lookup_block_group(fs_info, range->start);
8878
8879         while (cache) {
8880                 if (cache->key.objectid >= (range->start + range->len)) {
8881                         btrfs_put_block_group(cache);
8882                         break;
8883                 }
8884
8885                 start = max(range->start, cache->key.objectid);
8886                 end = min(range->start + range->len,
8887                                 cache->key.objectid + cache->key.offset);
8888
8889                 if (end - start >= range->minlen) {
8890                         if (!block_group_cache_done(cache)) {
8891                                 ret = cache_block_group(cache, 0);
8892                                 if (ret) {
8893                                         btrfs_put_block_group(cache);
8894                                         break;
8895                                 }
8896                                 ret = wait_block_group_cache_done(cache);
8897                                 if (ret) {
8898                                         btrfs_put_block_group(cache);
8899                                         break;
8900                                 }
8901                         }
8902                         ret = btrfs_trim_block_group(cache,
8903                                                      &group_trimmed,
8904                                                      start,
8905                                                      end,
8906                                                      range->minlen);
8907
8908                         trimmed += group_trimmed;
8909                         if (ret) {
8910                                 btrfs_put_block_group(cache);
8911                                 break;
8912                         }
8913                 }
8914
8915                 cache = next_block_group(fs_info->tree_root, cache);
8916         }
8917
8918         range->len = trimmed;
8919         return ret;
8920 }