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_delalloc_roots(root->fs_info, 0);
4020                 if (!current->journal_info)
4021                         btrfs_wait_ordered_roots(root->fs_info, -1);
4022         }
4023 }
4024
4025 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4026 {
4027         u64 bytes;
4028         int nr;
4029
4030         bytes = btrfs_calc_trans_metadata_size(root, 1);
4031         nr = (int)div64_u64(to_reclaim, bytes);
4032         if (!nr)
4033                 nr = 1;
4034         return nr;
4035 }
4036
4037 #define EXTENT_SIZE_PER_ITEM    (256 * 1024)
4038
4039 /*
4040  * shrink metadata reservation for delalloc
4041  */
4042 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4043                             bool wait_ordered)
4044 {
4045         struct btrfs_block_rsv *block_rsv;
4046         struct btrfs_space_info *space_info;
4047         struct btrfs_trans_handle *trans;
4048         u64 delalloc_bytes;
4049         u64 max_reclaim;
4050         long time_left;
4051         unsigned long nr_pages;
4052         int loops;
4053         int items;
4054         enum btrfs_reserve_flush_enum flush;
4055
4056         /* Calc the number of the pages we need flush for space reservation */
4057         items = calc_reclaim_items_nr(root, to_reclaim);
4058         to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4059
4060         trans = (struct btrfs_trans_handle *)current->journal_info;
4061         block_rsv = &root->fs_info->delalloc_block_rsv;
4062         space_info = block_rsv->space_info;
4063
4064         delalloc_bytes = percpu_counter_sum_positive(
4065                                                 &root->fs_info->delalloc_bytes);
4066         if (delalloc_bytes == 0) {
4067                 if (trans)
4068                         return;
4069                 if (wait_ordered)
4070                         btrfs_wait_ordered_roots(root->fs_info, items);
4071                 return;
4072         }
4073
4074         loops = 0;
4075         while (delalloc_bytes && loops < 3) {
4076                 max_reclaim = min(delalloc_bytes, to_reclaim);
4077                 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4078                 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4079                 /*
4080                  * We need to wait for the async pages to actually start before
4081                  * we do anything.
4082                  */
4083                 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4084                 if (!max_reclaim)
4085                         goto skip_async;
4086
4087                 if (max_reclaim <= nr_pages)
4088                         max_reclaim = 0;
4089                 else
4090                         max_reclaim -= nr_pages;
4091
4092                 wait_event(root->fs_info->async_submit_wait,
4093                            atomic_read(&root->fs_info->async_delalloc_pages) <=
4094                            (int)max_reclaim);
4095 skip_async:
4096                 if (!trans)
4097                         flush = BTRFS_RESERVE_FLUSH_ALL;
4098                 else
4099                         flush = BTRFS_RESERVE_NO_FLUSH;
4100                 spin_lock(&space_info->lock);
4101                 if (can_overcommit(root, space_info, orig, flush)) {
4102                         spin_unlock(&space_info->lock);
4103                         break;
4104                 }
4105                 spin_unlock(&space_info->lock);
4106
4107                 loops++;
4108                 if (wait_ordered && !trans) {
4109                         btrfs_wait_ordered_roots(root->fs_info, items);
4110                 } else {
4111                         time_left = schedule_timeout_killable(1);
4112                         if (time_left)
4113                                 break;
4114                 }
4115                 delalloc_bytes = percpu_counter_sum_positive(
4116                                                 &root->fs_info->delalloc_bytes);
4117         }
4118 }
4119
4120 /**
4121  * maybe_commit_transaction - possibly commit the transaction if its ok to
4122  * @root - the root we're allocating for
4123  * @bytes - the number of bytes we want to reserve
4124  * @force - force the commit
4125  *
4126  * This will check to make sure that committing the transaction will actually
4127  * get us somewhere and then commit the transaction if it does.  Otherwise it
4128  * will return -ENOSPC.
4129  */
4130 static int may_commit_transaction(struct btrfs_root *root,
4131                                   struct btrfs_space_info *space_info,
4132                                   u64 bytes, int force)
4133 {
4134         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4135         struct btrfs_trans_handle *trans;
4136
4137         trans = (struct btrfs_trans_handle *)current->journal_info;
4138         if (trans)
4139                 return -EAGAIN;
4140
4141         if (force)
4142                 goto commit;
4143
4144         /* See if there is enough pinned space to make this reservation */
4145         spin_lock(&space_info->lock);
4146         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4147                                    bytes) >= 0) {
4148                 spin_unlock(&space_info->lock);
4149                 goto commit;
4150         }
4151         spin_unlock(&space_info->lock);
4152
4153         /*
4154          * See if there is some space in the delayed insertion reservation for
4155          * this reservation.
4156          */
4157         if (space_info != delayed_rsv->space_info)
4158                 return -ENOSPC;
4159
4160         spin_lock(&space_info->lock);
4161         spin_lock(&delayed_rsv->lock);
4162         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4163                                    bytes - delayed_rsv->size) >= 0) {
4164                 spin_unlock(&delayed_rsv->lock);
4165                 spin_unlock(&space_info->lock);
4166                 return -ENOSPC;
4167         }
4168         spin_unlock(&delayed_rsv->lock);
4169         spin_unlock(&space_info->lock);
4170
4171 commit:
4172         trans = btrfs_join_transaction(root);
4173         if (IS_ERR(trans))
4174                 return -ENOSPC;
4175
4176         return btrfs_commit_transaction(trans, root);
4177 }
4178
4179 enum flush_state {
4180         FLUSH_DELAYED_ITEMS_NR  =       1,
4181         FLUSH_DELAYED_ITEMS     =       2,
4182         FLUSH_DELALLOC          =       3,
4183         FLUSH_DELALLOC_WAIT     =       4,
4184         ALLOC_CHUNK             =       5,
4185         COMMIT_TRANS            =       6,
4186 };
4187
4188 static int flush_space(struct btrfs_root *root,
4189                        struct btrfs_space_info *space_info, u64 num_bytes,
4190                        u64 orig_bytes, int state)
4191 {
4192         struct btrfs_trans_handle *trans;
4193         int nr;
4194         int ret = 0;
4195
4196         switch (state) {
4197         case FLUSH_DELAYED_ITEMS_NR:
4198         case FLUSH_DELAYED_ITEMS:
4199                 if (state == FLUSH_DELAYED_ITEMS_NR)
4200                         nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4201                 else
4202                         nr = -1;
4203
4204                 trans = btrfs_join_transaction(root);
4205                 if (IS_ERR(trans)) {
4206                         ret = PTR_ERR(trans);
4207                         break;
4208                 }
4209                 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4210                 btrfs_end_transaction(trans, root);
4211                 break;
4212         case FLUSH_DELALLOC:
4213         case FLUSH_DELALLOC_WAIT:
4214                 shrink_delalloc(root, num_bytes, orig_bytes,
4215                                 state == FLUSH_DELALLOC_WAIT);
4216                 break;
4217         case ALLOC_CHUNK:
4218                 trans = btrfs_join_transaction(root);
4219                 if (IS_ERR(trans)) {
4220                         ret = PTR_ERR(trans);
4221                         break;
4222                 }
4223                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4224                                      btrfs_get_alloc_profile(root, 0),
4225                                      CHUNK_ALLOC_NO_FORCE);
4226                 btrfs_end_transaction(trans, root);
4227                 if (ret == -ENOSPC)
4228                         ret = 0;
4229                 break;
4230         case COMMIT_TRANS:
4231                 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4232                 break;
4233         default:
4234                 ret = -ENOSPC;
4235                 break;
4236         }
4237
4238         return ret;
4239 }
4240 /**
4241  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4242  * @root - the root we're allocating for
4243  * @block_rsv - the block_rsv we're allocating for
4244  * @orig_bytes - the number of bytes we want
4245  * @flush - whether or not we can flush to make our reservation
4246  *
4247  * This will reserve orgi_bytes number of bytes from the space info associated
4248  * with the block_rsv.  If there is not enough space it will make an attempt to
4249  * flush out space to make room.  It will do this by flushing delalloc if
4250  * possible or committing the transaction.  If flush is 0 then no attempts to
4251  * regain reservations will be made and this will fail if there is not enough
4252  * space already.
4253  */
4254 static int reserve_metadata_bytes(struct btrfs_root *root,
4255                                   struct btrfs_block_rsv *block_rsv,
4256                                   u64 orig_bytes,
4257                                   enum btrfs_reserve_flush_enum flush)
4258 {
4259         struct btrfs_space_info *space_info = block_rsv->space_info;
4260         u64 used;
4261         u64 num_bytes = orig_bytes;
4262         int flush_state = FLUSH_DELAYED_ITEMS_NR;
4263         int ret = 0;
4264         bool flushing = false;
4265
4266 again:
4267         ret = 0;
4268         spin_lock(&space_info->lock);
4269         /*
4270          * We only want to wait if somebody other than us is flushing and we
4271          * are actually allowed to flush all things.
4272          */
4273         while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4274                space_info->flush) {
4275                 spin_unlock(&space_info->lock);
4276                 /*
4277                  * If we have a trans handle we can't wait because the flusher
4278                  * may have to commit the transaction, which would mean we would
4279                  * deadlock since we are waiting for the flusher to finish, but
4280                  * hold the current transaction open.
4281                  */
4282                 if (current->journal_info)
4283                         return -EAGAIN;
4284                 ret = wait_event_killable(space_info->wait, !space_info->flush);
4285                 /* Must have been killed, return */
4286                 if (ret)
4287                         return -EINTR;
4288
4289                 spin_lock(&space_info->lock);
4290         }
4291
4292         ret = -ENOSPC;
4293         used = space_info->bytes_used + space_info->bytes_reserved +
4294                 space_info->bytes_pinned + space_info->bytes_readonly +
4295                 space_info->bytes_may_use;
4296
4297         /*
4298          * The idea here is that we've not already over-reserved the block group
4299          * then we can go ahead and save our reservation first and then start
4300          * flushing if we need to.  Otherwise if we've already overcommitted
4301          * lets start flushing stuff first and then come back and try to make
4302          * our reservation.
4303          */
4304         if (used <= space_info->total_bytes) {
4305                 if (used + orig_bytes <= space_info->total_bytes) {
4306                         space_info->bytes_may_use += orig_bytes;
4307                         trace_btrfs_space_reservation(root->fs_info,
4308                                 "space_info", space_info->flags, orig_bytes, 1);
4309                         ret = 0;
4310                 } else {
4311                         /*
4312                          * Ok set num_bytes to orig_bytes since we aren't
4313                          * overocmmitted, this way we only try and reclaim what
4314                          * we need.
4315                          */
4316                         num_bytes = orig_bytes;
4317                 }
4318         } else {
4319                 /*
4320                  * Ok we're over committed, set num_bytes to the overcommitted
4321                  * amount plus the amount of bytes that we need for this
4322                  * reservation.
4323                  */
4324                 num_bytes = used - space_info->total_bytes +
4325                         (orig_bytes * 2);
4326         }
4327
4328         if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4329                 space_info->bytes_may_use += orig_bytes;
4330                 trace_btrfs_space_reservation(root->fs_info, "space_info",
4331                                               space_info->flags, orig_bytes,
4332                                               1);
4333                 ret = 0;
4334         }
4335
4336         /*
4337          * Couldn't make our reservation, save our place so while we're trying
4338          * to reclaim space we can actually use it instead of somebody else
4339          * stealing it from us.
4340          *
4341          * We make the other tasks wait for the flush only when we can flush
4342          * all things.
4343          */
4344         if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4345                 flushing = true;
4346                 space_info->flush = 1;
4347         }
4348
4349         spin_unlock(&space_info->lock);
4350
4351         if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4352                 goto out;
4353
4354         ret = flush_space(root, space_info, num_bytes, orig_bytes,
4355                           flush_state);
4356         flush_state++;
4357
4358         /*
4359          * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4360          * would happen. So skip delalloc flush.
4361          */
4362         if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4363             (flush_state == FLUSH_DELALLOC ||
4364              flush_state == FLUSH_DELALLOC_WAIT))
4365                 flush_state = ALLOC_CHUNK;
4366
4367         if (!ret)
4368                 goto again;
4369         else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4370                  flush_state < COMMIT_TRANS)
4371                 goto again;
4372         else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4373                  flush_state <= COMMIT_TRANS)
4374                 goto again;
4375
4376 out:
4377         if (ret == -ENOSPC &&
4378             unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4379                 struct btrfs_block_rsv *global_rsv =
4380                         &root->fs_info->global_block_rsv;
4381
4382                 if (block_rsv != global_rsv &&
4383                     !block_rsv_use_bytes(global_rsv, orig_bytes))
4384                         ret = 0;
4385         }
4386         if (ret == -ENOSPC)
4387                 trace_btrfs_space_reservation(root->fs_info,
4388                                               "space_info:enospc",
4389                                               space_info->flags, orig_bytes, 1);
4390         if (flushing) {
4391                 spin_lock(&space_info->lock);
4392                 space_info->flush = 0;
4393                 wake_up_all(&space_info->wait);
4394                 spin_unlock(&space_info->lock);
4395         }
4396         return ret;
4397 }
4398
4399 static struct btrfs_block_rsv *get_block_rsv(
4400                                         const struct btrfs_trans_handle *trans,
4401                                         const struct btrfs_root *root)
4402 {
4403         struct btrfs_block_rsv *block_rsv = NULL;
4404
4405         if (root->ref_cows)
4406                 block_rsv = trans->block_rsv;
4407
4408         if (root == root->fs_info->csum_root && trans->adding_csums)
4409                 block_rsv = trans->block_rsv;
4410
4411         if (root == root->fs_info->uuid_root)
4412                 block_rsv = trans->block_rsv;
4413
4414         if (!block_rsv)
4415                 block_rsv = root->block_rsv;
4416
4417         if (!block_rsv)
4418                 block_rsv = &root->fs_info->empty_block_rsv;
4419
4420         return block_rsv;
4421 }
4422
4423 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4424                                u64 num_bytes)
4425 {
4426         int ret = -ENOSPC;
4427         spin_lock(&block_rsv->lock);
4428         if (block_rsv->reserved >= num_bytes) {
4429                 block_rsv->reserved -= num_bytes;
4430                 if (block_rsv->reserved < block_rsv->size)
4431                         block_rsv->full = 0;
4432                 ret = 0;
4433         }
4434         spin_unlock(&block_rsv->lock);
4435         return ret;
4436 }
4437
4438 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4439                                 u64 num_bytes, int update_size)
4440 {
4441         spin_lock(&block_rsv->lock);
4442         block_rsv->reserved += num_bytes;
4443         if (update_size)
4444                 block_rsv->size += num_bytes;
4445         else if (block_rsv->reserved >= block_rsv->size)
4446                 block_rsv->full = 1;
4447         spin_unlock(&block_rsv->lock);
4448 }
4449
4450 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4451                              struct btrfs_block_rsv *dest, u64 num_bytes,
4452                              int min_factor)
4453 {
4454         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4455         u64 min_bytes;
4456
4457         if (global_rsv->space_info != dest->space_info)
4458                 return -ENOSPC;
4459
4460         spin_lock(&global_rsv->lock);
4461         min_bytes = div_factor(global_rsv->size, min_factor);
4462         if (global_rsv->reserved < min_bytes + num_bytes) {
4463                 spin_unlock(&global_rsv->lock);
4464                 return -ENOSPC;
4465         }
4466         global_rsv->reserved -= num_bytes;
4467         if (global_rsv->reserved < global_rsv->size)
4468                 global_rsv->full = 0;
4469         spin_unlock(&global_rsv->lock);
4470
4471         block_rsv_add_bytes(dest, num_bytes, 1);
4472         return 0;
4473 }
4474
4475 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4476                                     struct btrfs_block_rsv *block_rsv,
4477                                     struct btrfs_block_rsv *dest, u64 num_bytes)
4478 {
4479         struct btrfs_space_info *space_info = block_rsv->space_info;
4480
4481         spin_lock(&block_rsv->lock);
4482         if (num_bytes == (u64)-1)
4483                 num_bytes = block_rsv->size;
4484         block_rsv->size -= num_bytes;
4485         if (block_rsv->reserved >= block_rsv->size) {
4486                 num_bytes = block_rsv->reserved - block_rsv->size;
4487                 block_rsv->reserved = block_rsv->size;
4488                 block_rsv->full = 1;
4489         } else {
4490                 num_bytes = 0;
4491         }
4492         spin_unlock(&block_rsv->lock);
4493
4494         if (num_bytes > 0) {
4495                 if (dest) {
4496                         spin_lock(&dest->lock);
4497                         if (!dest->full) {
4498                                 u64 bytes_to_add;
4499
4500                                 bytes_to_add = dest->size - dest->reserved;
4501                                 bytes_to_add = min(num_bytes, bytes_to_add);
4502                                 dest->reserved += bytes_to_add;
4503                                 if (dest->reserved >= dest->size)
4504                                         dest->full = 1;
4505                                 num_bytes -= bytes_to_add;
4506                         }
4507                         spin_unlock(&dest->lock);
4508                 }
4509                 if (num_bytes) {
4510                         spin_lock(&space_info->lock);
4511                         space_info->bytes_may_use -= num_bytes;
4512                         trace_btrfs_space_reservation(fs_info, "space_info",
4513                                         space_info->flags, num_bytes, 0);
4514                         spin_unlock(&space_info->lock);
4515                 }
4516         }
4517 }
4518
4519 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4520                                    struct btrfs_block_rsv *dst, u64 num_bytes)
4521 {
4522         int ret;
4523
4524         ret = block_rsv_use_bytes(src, num_bytes);
4525         if (ret)
4526                 return ret;
4527
4528         block_rsv_add_bytes(dst, num_bytes, 1);
4529         return 0;
4530 }
4531
4532 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4533 {
4534         memset(rsv, 0, sizeof(*rsv));
4535         spin_lock_init(&rsv->lock);
4536         rsv->type = type;
4537 }
4538
4539 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4540                                               unsigned short type)
4541 {
4542         struct btrfs_block_rsv *block_rsv;
4543         struct btrfs_fs_info *fs_info = root->fs_info;
4544
4545         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4546         if (!block_rsv)
4547                 return NULL;
4548
4549         btrfs_init_block_rsv(block_rsv, type);
4550         block_rsv->space_info = __find_space_info(fs_info,
4551                                                   BTRFS_BLOCK_GROUP_METADATA);
4552         return block_rsv;
4553 }
4554
4555 void btrfs_free_block_rsv(struct btrfs_root *root,
4556                           struct btrfs_block_rsv *rsv)
4557 {
4558         if (!rsv)
4559                 return;
4560         btrfs_block_rsv_release(root, rsv, (u64)-1);
4561         kfree(rsv);
4562 }
4563
4564 int btrfs_block_rsv_add(struct btrfs_root *root,
4565                         struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4566                         enum btrfs_reserve_flush_enum flush)
4567 {
4568         int ret;
4569
4570         if (num_bytes == 0)
4571                 return 0;
4572
4573         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4574         if (!ret) {
4575                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4576                 return 0;
4577         }
4578
4579         return ret;
4580 }
4581
4582 int btrfs_block_rsv_check(struct btrfs_root *root,
4583                           struct btrfs_block_rsv *block_rsv, int min_factor)
4584 {
4585         u64 num_bytes = 0;
4586         int ret = -ENOSPC;
4587
4588         if (!block_rsv)
4589                 return 0;
4590
4591         spin_lock(&block_rsv->lock);
4592         num_bytes = div_factor(block_rsv->size, min_factor);
4593         if (block_rsv->reserved >= num_bytes)
4594                 ret = 0;
4595         spin_unlock(&block_rsv->lock);
4596
4597         return ret;
4598 }
4599
4600 int btrfs_block_rsv_refill(struct btrfs_root *root,
4601                            struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4602                            enum btrfs_reserve_flush_enum flush)
4603 {
4604         u64 num_bytes = 0;
4605         int ret = -ENOSPC;
4606
4607         if (!block_rsv)
4608                 return 0;
4609
4610         spin_lock(&block_rsv->lock);
4611         num_bytes = min_reserved;
4612         if (block_rsv->reserved >= num_bytes)
4613                 ret = 0;
4614         else
4615                 num_bytes -= block_rsv->reserved;
4616         spin_unlock(&block_rsv->lock);
4617
4618         if (!ret)
4619                 return 0;
4620
4621         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4622         if (!ret) {
4623                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4624                 return 0;
4625         }
4626
4627         return ret;
4628 }
4629
4630 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4631                             struct btrfs_block_rsv *dst_rsv,
4632                             u64 num_bytes)
4633 {
4634         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4635 }
4636
4637 void btrfs_block_rsv_release(struct btrfs_root *root,
4638                              struct btrfs_block_rsv *block_rsv,
4639                              u64 num_bytes)
4640 {
4641         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4642         if (global_rsv->full || global_rsv == block_rsv ||
4643             block_rsv->space_info != global_rsv->space_info)
4644                 global_rsv = NULL;
4645         block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4646                                 num_bytes);
4647 }
4648
4649 /*
4650  * helper to calculate size of global block reservation.
4651  * the desired value is sum of space used by extent tree,
4652  * checksum tree and root tree
4653  */
4654 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4655 {
4656         struct btrfs_space_info *sinfo;
4657         u64 num_bytes;
4658         u64 meta_used;
4659         u64 data_used;
4660         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4661
4662         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4663         spin_lock(&sinfo->lock);
4664         data_used = sinfo->bytes_used;
4665         spin_unlock(&sinfo->lock);
4666
4667         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4668         spin_lock(&sinfo->lock);
4669         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4670                 data_used = 0;
4671         meta_used = sinfo->bytes_used;
4672         spin_unlock(&sinfo->lock);
4673
4674         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4675                     csum_size * 2;
4676         num_bytes += div64_u64(data_used + meta_used, 50);
4677
4678         if (num_bytes * 3 > meta_used)
4679                 num_bytes = div64_u64(meta_used, 3);
4680
4681         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4682 }
4683
4684 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4685 {
4686         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4687         struct btrfs_space_info *sinfo = block_rsv->space_info;
4688         u64 num_bytes;
4689
4690         num_bytes = calc_global_metadata_size(fs_info);
4691
4692         spin_lock(&sinfo->lock);
4693         spin_lock(&block_rsv->lock);
4694
4695         block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4696
4697         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4698                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4699                     sinfo->bytes_may_use;
4700
4701         if (sinfo->total_bytes > num_bytes) {
4702                 num_bytes = sinfo->total_bytes - num_bytes;
4703                 block_rsv->reserved += num_bytes;
4704                 sinfo->bytes_may_use += num_bytes;
4705                 trace_btrfs_space_reservation(fs_info, "space_info",
4706                                       sinfo->flags, num_bytes, 1);
4707         }
4708
4709         if (block_rsv->reserved >= block_rsv->size) {
4710                 num_bytes = block_rsv->reserved - block_rsv->size;
4711                 sinfo->bytes_may_use -= num_bytes;
4712                 trace_btrfs_space_reservation(fs_info, "space_info",
4713                                       sinfo->flags, num_bytes, 0);
4714                 block_rsv->reserved = block_rsv->size;
4715                 block_rsv->full = 1;
4716         }
4717
4718         spin_unlock(&block_rsv->lock);
4719         spin_unlock(&sinfo->lock);
4720 }
4721
4722 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4723 {
4724         struct btrfs_space_info *space_info;
4725
4726         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4727         fs_info->chunk_block_rsv.space_info = space_info;
4728
4729         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4730         fs_info->global_block_rsv.space_info = space_info;
4731         fs_info->delalloc_block_rsv.space_info = space_info;
4732         fs_info->trans_block_rsv.space_info = space_info;
4733         fs_info->empty_block_rsv.space_info = space_info;
4734         fs_info->delayed_block_rsv.space_info = space_info;
4735
4736         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4737         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4738         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4739         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4740         if (fs_info->quota_root)
4741                 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4742         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4743
4744         update_global_block_rsv(fs_info);
4745 }
4746
4747 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4748 {
4749         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4750                                 (u64)-1);
4751         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4752         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4753         WARN_ON(fs_info->trans_block_rsv.size > 0);
4754         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4755         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4756         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4757         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4758         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4759 }
4760
4761 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4762                                   struct btrfs_root *root)
4763 {
4764         if (!trans->block_rsv)
4765                 return;
4766
4767         if (!trans->bytes_reserved)
4768                 return;
4769
4770         trace_btrfs_space_reservation(root->fs_info, "transaction",
4771                                       trans->transid, trans->bytes_reserved, 0);
4772         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4773         trans->bytes_reserved = 0;
4774 }
4775
4776 /* Can only return 0 or -ENOSPC */
4777 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4778                                   struct inode *inode)
4779 {
4780         struct btrfs_root *root = BTRFS_I(inode)->root;
4781         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4782         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4783
4784         /*
4785          * We need to hold space in order to delete our orphan item once we've
4786          * added it, so this takes the reservation so we can release it later
4787          * when we are truly done with the orphan item.
4788          */
4789         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4790         trace_btrfs_space_reservation(root->fs_info, "orphan",
4791                                       btrfs_ino(inode), num_bytes, 1);
4792         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4793 }
4794
4795 void btrfs_orphan_release_metadata(struct inode *inode)
4796 {
4797         struct btrfs_root *root = BTRFS_I(inode)->root;
4798         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4799         trace_btrfs_space_reservation(root->fs_info, "orphan",
4800                                       btrfs_ino(inode), num_bytes, 0);
4801         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4802 }
4803
4804 /*
4805  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4806  * root: the root of the parent directory
4807  * rsv: block reservation
4808  * items: the number of items that we need do reservation
4809  * qgroup_reserved: used to return the reserved size in qgroup
4810  *
4811  * This function is used to reserve the space for snapshot/subvolume
4812  * creation and deletion. Those operations are different with the
4813  * common file/directory operations, they change two fs/file trees
4814  * and root tree, the number of items that the qgroup reserves is
4815  * different with the free space reservation. So we can not use
4816  * the space reseravtion mechanism in start_transaction().
4817  */
4818 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4819                                      struct btrfs_block_rsv *rsv,
4820                                      int items,
4821                                      u64 *qgroup_reserved,
4822                                      bool use_global_rsv)
4823 {
4824         u64 num_bytes;
4825         int ret;
4826         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4827
4828         if (root->fs_info->quota_enabled) {
4829                 /* One for parent inode, two for dir entries */
4830                 num_bytes = 3 * root->leafsize;
4831                 ret = btrfs_qgroup_reserve(root, num_bytes);
4832                 if (ret)
4833                         return ret;
4834         } else {
4835                 num_bytes = 0;
4836         }
4837
4838         *qgroup_reserved = num_bytes;
4839
4840         num_bytes = btrfs_calc_trans_metadata_size(root, items);
4841         rsv->space_info = __find_space_info(root->fs_info,
4842                                             BTRFS_BLOCK_GROUP_METADATA);
4843         ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4844                                   BTRFS_RESERVE_FLUSH_ALL);
4845
4846         if (ret == -ENOSPC && use_global_rsv)
4847                 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4848
4849         if (ret) {
4850                 if (*qgroup_reserved)
4851                         btrfs_qgroup_free(root, *qgroup_reserved);
4852         }
4853
4854         return ret;
4855 }
4856
4857 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4858                                       struct btrfs_block_rsv *rsv,
4859                                       u64 qgroup_reserved)
4860 {
4861         btrfs_block_rsv_release(root, rsv, (u64)-1);
4862         if (qgroup_reserved)
4863                 btrfs_qgroup_free(root, qgroup_reserved);
4864 }
4865
4866 /**
4867  * drop_outstanding_extent - drop an outstanding extent
4868  * @inode: the inode we're dropping the extent for
4869  *
4870  * This is called when we are freeing up an outstanding extent, either called
4871  * after an error or after an extent is written.  This will return the number of
4872  * reserved extents that need to be freed.  This must be called with
4873  * BTRFS_I(inode)->lock held.
4874  */
4875 static unsigned drop_outstanding_extent(struct inode *inode)
4876 {
4877         unsigned drop_inode_space = 0;
4878         unsigned dropped_extents = 0;
4879
4880         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4881         BTRFS_I(inode)->outstanding_extents--;
4882
4883         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4884             test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4885                                &BTRFS_I(inode)->runtime_flags))
4886                 drop_inode_space = 1;
4887
4888         /*
4889          * If we have more or the same amount of outsanding extents than we have
4890          * reserved then we need to leave the reserved extents count alone.
4891          */
4892         if (BTRFS_I(inode)->outstanding_extents >=
4893             BTRFS_I(inode)->reserved_extents)
4894                 return drop_inode_space;
4895
4896         dropped_extents = BTRFS_I(inode)->reserved_extents -
4897                 BTRFS_I(inode)->outstanding_extents;
4898         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4899         return dropped_extents + drop_inode_space;
4900 }
4901
4902 /**
4903  * calc_csum_metadata_size - return the amount of metada space that must be
4904  *      reserved/free'd for the given bytes.
4905  * @inode: the inode we're manipulating
4906  * @num_bytes: the number of bytes in question
4907  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4908  *
4909  * This adjusts the number of csum_bytes in the inode and then returns the
4910  * correct amount of metadata that must either be reserved or freed.  We
4911  * calculate how many checksums we can fit into one leaf and then divide the
4912  * number of bytes that will need to be checksumed by this value to figure out
4913  * how many checksums will be required.  If we are adding bytes then the number
4914  * may go up and we will return the number of additional bytes that must be
4915  * reserved.  If it is going down we will return the number of bytes that must
4916  * be freed.
4917  *
4918  * This must be called with BTRFS_I(inode)->lock held.
4919  */
4920 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4921                                    int reserve)
4922 {
4923         struct btrfs_root *root = BTRFS_I(inode)->root;
4924         u64 csum_size;
4925         int num_csums_per_leaf;
4926         int num_csums;
4927         int old_csums;
4928
4929         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4930             BTRFS_I(inode)->csum_bytes == 0)
4931                 return 0;
4932
4933         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4934         if (reserve)
4935                 BTRFS_I(inode)->csum_bytes += num_bytes;
4936         else
4937                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4938         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4939         num_csums_per_leaf = (int)div64_u64(csum_size,
4940                                             sizeof(struct btrfs_csum_item) +
4941                                             sizeof(struct btrfs_disk_key));
4942         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4943         num_csums = num_csums + num_csums_per_leaf - 1;
4944         num_csums = num_csums / num_csums_per_leaf;
4945
4946         old_csums = old_csums + num_csums_per_leaf - 1;
4947         old_csums = old_csums / num_csums_per_leaf;
4948
4949         /* No change, no need to reserve more */
4950         if (old_csums == num_csums)
4951                 return 0;
4952
4953         if (reserve)
4954                 return btrfs_calc_trans_metadata_size(root,
4955                                                       num_csums - old_csums);
4956
4957         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4958 }
4959
4960 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4961 {
4962         struct btrfs_root *root = BTRFS_I(inode)->root;
4963         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4964         u64 to_reserve = 0;
4965         u64 csum_bytes;
4966         unsigned nr_extents = 0;
4967         int extra_reserve = 0;
4968         enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4969         int ret = 0;
4970         bool delalloc_lock = true;
4971         u64 to_free = 0;
4972         unsigned dropped;
4973
4974         /* If we are a free space inode we need to not flush since we will be in
4975          * the middle of a transaction commit.  We also don't need the delalloc
4976          * mutex since we won't race with anybody.  We need this mostly to make
4977          * lockdep shut its filthy mouth.
4978          */
4979         if (btrfs_is_free_space_inode(inode)) {
4980                 flush = BTRFS_RESERVE_NO_FLUSH;
4981                 delalloc_lock = false;
4982         }
4983
4984         if (flush != BTRFS_RESERVE_NO_FLUSH &&
4985             btrfs_transaction_in_commit(root->fs_info))
4986                 schedule_timeout(1);
4987
4988         if (delalloc_lock)
4989                 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4990
4991         num_bytes = ALIGN(num_bytes, root->sectorsize);
4992
4993         spin_lock(&BTRFS_I(inode)->lock);
4994         BTRFS_I(inode)->outstanding_extents++;
4995
4996         if (BTRFS_I(inode)->outstanding_extents >
4997             BTRFS_I(inode)->reserved_extents)
4998                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4999                         BTRFS_I(inode)->reserved_extents;
5000
5001         /*
5002          * Add an item to reserve for updating the inode when we complete the
5003          * delalloc io.
5004          */
5005         if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5006                       &BTRFS_I(inode)->runtime_flags)) {
5007                 nr_extents++;
5008                 extra_reserve = 1;
5009         }
5010
5011         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5012         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5013         csum_bytes = BTRFS_I(inode)->csum_bytes;
5014         spin_unlock(&BTRFS_I(inode)->lock);
5015
5016         if (root->fs_info->quota_enabled) {
5017                 ret = btrfs_qgroup_reserve(root, num_bytes +
5018                                            nr_extents * root->leafsize);
5019                 if (ret)
5020                         goto out_fail;
5021         }
5022
5023         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5024         if (unlikely(ret)) {
5025                 if (root->fs_info->quota_enabled)
5026                         btrfs_qgroup_free(root, num_bytes +
5027                                                 nr_extents * root->leafsize);
5028                 goto out_fail;
5029         }
5030
5031         spin_lock(&BTRFS_I(inode)->lock);
5032         if (extra_reserve) {
5033                 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5034                         &BTRFS_I(inode)->runtime_flags);
5035                 nr_extents--;
5036         }
5037         BTRFS_I(inode)->reserved_extents += nr_extents;
5038         spin_unlock(&BTRFS_I(inode)->lock);
5039
5040         if (delalloc_lock)
5041                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5042
5043         if (to_reserve)
5044                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5045                                               btrfs_ino(inode), to_reserve, 1);
5046         block_rsv_add_bytes(block_rsv, to_reserve, 1);
5047
5048         return 0;
5049
5050 out_fail:
5051         spin_lock(&BTRFS_I(inode)->lock);
5052         dropped = drop_outstanding_extent(inode);
5053         /*
5054          * If the inodes csum_bytes is the same as the original
5055          * csum_bytes then we know we haven't raced with any free()ers
5056          * so we can just reduce our inodes csum bytes and carry on.
5057          */
5058         if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5059                 calc_csum_metadata_size(inode, num_bytes, 0);
5060         } else {
5061                 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5062                 u64 bytes;
5063
5064                 /*
5065                  * This is tricky, but first we need to figure out how much we
5066                  * free'd from any free-ers that occured during this
5067                  * reservation, so we reset ->csum_bytes to the csum_bytes
5068                  * before we dropped our lock, and then call the free for the
5069                  * number of bytes that were freed while we were trying our
5070                  * reservation.
5071                  */
5072                 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5073                 BTRFS_I(inode)->csum_bytes = csum_bytes;
5074                 to_free = calc_csum_metadata_size(inode, bytes, 0);
5075
5076
5077                 /*
5078                  * Now we need to see how much we would have freed had we not
5079                  * been making this reservation and our ->csum_bytes were not
5080                  * artificially inflated.
5081                  */
5082                 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5083                 bytes = csum_bytes - orig_csum_bytes;
5084                 bytes = calc_csum_metadata_size(inode, bytes, 0);
5085
5086                 /*
5087                  * Now reset ->csum_bytes to what it should be.  If bytes is
5088                  * more than to_free then we would have free'd more space had we
5089                  * not had an artificially high ->csum_bytes, so we need to free
5090                  * the remainder.  If bytes is the same or less then we don't
5091                  * need to do anything, the other free-ers did the correct
5092                  * thing.
5093                  */
5094                 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5095                 if (bytes > to_free)
5096                         to_free = bytes - to_free;
5097                 else
5098                         to_free = 0;
5099         }
5100         spin_unlock(&BTRFS_I(inode)->lock);
5101         if (dropped)
5102                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5103
5104         if (to_free) {
5105                 btrfs_block_rsv_release(root, block_rsv, to_free);
5106                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5107                                               btrfs_ino(inode), to_free, 0);
5108         }
5109         if (delalloc_lock)
5110                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5111         return ret;
5112 }
5113
5114 /**
5115  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5116  * @inode: the inode to release the reservation for
5117  * @num_bytes: the number of bytes we're releasing
5118  *
5119  * This will release the metadata reservation for an inode.  This can be called
5120  * once we complete IO for a given set of bytes to release their metadata
5121  * reservations.
5122  */
5123 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5124 {
5125         struct btrfs_root *root = BTRFS_I(inode)->root;
5126         u64 to_free = 0;
5127         unsigned dropped;
5128
5129         num_bytes = ALIGN(num_bytes, root->sectorsize);
5130         spin_lock(&BTRFS_I(inode)->lock);
5131         dropped = drop_outstanding_extent(inode);
5132
5133         if (num_bytes)
5134                 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5135         spin_unlock(&BTRFS_I(inode)->lock);
5136         if (dropped > 0)
5137                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5138
5139         trace_btrfs_space_reservation(root->fs_info, "delalloc",
5140                                       btrfs_ino(inode), to_free, 0);
5141         if (root->fs_info->quota_enabled) {
5142                 btrfs_qgroup_free(root, num_bytes +
5143                                         dropped * root->leafsize);
5144         }
5145
5146         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5147                                 to_free);
5148 }
5149
5150 /**
5151  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5152  * @inode: inode we're writing to
5153  * @num_bytes: the number of bytes we want to allocate
5154  *
5155  * This will do the following things
5156  *
5157  * o reserve space in the data space info for num_bytes
5158  * o reserve space in the metadata space info based on number of outstanding
5159  *   extents and how much csums will be needed
5160  * o add to the inodes ->delalloc_bytes
5161  * o add it to the fs_info's delalloc inodes list.
5162  *
5163  * This will return 0 for success and -ENOSPC if there is no space left.
5164  */
5165 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5166 {
5167         int ret;
5168
5169         ret = btrfs_check_data_free_space(inode, num_bytes);
5170         if (ret)
5171                 return ret;
5172
5173         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5174         if (ret) {
5175                 btrfs_free_reserved_data_space(inode, num_bytes);
5176                 return ret;
5177         }
5178
5179         return 0;
5180 }
5181
5182 /**
5183  * btrfs_delalloc_release_space - release data and metadata space for delalloc
5184  * @inode: inode we're releasing space for
5185  * @num_bytes: the number of bytes we want to free up
5186  *
5187  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
5188  * called in the case that we don't need the metadata AND data reservations
5189  * anymore.  So if there is an error or we insert an inline extent.
5190  *
5191  * This function will release the metadata space that was not used and will
5192  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5193  * list if there are no delalloc bytes left.
5194  */
5195 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5196 {
5197         btrfs_delalloc_release_metadata(inode, num_bytes);
5198         btrfs_free_reserved_data_space(inode, num_bytes);
5199 }
5200
5201 static int update_block_group(struct btrfs_root *root,
5202                               u64 bytenr, u64 num_bytes, int alloc)
5203 {
5204         struct btrfs_block_group_cache *cache = NULL;
5205         struct btrfs_fs_info *info = root->fs_info;
5206         u64 total = num_bytes;
5207         u64 old_val;
5208         u64 byte_in_group;
5209         int factor;
5210
5211         /* block accounting for super block */
5212         spin_lock(&info->delalloc_root_lock);
5213         old_val = btrfs_super_bytes_used(info->super_copy);
5214         if (alloc)
5215                 old_val += num_bytes;
5216         else
5217                 old_val -= num_bytes;
5218         btrfs_set_super_bytes_used(info->super_copy, old_val);
5219         spin_unlock(&info->delalloc_root_lock);
5220
5221         while (total) {
5222                 cache = btrfs_lookup_block_group(info, bytenr);
5223                 if (!cache)
5224                         return -ENOENT;
5225                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5226                                     BTRFS_BLOCK_GROUP_RAID1 |
5227                                     BTRFS_BLOCK_GROUP_RAID10))
5228                         factor = 2;
5229                 else
5230                         factor = 1;
5231                 /*
5232                  * If this block group has free space cache written out, we
5233                  * need to make sure to load it if we are removing space.  This
5234                  * is because we need the unpinning stage to actually add the
5235                  * space back to the block group, otherwise we will leak space.
5236                  */
5237                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5238                         cache_block_group(cache, 1);
5239
5240                 byte_in_group = bytenr - cache->key.objectid;
5241                 WARN_ON(byte_in_group > cache->key.offset);
5242
5243                 spin_lock(&cache->space_info->lock);
5244                 spin_lock(&cache->lock);
5245
5246                 if (btrfs_test_opt(root, SPACE_CACHE) &&
5247                     cache->disk_cache_state < BTRFS_DC_CLEAR)
5248                         cache->disk_cache_state = BTRFS_DC_CLEAR;
5249
5250                 cache->dirty = 1;
5251                 old_val = btrfs_block_group_used(&cache->item);
5252                 num_bytes = min(total, cache->key.offset - byte_in_group);
5253                 if (alloc) {
5254                         old_val += num_bytes;
5255                         btrfs_set_block_group_used(&cache->item, old_val);
5256                         cache->reserved -= num_bytes;
5257                         cache->space_info->bytes_reserved -= num_bytes;
5258                         cache->space_info->bytes_used += num_bytes;
5259                         cache->space_info->disk_used += num_bytes * factor;
5260                         spin_unlock(&cache->lock);
5261                         spin_unlock(&cache->space_info->lock);
5262                 } else {
5263                         old_val -= num_bytes;
5264                         btrfs_set_block_group_used(&cache->item, old_val);
5265                         cache->pinned += num_bytes;
5266                         cache->space_info->bytes_pinned += num_bytes;
5267                         cache->space_info->bytes_used -= num_bytes;
5268                         cache->space_info->disk_used -= num_bytes * factor;
5269                         spin_unlock(&cache->lock);
5270                         spin_unlock(&cache->space_info->lock);
5271
5272                         set_extent_dirty(info->pinned_extents,
5273                                          bytenr, bytenr + num_bytes - 1,
5274                                          GFP_NOFS | __GFP_NOFAIL);
5275                 }
5276                 btrfs_put_block_group(cache);
5277                 total -= num_bytes;
5278                 bytenr += num_bytes;
5279         }
5280         return 0;
5281 }
5282
5283 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5284 {
5285         struct btrfs_block_group_cache *cache;
5286         u64 bytenr;
5287
5288         spin_lock(&root->fs_info->block_group_cache_lock);
5289         bytenr = root->fs_info->first_logical_byte;
5290         spin_unlock(&root->fs_info->block_group_cache_lock);
5291
5292         if (bytenr < (u64)-1)
5293                 return bytenr;
5294
5295         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5296         if (!cache)
5297                 return 0;
5298
5299         bytenr = cache->key.objectid;
5300         btrfs_put_block_group(cache);
5301
5302         return bytenr;
5303 }
5304
5305 static int pin_down_extent(struct btrfs_root *root,
5306                            struct btrfs_block_group_cache *cache,
5307                            u64 bytenr, u64 num_bytes, int reserved)
5308 {
5309         spin_lock(&cache->space_info->lock);
5310         spin_lock(&cache->lock);
5311         cache->pinned += num_bytes;
5312         cache->space_info->bytes_pinned += num_bytes;
5313         if (reserved) {
5314                 cache->reserved -= num_bytes;
5315                 cache->space_info->bytes_reserved -= num_bytes;
5316         }
5317         spin_unlock(&cache->lock);
5318         spin_unlock(&cache->space_info->lock);
5319
5320         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5321                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5322         if (reserved)
5323                 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5324         return 0;
5325 }
5326
5327 /*
5328  * this function must be called within transaction
5329  */
5330 int btrfs_pin_extent(struct btrfs_root *root,
5331                      u64 bytenr, u64 num_bytes, int reserved)
5332 {
5333         struct btrfs_block_group_cache *cache;
5334
5335         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5336         BUG_ON(!cache); /* Logic error */
5337
5338         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5339
5340         btrfs_put_block_group(cache);
5341         return 0;
5342 }
5343
5344 /*
5345  * this function must be called within transaction
5346  */
5347 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5348                                     u64 bytenr, u64 num_bytes)
5349 {
5350         struct btrfs_block_group_cache *cache;
5351         int ret;
5352
5353         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5354         if (!cache)
5355                 return -EINVAL;
5356
5357         /*
5358          * pull in the free space cache (if any) so that our pin
5359          * removes the free space from the cache.  We have load_only set
5360          * to one because the slow code to read in the free extents does check
5361          * the pinned extents.
5362          */
5363         cache_block_group(cache, 1);
5364
5365         pin_down_extent(root, cache, bytenr, num_bytes, 0);
5366
5367         /* remove us from the free space cache (if we're there at all) */
5368         ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5369         btrfs_put_block_group(cache);
5370         return ret;
5371 }
5372
5373 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5374 {
5375         int ret;
5376         struct btrfs_block_group_cache *block_group;
5377         struct btrfs_caching_control *caching_ctl;
5378
5379         block_group = btrfs_lookup_block_group(root->fs_info, start);
5380         if (!block_group)
5381                 return -EINVAL;
5382
5383         cache_block_group(block_group, 0);
5384         caching_ctl = get_caching_control(block_group);
5385
5386         if (!caching_ctl) {
5387                 /* Logic error */
5388                 BUG_ON(!block_group_cache_done(block_group));
5389                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5390         } else {
5391                 mutex_lock(&caching_ctl->mutex);
5392
5393                 if (start >= caching_ctl->progress) {
5394                         ret = add_excluded_extent(root, start, num_bytes);
5395                 } else if (start + num_bytes <= caching_ctl->progress) {
5396                         ret = btrfs_remove_free_space(block_group,
5397                                                       start, num_bytes);
5398                 } else {
5399                         num_bytes = caching_ctl->progress - start;
5400                         ret = btrfs_remove_free_space(block_group,
5401                                                       start, num_bytes);
5402                         if (ret)
5403                                 goto out_lock;
5404
5405                         num_bytes = (start + num_bytes) -
5406                                 caching_ctl->progress;
5407                         start = caching_ctl->progress;
5408                         ret = add_excluded_extent(root, start, num_bytes);
5409                 }
5410 out_lock:
5411                 mutex_unlock(&caching_ctl->mutex);
5412                 put_caching_control(caching_ctl);
5413         }
5414         btrfs_put_block_group(block_group);
5415         return ret;
5416 }
5417
5418 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5419                                  struct extent_buffer *eb)
5420 {
5421         struct btrfs_file_extent_item *item;
5422         struct btrfs_key key;
5423         int found_type;
5424         int i;
5425
5426         if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5427                 return 0;
5428
5429         for (i = 0; i < btrfs_header_nritems(eb); i++) {
5430                 btrfs_item_key_to_cpu(eb, &key, i);
5431                 if (key.type != BTRFS_EXTENT_DATA_KEY)
5432                         continue;
5433                 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5434                 found_type = btrfs_file_extent_type(eb, item);
5435                 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5436                         continue;
5437                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5438                         continue;
5439                 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5440                 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5441                 __exclude_logged_extent(log, key.objectid, key.offset);
5442         }
5443
5444         return 0;
5445 }
5446
5447 /**
5448  * btrfs_update_reserved_bytes - update the block_group and space info counters
5449  * @cache:      The cache we are manipulating
5450  * @num_bytes:  The number of bytes in question
5451  * @reserve:    One of the reservation enums
5452  *
5453  * This is called by the allocator when it reserves space, or by somebody who is
5454  * freeing space that was never actually used on disk.  For example if you
5455  * reserve some space for a new leaf in transaction A and before transaction A
5456  * commits you free that leaf, you call this with reserve set to 0 in order to
5457  * clear the reservation.
5458  *
5459  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5460  * ENOSPC accounting.  For data we handle the reservation through clearing the
5461  * delalloc bits in the io_tree.  We have to do this since we could end up
5462  * allocating less disk space for the amount of data we have reserved in the
5463  * case of compression.
5464  *
5465  * If this is a reservation and the block group has become read only we cannot
5466  * make the reservation and return -EAGAIN, otherwise this function always
5467  * succeeds.
5468  */
5469 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5470                                        u64 num_bytes, int reserve)
5471 {
5472         struct btrfs_space_info *space_info = cache->space_info;
5473         int ret = 0;
5474
5475         spin_lock(&space_info->lock);
5476         spin_lock(&cache->lock);
5477         if (reserve != RESERVE_FREE) {
5478                 if (cache->ro) {
5479                         ret = -EAGAIN;
5480                 } else {
5481                         cache->reserved += num_bytes;
5482                         space_info->bytes_reserved += num_bytes;
5483                         if (reserve == RESERVE_ALLOC) {
5484                                 trace_btrfs_space_reservation(cache->fs_info,
5485                                                 "space_info", space_info->flags,
5486                                                 num_bytes, 0);
5487                                 space_info->bytes_may_use -= num_bytes;
5488                         }
5489                 }
5490         } else {
5491                 if (cache->ro)
5492                         space_info->bytes_readonly += num_bytes;
5493                 cache->reserved -= num_bytes;
5494                 space_info->bytes_reserved -= num_bytes;
5495         }
5496         spin_unlock(&cache->lock);
5497         spin_unlock(&space_info->lock);
5498         return ret;
5499 }
5500
5501 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5502                                 struct btrfs_root *root)
5503 {
5504         struct btrfs_fs_info *fs_info = root->fs_info;
5505         struct btrfs_caching_control *next;
5506         struct btrfs_caching_control *caching_ctl;
5507         struct btrfs_block_group_cache *cache;
5508         struct btrfs_space_info *space_info;
5509
5510         down_write(&fs_info->extent_commit_sem);
5511
5512         list_for_each_entry_safe(caching_ctl, next,
5513                                  &fs_info->caching_block_groups, list) {
5514                 cache = caching_ctl->block_group;
5515                 if (block_group_cache_done(cache)) {
5516                         cache->last_byte_to_unpin = (u64)-1;
5517                         list_del_init(&caching_ctl->list);
5518                         put_caching_control(caching_ctl);
5519                 } else {
5520                         cache->last_byte_to_unpin = caching_ctl->progress;
5521                 }
5522         }
5523
5524         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5525                 fs_info->pinned_extents = &fs_info->freed_extents[1];
5526         else
5527                 fs_info->pinned_extents = &fs_info->freed_extents[0];
5528
5529         up_write(&fs_info->extent_commit_sem);
5530
5531         list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5532                 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5533
5534         update_global_block_rsv(fs_info);
5535 }
5536
5537 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5538 {
5539         struct btrfs_fs_info *fs_info = root->fs_info;
5540         struct btrfs_block_group_cache *cache = NULL;
5541         struct btrfs_space_info *space_info;
5542         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5543         u64 len;
5544         bool readonly;
5545
5546         while (start <= end) {
5547                 readonly = false;
5548                 if (!cache ||
5549                     start >= cache->key.objectid + cache->key.offset) {
5550                         if (cache)
5551                                 btrfs_put_block_group(cache);
5552                         cache = btrfs_lookup_block_group(fs_info, start);
5553                         BUG_ON(!cache); /* Logic error */
5554                 }
5555
5556                 len = cache->key.objectid + cache->key.offset - start;
5557                 len = min(len, end + 1 - start);
5558
5559                 if (start < cache->last_byte_to_unpin) {
5560                         len = min(len, cache->last_byte_to_unpin - start);
5561                         btrfs_add_free_space(cache, start, len);
5562                 }
5563
5564                 start += len;
5565                 space_info = cache->space_info;
5566
5567                 spin_lock(&space_info->lock);
5568                 spin_lock(&cache->lock);
5569                 cache->pinned -= len;
5570                 space_info->bytes_pinned -= len;
5571                 if (cache->ro) {
5572                         space_info->bytes_readonly += len;
5573                         readonly = true;
5574                 }
5575                 spin_unlock(&cache->lock);
5576                 if (!readonly && global_rsv->space_info == space_info) {
5577                         spin_lock(&global_rsv->lock);
5578                         if (!global_rsv->full) {
5579                                 len = min(len, global_rsv->size -
5580                                           global_rsv->reserved);
5581                                 global_rsv->reserved += len;
5582                                 space_info->bytes_may_use += len;
5583                                 if (global_rsv->reserved >= global_rsv->size)
5584                                         global_rsv->full = 1;
5585                         }
5586                         spin_unlock(&global_rsv->lock);
5587                 }
5588                 spin_unlock(&space_info->lock);
5589         }
5590
5591         if (cache)
5592                 btrfs_put_block_group(cache);
5593         return 0;
5594 }
5595
5596 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5597                                struct btrfs_root *root)
5598 {
5599         struct btrfs_fs_info *fs_info = root->fs_info;
5600         struct extent_io_tree *unpin;
5601         u64 start;
5602         u64 end;
5603         int ret;
5604
5605         if (trans->aborted)
5606                 return 0;
5607
5608         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5609                 unpin = &fs_info->freed_extents[1];
5610         else
5611                 unpin = &fs_info->freed_extents[0];
5612
5613         while (1) {
5614                 ret = find_first_extent_bit(unpin, 0, &start, &end,
5615                                             EXTENT_DIRTY, NULL);
5616                 if (ret)
5617                         break;
5618
5619                 if (btrfs_test_opt(root, DISCARD))
5620                         ret = btrfs_discard_extent(root, start,
5621                                                    end + 1 - start, NULL);
5622
5623                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5624                 unpin_extent_range(root, start, end);
5625                 cond_resched();
5626         }
5627
5628         return 0;
5629 }
5630
5631 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5632                              u64 owner, u64 root_objectid)
5633 {
5634         struct btrfs_space_info *space_info;
5635         u64 flags;
5636
5637         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5638                 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5639                         flags = BTRFS_BLOCK_GROUP_SYSTEM;
5640                 else
5641                         flags = BTRFS_BLOCK_GROUP_METADATA;
5642         } else {
5643                 flags = BTRFS_BLOCK_GROUP_DATA;
5644         }
5645
5646         space_info = __find_space_info(fs_info, flags);
5647         BUG_ON(!space_info); /* Logic bug */
5648         percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5649 }
5650
5651
5652 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5653                                 struct btrfs_root *root,
5654                                 u64 bytenr, u64 num_bytes, u64 parent,
5655                                 u64 root_objectid, u64 owner_objectid,
5656                                 u64 owner_offset, int refs_to_drop,
5657                                 struct btrfs_delayed_extent_op *extent_op)
5658 {
5659         struct btrfs_key key;
5660         struct btrfs_path *path;
5661         struct btrfs_fs_info *info = root->fs_info;
5662         struct btrfs_root *extent_root = info->extent_root;
5663         struct extent_buffer *leaf;
5664         struct btrfs_extent_item *ei;
5665         struct btrfs_extent_inline_ref *iref;
5666         int ret;
5667         int is_data;
5668         int extent_slot = 0;
5669         int found_extent = 0;
5670         int num_to_del = 1;
5671         u32 item_size;
5672         u64 refs;
5673         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5674                                                  SKINNY_METADATA);
5675
5676         path = btrfs_alloc_path();
5677         if (!path)
5678                 return -ENOMEM;
5679
5680         path->reada = 1;
5681         path->leave_spinning = 1;
5682
5683         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5684         BUG_ON(!is_data && refs_to_drop != 1);
5685
5686         if (is_data)
5687                 skinny_metadata = 0;
5688
5689         ret = lookup_extent_backref(trans, extent_root, path, &iref,
5690                                     bytenr, num_bytes, parent,
5691                                     root_objectid, owner_objectid,
5692                                     owner_offset);
5693         if (ret == 0) {
5694                 extent_slot = path->slots[0];
5695                 while (extent_slot >= 0) {
5696                         btrfs_item_key_to_cpu(path->nodes[0], &key,
5697                                               extent_slot);
5698                         if (key.objectid != bytenr)
5699                                 break;
5700                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5701                             key.offset == num_bytes) {
5702                                 found_extent = 1;
5703                                 break;
5704                         }
5705                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
5706                             key.offset == owner_objectid) {
5707                                 found_extent = 1;
5708                                 break;
5709                         }
5710                         if (path->slots[0] - extent_slot > 5)
5711                                 break;
5712                         extent_slot--;
5713                 }
5714 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5715                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5716                 if (found_extent && item_size < sizeof(*ei))
5717                         found_extent = 0;
5718 #endif
5719                 if (!found_extent) {
5720                         BUG_ON(iref);
5721                         ret = remove_extent_backref(trans, extent_root, path,
5722                                                     NULL, refs_to_drop,
5723                                                     is_data);
5724                         if (ret) {
5725                                 btrfs_abort_transaction(trans, extent_root, ret);
5726                                 goto out;
5727                         }
5728                         btrfs_release_path(path);
5729                         path->leave_spinning = 1;
5730
5731                         key.objectid = bytenr;
5732                         key.type = BTRFS_EXTENT_ITEM_KEY;
5733                         key.offset = num_bytes;
5734
5735                         if (!is_data && skinny_metadata) {
5736                                 key.type = BTRFS_METADATA_ITEM_KEY;
5737                                 key.offset = owner_objectid;
5738                         }
5739
5740                         ret = btrfs_search_slot(trans, extent_root,
5741                                                 &key, path, -1, 1);
5742                         if (ret > 0 && skinny_metadata && path->slots[0]) {
5743                                 /*
5744                                  * Couldn't find our skinny metadata item,
5745                                  * see if we have ye olde extent item.
5746                                  */
5747                                 path->slots[0]--;
5748                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
5749                                                       path->slots[0]);
5750                                 if (key.objectid == bytenr &&
5751                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
5752                                     key.offset == num_bytes)
5753                                         ret = 0;
5754                         }
5755
5756                         if (ret > 0 && skinny_metadata) {
5757                                 skinny_metadata = false;
5758                                 key.type = BTRFS_EXTENT_ITEM_KEY;
5759                                 key.offset = num_bytes;
5760                                 btrfs_release_path(path);
5761                                 ret = btrfs_search_slot(trans, extent_root,
5762                                                         &key, path, -1, 1);
5763                         }
5764
5765                         if (ret) {
5766                                 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5767                                         ret, bytenr);
5768                                 if (ret > 0)
5769                                         btrfs_print_leaf(extent_root,
5770                                                          path->nodes[0]);
5771                         }
5772                         if (ret < 0) {
5773                                 btrfs_abort_transaction(trans, extent_root, ret);
5774                                 goto out;
5775                         }
5776                         extent_slot = path->slots[0];
5777                 }
5778         } else if (WARN_ON(ret == -ENOENT)) {
5779                 btrfs_print_leaf(extent_root, path->nodes[0]);
5780                 btrfs_err(info,
5781                         "unable to find ref byte nr %llu parent %llu root %llu  owner %llu offset %llu",
5782                         bytenr, parent, root_objectid, owner_objectid,
5783                         owner_offset);
5784         } else {
5785                 btrfs_abort_transaction(trans, extent_root, ret);
5786                 goto out;
5787         }
5788
5789         leaf = path->nodes[0];
5790         item_size = btrfs_item_size_nr(leaf, extent_slot);
5791 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5792         if (item_size < sizeof(*ei)) {
5793                 BUG_ON(found_extent || extent_slot != path->slots[0]);
5794                 ret = convert_extent_item_v0(trans, extent_root, path,
5795                                              owner_objectid, 0);
5796                 if (ret < 0) {
5797                         btrfs_abort_transaction(trans, extent_root, ret);
5798                         goto out;
5799                 }
5800
5801                 btrfs_release_path(path);
5802                 path->leave_spinning = 1;
5803
5804                 key.objectid = bytenr;
5805                 key.type = BTRFS_EXTENT_ITEM_KEY;
5806                 key.offset = num_bytes;
5807
5808                 ret = btrfs_search_slot(trans, extent_root, &key, path,
5809                                         -1, 1);
5810                 if (ret) {
5811                         btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5812                                 ret, bytenr);
5813                         btrfs_print_leaf(extent_root, path->nodes[0]);
5814                 }
5815                 if (ret < 0) {
5816                         btrfs_abort_transaction(trans, extent_root, ret);
5817                         goto out;
5818                 }
5819
5820                 extent_slot = path->slots[0];
5821                 leaf = path->nodes[0];
5822                 item_size = btrfs_item_size_nr(leaf, extent_slot);
5823         }
5824 #endif
5825         BUG_ON(item_size < sizeof(*ei));
5826         ei = btrfs_item_ptr(leaf, extent_slot,
5827                             struct btrfs_extent_item);
5828         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5829             key.type == BTRFS_EXTENT_ITEM_KEY) {
5830                 struct btrfs_tree_block_info *bi;
5831                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5832                 bi = (struct btrfs_tree_block_info *)(ei + 1);
5833                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5834         }
5835
5836         refs = btrfs_extent_refs(leaf, ei);
5837         if (refs < refs_to_drop) {
5838                 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5839                           "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5840                 ret = -EINVAL;
5841                 btrfs_abort_transaction(trans, extent_root, ret);
5842                 goto out;
5843         }
5844         refs -= refs_to_drop;
5845
5846         if (refs > 0) {
5847                 if (extent_op)
5848                         __run_delayed_extent_op(extent_op, leaf, ei);
5849                 /*
5850                  * In the case of inline back ref, reference count will
5851                  * be updated by remove_extent_backref
5852                  */
5853                 if (iref) {
5854                         BUG_ON(!found_extent);
5855                 } else {
5856                         btrfs_set_extent_refs(leaf, ei, refs);
5857                         btrfs_mark_buffer_dirty(leaf);
5858                 }
5859                 if (found_extent) {
5860                         ret = remove_extent_backref(trans, extent_root, path,
5861                                                     iref, refs_to_drop,
5862                                                     is_data);
5863                         if (ret) {
5864                                 btrfs_abort_transaction(trans, extent_root, ret);
5865                                 goto out;
5866                         }
5867                 }
5868                 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5869                                  root_objectid);
5870         } else {
5871                 if (found_extent) {
5872                         BUG_ON(is_data && refs_to_drop !=
5873                                extent_data_ref_count(root, path, iref));
5874                         if (iref) {
5875                                 BUG_ON(path->slots[0] != extent_slot);
5876                         } else {
5877                                 BUG_ON(path->slots[0] != extent_slot + 1);
5878                                 path->slots[0] = extent_slot;
5879                                 num_to_del = 2;
5880                         }
5881                 }
5882
5883                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5884                                       num_to_del);
5885                 if (ret) {
5886                         btrfs_abort_transaction(trans, extent_root, ret);
5887                         goto out;
5888                 }
5889                 btrfs_release_path(path);
5890
5891                 if (is_data) {
5892                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5893                         if (ret) {
5894                                 btrfs_abort_transaction(trans, extent_root, ret);
5895                                 goto out;
5896                         }
5897                 }
5898
5899                 ret = update_block_group(root, bytenr, num_bytes, 0);
5900                 if (ret) {
5901                         btrfs_abort_transaction(trans, extent_root, ret);
5902                         goto out;
5903                 }
5904         }
5905 out:
5906         btrfs_free_path(path);
5907         return ret;
5908 }
5909
5910 /*
5911  * when we free an block, it is possible (and likely) that we free the last
5912  * delayed ref for that extent as well.  This searches the delayed ref tree for
5913  * a given extent, and if there are no other delayed refs to be processed, it
5914  * removes it from the tree.
5915  */
5916 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5917                                       struct btrfs_root *root, u64 bytenr)
5918 {
5919         struct btrfs_delayed_ref_head *head;
5920         struct btrfs_delayed_ref_root *delayed_refs;
5921         struct btrfs_delayed_ref_node *ref;
5922         struct rb_node *node;
5923         int ret = 0;
5924
5925         delayed_refs = &trans->transaction->delayed_refs;
5926         spin_lock(&delayed_refs->lock);
5927         head = btrfs_find_delayed_ref_head(trans, bytenr);
5928         if (!head)
5929                 goto out;
5930
5931         node = rb_prev(&head->node.rb_node);
5932         if (!node)
5933                 goto out;
5934
5935         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5936
5937         /* there are still entries for this ref, we can't drop it */
5938         if (ref->bytenr == bytenr)
5939                 goto out;
5940
5941         if (head->extent_op) {
5942                 if (!head->must_insert_reserved)
5943                         goto out;
5944                 btrfs_free_delayed_extent_op(head->extent_op);
5945                 head->extent_op = NULL;
5946         }
5947
5948         /*
5949          * waiting for the lock here would deadlock.  If someone else has it
5950          * locked they are already in the process of dropping it anyway
5951          */
5952         if (!mutex_trylock(&head->mutex))
5953                 goto out;
5954
5955         /*
5956          * at this point we have a head with no other entries.  Go
5957          * ahead and process it.
5958          */
5959         head->node.in_tree = 0;
5960         rb_erase(&head->node.rb_node, &delayed_refs->root);
5961
5962         delayed_refs->num_entries--;
5963
5964         /*
5965          * we don't take a ref on the node because we're removing it from the
5966          * tree, so we just steal the ref the tree was holding.
5967          */
5968         delayed_refs->num_heads--;
5969         if (list_empty(&head->cluster))
5970                 delayed_refs->num_heads_ready--;
5971
5972         list_del_init(&head->cluster);
5973         spin_unlock(&delayed_refs->lock);
5974
5975         BUG_ON(head->extent_op);
5976         if (head->must_insert_reserved)
5977                 ret = 1;
5978
5979         mutex_unlock(&head->mutex);
5980         btrfs_put_delayed_ref(&head->node);
5981         return ret;
5982 out:
5983         spin_unlock(&delayed_refs->lock);
5984         return 0;
5985 }
5986
5987 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5988                            struct btrfs_root *root,
5989                            struct extent_buffer *buf,
5990                            u64 parent, int last_ref)
5991 {
5992         struct btrfs_block_group_cache *cache = NULL;
5993         int pin = 1;
5994         int ret;
5995
5996         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5997                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5998                                         buf->start, buf->len,
5999                                         parent, root->root_key.objectid,
6000                                         btrfs_header_level(buf),
6001                                         BTRFS_DROP_DELAYED_REF, NULL, 0);
6002                 BUG_ON(ret); /* -ENOMEM */
6003         }
6004
6005         if (!last_ref)
6006                 return;
6007
6008         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6009
6010         if (btrfs_header_generation(buf) == trans->transid) {
6011                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6012                         ret = check_ref_cleanup(trans, root, buf->start);
6013                         if (!ret)
6014                                 goto out;
6015                 }
6016
6017                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6018                         pin_down_extent(root, cache, buf->start, buf->len, 1);
6019                         goto out;
6020                 }
6021
6022                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6023
6024                 btrfs_add_free_space(cache, buf->start, buf->len);
6025                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
6026                 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6027                 pin = 0;
6028         }
6029 out:
6030         if (pin)
6031                 add_pinned_bytes(root->fs_info, buf->len,
6032                                  btrfs_header_level(buf),
6033                                  root->root_key.objectid);
6034
6035         /*
6036          * Deleting the buffer, clear the corrupt flag since it doesn't matter
6037          * anymore.
6038          */
6039         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6040         btrfs_put_block_group(cache);
6041 }
6042
6043 /* Can return -ENOMEM */
6044 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6045                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6046                       u64 owner, u64 offset, int for_cow)
6047 {
6048         int ret;
6049         struct btrfs_fs_info *fs_info = root->fs_info;
6050
6051         add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6052
6053         /*
6054          * tree log blocks never actually go into the extent allocation
6055          * tree, just update pinning info and exit early.
6056          */
6057         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6058                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6059                 /* unlocks the pinned mutex */
6060                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6061                 ret = 0;
6062         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6063                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6064                                         num_bytes,
6065                                         parent, root_objectid, (int)owner,
6066                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6067         } else {
6068                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6069                                                 num_bytes,
6070                                                 parent, root_objectid, owner,
6071                                                 offset, BTRFS_DROP_DELAYED_REF,
6072                                                 NULL, for_cow);
6073         }
6074         return ret;
6075 }
6076
6077 static u64 stripe_align(struct btrfs_root *root,
6078                         struct btrfs_block_group_cache *cache,
6079                         u64 val, u64 num_bytes)
6080 {
6081         u64 ret = ALIGN(val, root->stripesize);
6082         return ret;
6083 }
6084
6085 /*
6086  * when we wait for progress in the block group caching, its because
6087  * our allocation attempt failed at least once.  So, we must sleep
6088  * and let some progress happen before we try again.
6089  *
6090  * This function will sleep at least once waiting for new free space to
6091  * show up, and then it will check the block group free space numbers
6092  * for our min num_bytes.  Another option is to have it go ahead
6093  * and look in the rbtree for a free extent of a given size, but this
6094  * is a good start.
6095  *
6096  * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6097  * any of the information in this block group.
6098  */
6099 static noinline void
6100 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6101                                 u64 num_bytes)
6102 {
6103         struct btrfs_caching_control *caching_ctl;
6104
6105         caching_ctl = get_caching_control(cache);
6106         if (!caching_ctl)
6107                 return;
6108
6109         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6110                    (cache->free_space_ctl->free_space >= num_bytes));
6111
6112         put_caching_control(caching_ctl);
6113 }
6114
6115 static noinline int
6116 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6117 {
6118         struct btrfs_caching_control *caching_ctl;
6119         int ret = 0;
6120
6121         caching_ctl = get_caching_control(cache);
6122         if (!caching_ctl)
6123                 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6124
6125         wait_event(caching_ctl->wait, block_group_cache_done(cache));
6126         if (cache->cached == BTRFS_CACHE_ERROR)
6127                 ret = -EIO;
6128         put_caching_control(caching_ctl);
6129         return ret;
6130 }
6131
6132 int __get_raid_index(u64 flags)
6133 {
6134         if (flags & BTRFS_BLOCK_GROUP_RAID10)
6135                 return BTRFS_RAID_RAID10;
6136         else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6137                 return BTRFS_RAID_RAID1;
6138         else if (flags & BTRFS_BLOCK_GROUP_DUP)
6139                 return BTRFS_RAID_DUP;
6140         else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6141                 return BTRFS_RAID_RAID0;
6142         else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6143                 return BTRFS_RAID_RAID5;
6144         else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6145                 return BTRFS_RAID_RAID6;
6146
6147         return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6148 }
6149
6150 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6151 {
6152         return __get_raid_index(cache->flags);
6153 }
6154
6155 enum btrfs_loop_type {
6156         LOOP_CACHING_NOWAIT = 0,
6157         LOOP_CACHING_WAIT = 1,
6158         LOOP_ALLOC_CHUNK = 2,
6159         LOOP_NO_EMPTY_SIZE = 3,
6160 };
6161
6162 /*
6163  * walks the btree of allocated extents and find a hole of a given size.
6164  * The key ins is changed to record the hole:
6165  * ins->objectid == start position
6166  * ins->flags = BTRFS_EXTENT_ITEM_KEY
6167  * ins->offset == the size of the hole.
6168  * Any available blocks before search_start are skipped.
6169  *
6170  * If there is no suitable free space, we will record the max size of
6171  * the free space extent currently.
6172  */
6173 static noinline int find_free_extent(struct btrfs_root *orig_root,
6174                                      u64 num_bytes, u64 empty_size,
6175                                      u64 hint_byte, struct btrfs_key *ins,
6176                                      u64 flags)
6177 {
6178         int ret = 0;
6179         struct btrfs_root *root = orig_root->fs_info->extent_root;
6180         struct btrfs_free_cluster *last_ptr = NULL;
6181         struct btrfs_block_group_cache *block_group = NULL;
6182         struct btrfs_block_group_cache *used_block_group;
6183         u64 search_start = 0;
6184         u64 max_extent_size = 0;
6185         int empty_cluster = 2 * 1024 * 1024;
6186         struct btrfs_space_info *space_info;
6187         int loop = 0;
6188         int index = __get_raid_index(flags);
6189         int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6190                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6191         bool found_uncached_bg = false;
6192         bool failed_cluster_refill = false;
6193         bool failed_alloc = false;
6194         bool use_cluster = true;
6195         bool have_caching_bg = false;
6196
6197         WARN_ON(num_bytes < root->sectorsize);
6198         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6199         ins->objectid = 0;
6200         ins->offset = 0;
6201
6202         trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6203
6204         space_info = __find_space_info(root->fs_info, flags);
6205         if (!space_info) {
6206                 btrfs_err(root->fs_info, "No space info for %llu", flags);
6207                 return -ENOSPC;
6208         }
6209
6210         /*
6211          * If the space info is for both data and metadata it means we have a
6212          * small filesystem and we can't use the clustering stuff.
6213          */
6214         if (btrfs_mixed_space_info(space_info))
6215                 use_cluster = false;
6216
6217         if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6218                 last_ptr = &root->fs_info->meta_alloc_cluster;
6219                 if (!btrfs_test_opt(root, SSD))
6220                         empty_cluster = 64 * 1024;
6221         }
6222
6223         if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6224             btrfs_test_opt(root, SSD)) {
6225                 last_ptr = &root->fs_info->data_alloc_cluster;
6226         }
6227
6228         if (last_ptr) {
6229                 spin_lock(&last_ptr->lock);
6230                 if (last_ptr->block_group)
6231                         hint_byte = last_ptr->window_start;
6232                 spin_unlock(&last_ptr->lock);
6233         }
6234
6235         search_start = max(search_start, first_logical_byte(root, 0));
6236         search_start = max(search_start, hint_byte);
6237
6238         if (!last_ptr)
6239                 empty_cluster = 0;
6240
6241         if (search_start == hint_byte) {
6242                 block_group = btrfs_lookup_block_group(root->fs_info,
6243                                                        search_start);
6244                 used_block_group = block_group;
6245                 /*
6246                  * we don't want to use the block group if it doesn't match our
6247                  * allocation bits, or if its not cached.
6248                  *
6249                  * However if we are re-searching with an ideal block group
6250                  * picked out then we don't care that the block group is cached.
6251                  */
6252                 if (block_group && block_group_bits(block_group, flags) &&
6253                     block_group->cached != BTRFS_CACHE_NO) {
6254                         down_read(&space_info->groups_sem);
6255                         if (list_empty(&block_group->list) ||
6256                             block_group->ro) {
6257                                 /*
6258                                  * someone is removing this block group,
6259                                  * we can't jump into the have_block_group
6260                                  * target because our list pointers are not
6261                                  * valid
6262                                  */
6263                                 btrfs_put_block_group(block_group);
6264                                 up_read(&space_info->groups_sem);
6265                         } else {
6266                                 index = get_block_group_index(block_group);
6267                                 goto have_block_group;
6268                         }
6269                 } else if (block_group) {
6270                         btrfs_put_block_group(block_group);
6271                 }
6272         }
6273 search:
6274         have_caching_bg = false;
6275         down_read(&space_info->groups_sem);
6276         list_for_each_entry(block_group, &space_info->block_groups[index],
6277                             list) {
6278                 u64 offset;
6279                 int cached;
6280
6281                 used_block_group = block_group;
6282                 btrfs_get_block_group(block_group);
6283                 search_start = block_group->key.objectid;
6284
6285                 /*
6286                  * this can happen if we end up cycling through all the
6287                  * raid types, but we want to make sure we only allocate
6288                  * for the proper type.
6289                  */
6290                 if (!block_group_bits(block_group, flags)) {
6291                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
6292                                 BTRFS_BLOCK_GROUP_RAID1 |
6293                                 BTRFS_BLOCK_GROUP_RAID5 |
6294                                 BTRFS_BLOCK_GROUP_RAID6 |
6295                                 BTRFS_BLOCK_GROUP_RAID10;
6296
6297                         /*
6298                          * if they asked for extra copies and this block group
6299                          * doesn't provide them, bail.  This does allow us to
6300                          * fill raid0 from raid1.
6301                          */
6302                         if ((flags & extra) && !(block_group->flags & extra))
6303                                 goto loop;
6304                 }
6305
6306 have_block_group:
6307                 cached = block_group_cache_done(block_group);
6308                 if (unlikely(!cached)) {
6309                         found_uncached_bg = true;
6310                         ret = cache_block_group(block_group, 0);
6311                         BUG_ON(ret < 0);
6312                         ret = 0;
6313                 }
6314
6315                 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6316                         goto loop;
6317                 if (unlikely(block_group->ro))
6318                         goto loop;
6319
6320                 /*
6321                  * Ok we want to try and use the cluster allocator, so
6322                  * lets look there
6323                  */
6324                 if (last_ptr) {
6325                         unsigned long aligned_cluster;
6326                         /*
6327                          * the refill lock keeps out other
6328                          * people trying to start a new cluster
6329                          */
6330                         spin_lock(&last_ptr->refill_lock);
6331                         used_block_group = last_ptr->block_group;
6332                         if (used_block_group != block_group &&
6333                             (!used_block_group ||
6334                              used_block_group->ro ||
6335                              !block_group_bits(used_block_group, flags))) {
6336                                 used_block_group = block_group;
6337                                 goto refill_cluster;
6338                         }
6339
6340                         if (used_block_group != block_group)
6341                                 btrfs_get_block_group(used_block_group);
6342
6343                         offset = btrfs_alloc_from_cluster(used_block_group,
6344                                                 last_ptr,
6345                                                 num_bytes,
6346                                                 used_block_group->key.objectid,
6347                                                 &max_extent_size);
6348                         if (offset) {
6349                                 /* we have a block, we're done */
6350                                 spin_unlock(&last_ptr->refill_lock);
6351                                 trace_btrfs_reserve_extent_cluster(root,
6352                                         block_group, search_start, num_bytes);
6353                                 goto checks;
6354                         }
6355
6356                         WARN_ON(last_ptr->block_group != used_block_group);
6357                         if (used_block_group != block_group) {
6358                                 btrfs_put_block_group(used_block_group);
6359                                 used_block_group = block_group;
6360                         }
6361 refill_cluster:
6362                         BUG_ON(used_block_group != block_group);
6363                         /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6364                          * set up a new clusters, so lets just skip it
6365                          * and let the allocator find whatever block
6366                          * it can find.  If we reach this point, we
6367                          * will have tried the cluster allocator
6368                          * plenty of times and not have found
6369                          * anything, so we are likely way too
6370                          * fragmented for the clustering stuff to find
6371                          * anything.
6372                          *
6373                          * However, if the cluster is taken from the
6374                          * current block group, release the cluster
6375                          * first, so that we stand a better chance of
6376                          * succeeding in the unclustered
6377                          * allocation.  */
6378                         if (loop >= LOOP_NO_EMPTY_SIZE &&
6379                             last_ptr->block_group != block_group) {
6380                                 spin_unlock(&last_ptr->refill_lock);
6381                                 goto unclustered_alloc;
6382                         }
6383
6384                         /*
6385                          * this cluster didn't work out, free it and
6386                          * start over
6387                          */
6388                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
6389
6390                         if (loop >= LOOP_NO_EMPTY_SIZE) {
6391                                 spin_unlock(&last_ptr->refill_lock);
6392                                 goto unclustered_alloc;
6393                         }
6394
6395                         aligned_cluster = max_t(unsigned long,
6396                                                 empty_cluster + empty_size,
6397                                               block_group->full_stripe_len);
6398
6399                         /* allocate a cluster in this block group */
6400                         ret = btrfs_find_space_cluster(root, block_group,
6401                                                        last_ptr, search_start,
6402                                                        num_bytes,
6403                                                        aligned_cluster);
6404                         if (ret == 0) {
6405                                 /*
6406                                  * now pull our allocation out of this
6407                                  * cluster
6408                                  */
6409                                 offset = btrfs_alloc_from_cluster(block_group,
6410                                                         last_ptr,
6411                                                         num_bytes,
6412                                                         search_start,
6413                                                         &max_extent_size);
6414                                 if (offset) {
6415                                         /* we found one, proceed */
6416                                         spin_unlock(&last_ptr->refill_lock);
6417                                         trace_btrfs_reserve_extent_cluster(root,
6418                                                 block_group, search_start,
6419                                                 num_bytes);
6420                                         goto checks;
6421                                 }
6422                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
6423                                    && !failed_cluster_refill) {
6424                                 spin_unlock(&last_ptr->refill_lock);
6425
6426                                 failed_cluster_refill = true;
6427                                 wait_block_group_cache_progress(block_group,
6428                                        num_bytes + empty_cluster + empty_size);
6429                                 goto have_block_group;
6430                         }
6431
6432                         /*
6433                          * at this point we either didn't find a cluster
6434                          * or we weren't able to allocate a block from our
6435                          * cluster.  Free the cluster we've been trying
6436                          * to use, and go to the next block group
6437                          */
6438                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
6439                         spin_unlock(&last_ptr->refill_lock);
6440                         goto loop;
6441                 }
6442
6443 unclustered_alloc:
6444                 spin_lock(&block_group->free_space_ctl->tree_lock);
6445                 if (cached &&
6446                     block_group->free_space_ctl->free_space <
6447                     num_bytes + empty_cluster + empty_size) {
6448                         if (block_group->free_space_ctl->free_space >
6449                             max_extent_size)
6450                                 max_extent_size =
6451                                         block_group->free_space_ctl->free_space;
6452                         spin_unlock(&block_group->free_space_ctl->tree_lock);
6453                         goto loop;
6454                 }
6455                 spin_unlock(&block_group->free_space_ctl->tree_lock);
6456
6457                 offset = btrfs_find_space_for_alloc(block_group, search_start,
6458                                                     num_bytes, empty_size,
6459                                                     &max_extent_size);
6460                 /*
6461                  * If we didn't find a chunk, and we haven't failed on this
6462                  * block group before, and this block group is in the middle of
6463                  * caching and we are ok with waiting, then go ahead and wait
6464                  * for progress to be made, and set failed_alloc to true.
6465                  *
6466                  * If failed_alloc is true then we've already waited on this
6467                  * block group once and should move on to the next block group.
6468                  */
6469                 if (!offset && !failed_alloc && !cached &&
6470                     loop > LOOP_CACHING_NOWAIT) {
6471                         wait_block_group_cache_progress(block_group,
6472                                                 num_bytes + empty_size);
6473                         failed_alloc = true;
6474                         goto have_block_group;
6475                 } else if (!offset) {
6476                         if (!cached)
6477                                 have_caching_bg = true;
6478                         goto loop;
6479                 }
6480 checks:
6481                 search_start = stripe_align(root, used_block_group,
6482                                             offset, num_bytes);
6483
6484                 /* move on to the next group */
6485                 if (search_start + num_bytes >
6486                     used_block_group->key.objectid + used_block_group->key.offset) {
6487                         btrfs_add_free_space(used_block_group, offset, num_bytes);
6488                         goto loop;
6489                 }
6490
6491                 if (offset < search_start)
6492                         btrfs_add_free_space(used_block_group, offset,
6493                                              search_start - offset);
6494                 BUG_ON(offset > search_start);
6495
6496                 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6497                                                   alloc_type);
6498                 if (ret == -EAGAIN) {
6499                         btrfs_add_free_space(used_block_group, offset, num_bytes);
6500                         goto loop;
6501                 }
6502
6503                 /* we are all good, lets return */
6504                 ins->objectid = search_start;
6505                 ins->offset = num_bytes;
6506
6507                 trace_btrfs_reserve_extent(orig_root, block_group,
6508                                            search_start, num_bytes);
6509                 if (used_block_group != block_group)
6510                         btrfs_put_block_group(used_block_group);
6511                 btrfs_put_block_group(block_group);
6512                 break;
6513 loop:
6514                 failed_cluster_refill = false;
6515                 failed_alloc = false;
6516                 BUG_ON(index != get_block_group_index(block_group));
6517                 if (used_block_group != block_group)
6518                         btrfs_put_block_group(used_block_group);
6519                 btrfs_put_block_group(block_group);
6520         }
6521         up_read(&space_info->groups_sem);
6522
6523         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6524                 goto search;
6525
6526         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6527                 goto search;
6528
6529         /*
6530          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6531          *                      caching kthreads as we move along
6532          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6533          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6534          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6535          *                      again
6536          */
6537         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6538                 index = 0;
6539                 loop++;
6540                 if (loop == LOOP_ALLOC_CHUNK) {
6541                         struct btrfs_trans_handle *trans;
6542
6543                         trans = btrfs_join_transaction(root);
6544                         if (IS_ERR(trans)) {
6545                                 ret = PTR_ERR(trans);
6546                                 goto out;
6547                         }
6548
6549                         ret = do_chunk_alloc(trans, root, flags,
6550                                              CHUNK_ALLOC_FORCE);
6551                         /*
6552                          * Do not bail out on ENOSPC since we
6553                          * can do more things.
6554                          */
6555                         if (ret < 0 && ret != -ENOSPC)
6556                                 btrfs_abort_transaction(trans,
6557                                                         root, ret);
6558                         else
6559                                 ret = 0;
6560                         btrfs_end_transaction(trans, root);
6561                         if (ret)
6562                                 goto out;
6563                 }
6564
6565                 if (loop == LOOP_NO_EMPTY_SIZE) {
6566                         empty_size = 0;
6567                         empty_cluster = 0;
6568                 }
6569
6570                 goto search;
6571         } else if (!ins->objectid) {
6572                 ret = -ENOSPC;
6573         } else if (ins->objectid) {
6574                 ret = 0;
6575         }
6576 out:
6577         if (ret == -ENOSPC)
6578                 ins->offset = max_extent_size;
6579         return ret;
6580 }
6581
6582 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6583                             int dump_block_groups)
6584 {
6585         struct btrfs_block_group_cache *cache;
6586         int index = 0;
6587
6588         spin_lock(&info->lock);
6589         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6590                info->flags,
6591                info->total_bytes - info->bytes_used - info->bytes_pinned -
6592                info->bytes_reserved - info->bytes_readonly,
6593                (info->full) ? "" : "not ");
6594         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6595                "reserved=%llu, may_use=%llu, readonly=%llu\n",
6596                info->total_bytes, info->bytes_used, info->bytes_pinned,
6597                info->bytes_reserved, info->bytes_may_use,
6598                info->bytes_readonly);
6599         spin_unlock(&info->lock);
6600
6601         if (!dump_block_groups)
6602                 return;
6603
6604         down_read(&info->groups_sem);
6605 again:
6606         list_for_each_entry(cache, &info->block_groups[index], list) {
6607                 spin_lock(&cache->lock);
6608                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6609                        cache->key.objectid, cache->key.offset,
6610                        btrfs_block_group_used(&cache->item), cache->pinned,
6611                        cache->reserved, cache->ro ? "[readonly]" : "");
6612                 btrfs_dump_free_space(cache, bytes);
6613                 spin_unlock(&cache->lock);
6614         }
6615         if (++index < BTRFS_NR_RAID_TYPES)
6616                 goto again;
6617         up_read(&info->groups_sem);
6618 }
6619
6620 int btrfs_reserve_extent(struct btrfs_root *root,
6621                          u64 num_bytes, u64 min_alloc_size,
6622                          u64 empty_size, u64 hint_byte,
6623                          struct btrfs_key *ins, int is_data)
6624 {
6625         bool final_tried = false;
6626         u64 flags;
6627         int ret;
6628
6629         flags = btrfs_get_alloc_profile(root, is_data);
6630 again:
6631         WARN_ON(num_bytes < root->sectorsize);
6632         ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6633                                flags);
6634
6635         if (ret == -ENOSPC) {
6636                 if (!final_tried && ins->offset) {
6637                         num_bytes = min(num_bytes >> 1, ins->offset);
6638                         num_bytes = round_down(num_bytes, root->sectorsize);
6639                         num_bytes = max(num_bytes, min_alloc_size);
6640                         if (num_bytes == min_alloc_size)
6641                                 final_tried = true;
6642                         goto again;
6643                 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6644                         struct btrfs_space_info *sinfo;
6645
6646                         sinfo = __find_space_info(root->fs_info, flags);
6647                         btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6648                                 flags, num_bytes);
6649                         if (sinfo)
6650                                 dump_space_info(sinfo, num_bytes, 1);
6651                 }
6652         }
6653
6654         return ret;
6655 }
6656
6657 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6658                                         u64 start, u64 len, int pin)
6659 {
6660         struct btrfs_block_group_cache *cache;
6661         int ret = 0;
6662
6663         cache = btrfs_lookup_block_group(root->fs_info, start);
6664         if (!cache) {
6665                 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6666                         start);
6667                 return -ENOSPC;
6668         }
6669
6670         if (btrfs_test_opt(root, DISCARD))
6671                 ret = btrfs_discard_extent(root, start, len, NULL);
6672
6673         if (pin)
6674                 pin_down_extent(root, cache, start, len, 1);
6675         else {
6676                 btrfs_add_free_space(cache, start, len);
6677                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6678         }
6679         btrfs_put_block_group(cache);
6680
6681         trace_btrfs_reserved_extent_free(root, start, len);
6682
6683         return ret;
6684 }
6685
6686 int btrfs_free_reserved_extent(struct btrfs_root *root,
6687                                         u64 start, u64 len)
6688 {
6689         return __btrfs_free_reserved_extent(root, start, len, 0);
6690 }
6691
6692 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6693                                        u64 start, u64 len)
6694 {
6695         return __btrfs_free_reserved_extent(root, start, len, 1);
6696 }
6697
6698 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6699                                       struct btrfs_root *root,
6700                                       u64 parent, u64 root_objectid,
6701                                       u64 flags, u64 owner, u64 offset,
6702                                       struct btrfs_key *ins, int ref_mod)
6703 {
6704         int ret;
6705         struct btrfs_fs_info *fs_info = root->fs_info;
6706         struct btrfs_extent_item *extent_item;
6707         struct btrfs_extent_inline_ref *iref;
6708         struct btrfs_path *path;
6709         struct extent_buffer *leaf;
6710         int type;
6711         u32 size;
6712
6713         if (parent > 0)
6714                 type = BTRFS_SHARED_DATA_REF_KEY;
6715         else
6716                 type = BTRFS_EXTENT_DATA_REF_KEY;
6717
6718         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6719
6720         path = btrfs_alloc_path();
6721         if (!path)
6722                 return -ENOMEM;
6723
6724         path->leave_spinning = 1;
6725         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6726                                       ins, size);
6727         if (ret) {
6728                 btrfs_free_path(path);
6729                 return ret;
6730         }
6731
6732         leaf = path->nodes[0];
6733         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6734                                      struct btrfs_extent_item);
6735         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6736         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6737         btrfs_set_extent_flags(leaf, extent_item,
6738                                flags | BTRFS_EXTENT_FLAG_DATA);
6739
6740         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6741         btrfs_set_extent_inline_ref_type(leaf, iref, type);
6742         if (parent > 0) {
6743                 struct btrfs_shared_data_ref *ref;
6744                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6745                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6746                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6747         } else {
6748                 struct btrfs_extent_data_ref *ref;
6749                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6750                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6751                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6752                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6753                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6754         }
6755
6756         btrfs_mark_buffer_dirty(path->nodes[0]);
6757         btrfs_free_path(path);
6758
6759         ret = update_block_group(root, ins->objectid, ins->offset, 1);
6760         if (ret) { /* -ENOENT, logic error */
6761                 btrfs_err(fs_info, "update block group failed for %llu %llu",
6762                         ins->objectid, ins->offset);
6763                 BUG();
6764         }
6765         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6766         return ret;
6767 }
6768
6769 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6770                                      struct btrfs_root *root,
6771                                      u64 parent, u64 root_objectid,
6772                                      u64 flags, struct btrfs_disk_key *key,
6773                                      int level, struct btrfs_key *ins)
6774 {
6775         int ret;
6776         struct btrfs_fs_info *fs_info = root->fs_info;
6777         struct btrfs_extent_item *extent_item;
6778         struct btrfs_tree_block_info *block_info;
6779         struct btrfs_extent_inline_ref *iref;
6780         struct btrfs_path *path;
6781         struct extent_buffer *leaf;
6782         u32 size = sizeof(*extent_item) + sizeof(*iref);
6783         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6784                                                  SKINNY_METADATA);
6785
6786         if (!skinny_metadata)
6787                 size += sizeof(*block_info);
6788
6789         path = btrfs_alloc_path();
6790         if (!path) {
6791                 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6792                                                    root->leafsize);
6793                 return -ENOMEM;
6794         }
6795
6796         path->leave_spinning = 1;
6797         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6798                                       ins, size);
6799         if (ret) {
6800                 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6801                                                    root->leafsize);
6802                 btrfs_free_path(path);
6803                 return ret;
6804         }
6805
6806         leaf = path->nodes[0];
6807         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6808                                      struct btrfs_extent_item);
6809         btrfs_set_extent_refs(leaf, extent_item, 1);
6810         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6811         btrfs_set_extent_flags(leaf, extent_item,
6812                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6813
6814         if (skinny_metadata) {
6815                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6816         } else {
6817                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6818                 btrfs_set_tree_block_key(leaf, block_info, key);
6819                 btrfs_set_tree_block_level(leaf, block_info, level);
6820                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6821         }
6822
6823         if (parent > 0) {
6824                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6825                 btrfs_set_extent_inline_ref_type(leaf, iref,
6826                                                  BTRFS_SHARED_BLOCK_REF_KEY);
6827                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6828         } else {
6829                 btrfs_set_extent_inline_ref_type(leaf, iref,
6830                                                  BTRFS_TREE_BLOCK_REF_KEY);
6831                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6832         }
6833
6834         btrfs_mark_buffer_dirty(leaf);
6835         btrfs_free_path(path);
6836
6837         ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6838         if (ret) { /* -ENOENT, logic error */
6839                 btrfs_err(fs_info, "update block group failed for %llu %llu",
6840                         ins->objectid, ins->offset);
6841                 BUG();
6842         }
6843
6844         trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6845         return ret;
6846 }
6847
6848 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6849                                      struct btrfs_root *root,
6850                                      u64 root_objectid, u64 owner,
6851                                      u64 offset, struct btrfs_key *ins)
6852 {
6853         int ret;
6854
6855         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6856
6857         ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6858                                          ins->offset, 0,
6859                                          root_objectid, owner, offset,
6860                                          BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6861         return ret;
6862 }
6863
6864 /*
6865  * this is used by the tree logging recovery code.  It records that
6866  * an extent has been allocated and makes sure to clear the free
6867  * space cache bits as well
6868  */
6869 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6870                                    struct btrfs_root *root,
6871                                    u64 root_objectid, u64 owner, u64 offset,
6872                                    struct btrfs_key *ins)
6873 {
6874         int ret;
6875         struct btrfs_block_group_cache *block_group;
6876
6877         /*
6878          * Mixed block groups will exclude before processing the log so we only
6879          * need to do the exlude dance if this fs isn't mixed.
6880          */
6881         if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6882                 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6883                 if (ret)
6884                         return ret;
6885         }
6886
6887         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6888         if (!block_group)
6889                 return -EINVAL;
6890
6891         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6892                                           RESERVE_ALLOC_NO_ACCOUNT);
6893         BUG_ON(ret); /* logic error */
6894         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6895                                          0, owner, offset, ins, 1);
6896         btrfs_put_block_group(block_group);
6897         return ret;
6898 }
6899
6900 static struct extent_buffer *
6901 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6902                       u64 bytenr, u32 blocksize, int level)
6903 {
6904         struct extent_buffer *buf;
6905
6906         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6907         if (!buf)
6908                 return ERR_PTR(-ENOMEM);
6909         btrfs_set_header_generation(buf, trans->transid);
6910         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6911         btrfs_tree_lock(buf);
6912         clean_tree_block(trans, root, buf);
6913         clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6914
6915         btrfs_set_lock_blocking(buf);
6916         btrfs_set_buffer_uptodate(buf);
6917
6918         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6919                 /*
6920                  * we allow two log transactions at a time, use different
6921                  * EXENT bit to differentiate dirty pages.
6922                  */
6923                 if (root->log_transid % 2 == 0)
6924                         set_extent_dirty(&root->dirty_log_pages, buf->start,
6925                                         buf->start + buf->len - 1, GFP_NOFS);
6926                 else
6927                         set_extent_new(&root->dirty_log_pages, buf->start,
6928                                         buf->start + buf->len - 1, GFP_NOFS);
6929         } else {
6930                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6931                          buf->start + buf->len - 1, GFP_NOFS);
6932         }
6933         trans->blocks_used++;
6934         /* this returns a buffer locked for blocking */
6935         return buf;
6936 }
6937
6938 static struct btrfs_block_rsv *
6939 use_block_rsv(struct btrfs_trans_handle *trans,
6940               struct btrfs_root *root, u32 blocksize)
6941 {
6942         struct btrfs_block_rsv *block_rsv;
6943         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6944         int ret;
6945         bool global_updated = false;
6946
6947         block_rsv = get_block_rsv(trans, root);
6948
6949         if (unlikely(block_rsv->size == 0))
6950                 goto try_reserve;
6951 again:
6952         ret = block_rsv_use_bytes(block_rsv, blocksize);
6953         if (!ret)
6954                 return block_rsv;
6955
6956         if (block_rsv->failfast)
6957                 return ERR_PTR(ret);
6958
6959         if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6960                 global_updated = true;
6961                 update_global_block_rsv(root->fs_info);
6962                 goto again;
6963         }
6964
6965         if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6966                 static DEFINE_RATELIMIT_STATE(_rs,
6967                                 DEFAULT_RATELIMIT_INTERVAL * 10,
6968                                 /*DEFAULT_RATELIMIT_BURST*/ 1);
6969                 if (__ratelimit(&_rs))
6970                         WARN(1, KERN_DEBUG
6971                                 "btrfs: block rsv returned %d\n", ret);
6972         }
6973 try_reserve:
6974         ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6975                                      BTRFS_RESERVE_NO_FLUSH);
6976         if (!ret)
6977                 return block_rsv;
6978         /*
6979          * If we couldn't reserve metadata bytes try and use some from
6980          * the global reserve if its space type is the same as the global
6981          * reservation.
6982          */
6983         if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6984             block_rsv->space_info == global_rsv->space_info) {
6985                 ret = block_rsv_use_bytes(global_rsv, blocksize);
6986                 if (!ret)
6987                         return global_rsv;
6988         }
6989         return ERR_PTR(ret);
6990 }
6991
6992 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6993                             struct btrfs_block_rsv *block_rsv, u32 blocksize)
6994 {
6995         block_rsv_add_bytes(block_rsv, blocksize, 0);
6996         block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6997 }
6998
6999 /*
7000  * finds a free extent and does all the dirty work required for allocation
7001  * returns the key for the extent through ins, and a tree buffer for
7002  * the first block of the extent through buf.
7003  *
7004  * returns the tree buffer or NULL.
7005  */
7006 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7007                                         struct btrfs_root *root, u32 blocksize,
7008                                         u64 parent, u64 root_objectid,
7009                                         struct btrfs_disk_key *key, int level,
7010                                         u64 hint, u64 empty_size)
7011 {
7012         struct btrfs_key ins;
7013         struct btrfs_block_rsv *block_rsv;
7014         struct extent_buffer *buf;
7015         u64 flags = 0;
7016         int ret;
7017         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7018                                                  SKINNY_METADATA);
7019
7020         block_rsv = use_block_rsv(trans, root, blocksize);
7021         if (IS_ERR(block_rsv))
7022                 return ERR_CAST(block_rsv);
7023
7024         ret = btrfs_reserve_extent(root, blocksize, blocksize,
7025                                    empty_size, hint, &ins, 0);
7026         if (ret) {
7027                 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7028                 return ERR_PTR(ret);
7029         }
7030
7031         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7032                                     blocksize, level);
7033         BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7034
7035         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7036                 if (parent == 0)
7037                         parent = ins.objectid;
7038                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7039         } else
7040                 BUG_ON(parent > 0);
7041
7042         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7043                 struct btrfs_delayed_extent_op *extent_op;
7044                 extent_op = btrfs_alloc_delayed_extent_op();
7045                 BUG_ON(!extent_op); /* -ENOMEM */
7046                 if (key)
7047                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
7048                 else
7049                         memset(&extent_op->key, 0, sizeof(extent_op->key));
7050                 extent_op->flags_to_set = flags;
7051                 if (skinny_metadata)
7052                         extent_op->update_key = 0;
7053                 else
7054                         extent_op->update_key = 1;
7055                 extent_op->update_flags = 1;
7056                 extent_op->is_data = 0;
7057                 extent_op->level = level;
7058
7059                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7060                                         ins.objectid,
7061                                         ins.offset, parent, root_objectid,
7062                                         level, BTRFS_ADD_DELAYED_EXTENT,
7063                                         extent_op, 0);
7064                 BUG_ON(ret); /* -ENOMEM */
7065         }
7066         return buf;
7067 }
7068
7069 struct walk_control {
7070         u64 refs[BTRFS_MAX_LEVEL];
7071         u64 flags[BTRFS_MAX_LEVEL];
7072         struct btrfs_key update_progress;
7073         int stage;
7074         int level;
7075         int shared_level;
7076         int update_ref;
7077         int keep_locks;
7078         int reada_slot;
7079         int reada_count;
7080         int for_reloc;
7081 };
7082
7083 #define DROP_REFERENCE  1
7084 #define UPDATE_BACKREF  2
7085
7086 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7087                                      struct btrfs_root *root,
7088                                      struct walk_control *wc,
7089                                      struct btrfs_path *path)
7090 {
7091         u64 bytenr;
7092         u64 generation;
7093         u64 refs;
7094         u64 flags;
7095         u32 nritems;
7096         u32 blocksize;
7097         struct btrfs_key key;
7098         struct extent_buffer *eb;
7099         int ret;
7100         int slot;
7101         int nread = 0;
7102
7103         if (path->slots[wc->level] < wc->reada_slot) {
7104                 wc->reada_count = wc->reada_count * 2 / 3;
7105                 wc->reada_count = max(wc->reada_count, 2);
7106         } else {
7107                 wc->reada_count = wc->reada_count * 3 / 2;
7108                 wc->reada_count = min_t(int, wc->reada_count,
7109                                         BTRFS_NODEPTRS_PER_BLOCK(root));
7110         }
7111
7112         eb = path->nodes[wc->level];
7113         nritems = btrfs_header_nritems(eb);
7114         blocksize = btrfs_level_size(root, wc->level - 1);
7115
7116         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7117                 if (nread >= wc->reada_count)
7118                         break;
7119
7120                 cond_resched();
7121                 bytenr = btrfs_node_blockptr(eb, slot);
7122                 generation = btrfs_node_ptr_generation(eb, slot);
7123
7124                 if (slot == path->slots[wc->level])
7125                         goto reada;
7126
7127                 if (wc->stage == UPDATE_BACKREF &&
7128                     generation <= root->root_key.offset)
7129                         continue;
7130
7131                 /* We don't lock the tree block, it's OK to be racy here */
7132                 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7133                                                wc->level - 1, 1, &refs,
7134                                                &flags);
7135                 /* We don't care about errors in readahead. */
7136                 if (ret < 0)
7137                         continue;
7138                 BUG_ON(refs == 0);
7139
7140                 if (wc->stage == DROP_REFERENCE) {
7141                         if (refs == 1)
7142                                 goto reada;
7143
7144                         if (wc->level == 1 &&
7145                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7146                                 continue;
7147                         if (!wc->update_ref ||
7148                             generation <= root->root_key.offset)
7149                                 continue;
7150                         btrfs_node_key_to_cpu(eb, &key, slot);
7151                         ret = btrfs_comp_cpu_keys(&key,
7152                                                   &wc->update_progress);
7153                         if (ret < 0)
7154                                 continue;
7155                 } else {
7156                         if (wc->level == 1 &&
7157                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7158                                 continue;
7159                 }
7160 reada:
7161                 ret = readahead_tree_block(root, bytenr, blocksize,
7162                                            generation);
7163                 if (ret)
7164                         break;
7165                 nread++;
7166         }
7167         wc->reada_slot = slot;
7168 }
7169
7170 /*
7171  * helper to process tree block while walking down the tree.
7172  *
7173  * when wc->stage == UPDATE_BACKREF, this function updates
7174  * back refs for pointers in the block.
7175  *
7176  * NOTE: return value 1 means we should stop walking down.
7177  */
7178 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7179                                    struct btrfs_root *root,
7180                                    struct btrfs_path *path,
7181                                    struct walk_control *wc, int lookup_info)
7182 {
7183         int level = wc->level;
7184         struct extent_buffer *eb = path->nodes[level];
7185         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7186         int ret;
7187
7188         if (wc->stage == UPDATE_BACKREF &&
7189             btrfs_header_owner(eb) != root->root_key.objectid)
7190                 return 1;
7191
7192         /*
7193          * when reference count of tree block is 1, it won't increase
7194          * again. once full backref flag is set, we never clear it.
7195          */
7196         if (lookup_info &&
7197             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7198              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7199                 BUG_ON(!path->locks[level]);
7200                 ret = btrfs_lookup_extent_info(trans, root,
7201                                                eb->start, level, 1,
7202                                                &wc->refs[level],
7203                                                &wc->flags[level]);
7204                 BUG_ON(ret == -ENOMEM);
7205                 if (ret)
7206                         return ret;
7207                 BUG_ON(wc->refs[level] == 0);
7208         }
7209
7210         if (wc->stage == DROP_REFERENCE) {
7211                 if (wc->refs[level] > 1)
7212                         return 1;
7213
7214                 if (path->locks[level] && !wc->keep_locks) {
7215                         btrfs_tree_unlock_rw(eb, path->locks[level]);
7216                         path->locks[level] = 0;
7217                 }
7218                 return 0;
7219         }
7220
7221         /* wc->stage == UPDATE_BACKREF */
7222         if (!(wc->flags[level] & flag)) {
7223                 BUG_ON(!path->locks[level]);
7224                 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7225                 BUG_ON(ret); /* -ENOMEM */
7226                 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7227                 BUG_ON(ret); /* -ENOMEM */
7228                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7229                                                   eb->len, flag,
7230                                                   btrfs_header_level(eb), 0);
7231                 BUG_ON(ret); /* -ENOMEM */
7232                 wc->flags[level] |= flag;
7233         }
7234
7235         /*
7236          * the block is shared by multiple trees, so it's not good to
7237          * keep the tree lock
7238          */
7239         if (path->locks[level] && level > 0) {
7240                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7241                 path->locks[level] = 0;
7242         }
7243         return 0;
7244 }
7245
7246 /*
7247  * helper to process tree block pointer.
7248  *
7249  * when wc->stage == DROP_REFERENCE, this function checks
7250  * reference count of the block pointed to. if the block
7251  * is shared and we need update back refs for the subtree
7252  * rooted at the block, this function changes wc->stage to
7253  * UPDATE_BACKREF. if the block is shared and there is no
7254  * need to update back, this function drops the reference
7255  * to the block.
7256  *
7257  * NOTE: return value 1 means we should stop walking down.
7258  */
7259 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7260                                  struct btrfs_root *root,
7261                                  struct btrfs_path *path,
7262                                  struct walk_control *wc, int *lookup_info)
7263 {
7264         u64 bytenr;
7265         u64 generation;
7266         u64 parent;
7267         u32 blocksize;
7268         struct btrfs_key key;
7269         struct extent_buffer *next;
7270         int level = wc->level;
7271         int reada = 0;
7272         int ret = 0;
7273
7274         generation = btrfs_node_ptr_generation(path->nodes[level],
7275                                                path->slots[level]);
7276         /*
7277          * if the lower level block was created before the snapshot
7278          * was created, we know there is no need to update back refs
7279          * for the subtree
7280          */
7281         if (wc->stage == UPDATE_BACKREF &&
7282             generation <= root->root_key.offset) {
7283                 *lookup_info = 1;
7284                 return 1;
7285         }
7286
7287         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7288         blocksize = btrfs_level_size(root, level - 1);
7289
7290         next = btrfs_find_tree_block(root, bytenr, blocksize);
7291         if (!next) {
7292                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7293                 if (!next)
7294                         return -ENOMEM;
7295                 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7296                                                level - 1);
7297                 reada = 1;
7298         }
7299         btrfs_tree_lock(next);
7300         btrfs_set_lock_blocking(next);
7301
7302         ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7303                                        &wc->refs[level - 1],
7304                                        &wc->flags[level - 1]);
7305         if (ret < 0) {
7306                 btrfs_tree_unlock(next);
7307                 return ret;
7308         }
7309
7310         if (unlikely(wc->refs[level - 1] == 0)) {
7311                 btrfs_err(root->fs_info, "Missing references.");
7312                 BUG();
7313         }
7314         *lookup_info = 0;
7315
7316         if (wc->stage == DROP_REFERENCE) {
7317                 if (wc->refs[level - 1] > 1) {
7318                         if (level == 1 &&
7319                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7320                                 goto skip;
7321
7322                         if (!wc->update_ref ||
7323                             generation <= root->root_key.offset)
7324                                 goto skip;
7325
7326                         btrfs_node_key_to_cpu(path->nodes[level], &key,
7327                                               path->slots[level]);
7328                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7329                         if (ret < 0)
7330                                 goto skip;
7331
7332                         wc->stage = UPDATE_BACKREF;
7333                         wc->shared_level = level - 1;
7334                 }
7335         } else {
7336                 if (level == 1 &&
7337                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7338                         goto skip;
7339         }
7340
7341         if (!btrfs_buffer_uptodate(next, generation, 0)) {
7342                 btrfs_tree_unlock(next);
7343                 free_extent_buffer(next);
7344                 next = NULL;
7345                 *lookup_info = 1;
7346         }
7347
7348         if (!next) {
7349                 if (reada && level == 1)
7350                         reada_walk_down(trans, root, wc, path);
7351                 next = read_tree_block(root, bytenr, blocksize, generation);
7352                 if (!next || !extent_buffer_uptodate(next)) {
7353                         free_extent_buffer(next);
7354                         return -EIO;
7355                 }
7356                 btrfs_tree_lock(next);
7357                 btrfs_set_lock_blocking(next);
7358         }
7359
7360         level--;
7361         BUG_ON(level != btrfs_header_level(next));
7362         path->nodes[level] = next;
7363         path->slots[level] = 0;
7364         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7365         wc->level = level;
7366         if (wc->level == 1)
7367                 wc->reada_slot = 0;
7368         return 0;
7369 skip:
7370         wc->refs[level - 1] = 0;
7371         wc->flags[level - 1] = 0;
7372         if (wc->stage == DROP_REFERENCE) {
7373                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7374                         parent = path->nodes[level]->start;
7375                 } else {
7376                         BUG_ON(root->root_key.objectid !=
7377                                btrfs_header_owner(path->nodes[level]));
7378                         parent = 0;
7379                 }
7380
7381                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7382                                 root->root_key.objectid, level - 1, 0, 0);
7383                 BUG_ON(ret); /* -ENOMEM */
7384         }
7385         btrfs_tree_unlock(next);
7386         free_extent_buffer(next);
7387         *lookup_info = 1;
7388         return 1;
7389 }
7390
7391 /*
7392  * helper to process tree block while walking up the tree.
7393  *
7394  * when wc->stage == DROP_REFERENCE, this function drops
7395  * reference count on the block.
7396  *
7397  * when wc->stage == UPDATE_BACKREF, this function changes
7398  * wc->stage back to DROP_REFERENCE if we changed wc->stage
7399  * to UPDATE_BACKREF previously while processing the block.
7400  *
7401  * NOTE: return value 1 means we should stop walking up.
7402  */
7403 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7404                                  struct btrfs_root *root,
7405                                  struct btrfs_path *path,
7406                                  struct walk_control *wc)
7407 {
7408         int ret;
7409         int level = wc->level;
7410         struct extent_buffer *eb = path->nodes[level];
7411         u64 parent = 0;
7412
7413         if (wc->stage == UPDATE_BACKREF) {
7414                 BUG_ON(wc->shared_level < level);
7415                 if (level < wc->shared_level)
7416                         goto out;
7417
7418                 ret = find_next_key(path, level + 1, &wc->update_progress);
7419                 if (ret > 0)
7420                         wc->update_ref = 0;
7421
7422                 wc->stage = DROP_REFERENCE;
7423                 wc->shared_level = -1;
7424                 path->slots[level] = 0;
7425
7426                 /*
7427                  * check reference count again if the block isn't locked.
7428                  * we should start walking down the tree again if reference
7429                  * count is one.
7430                  */
7431                 if (!path->locks[level]) {
7432                         BUG_ON(level == 0);
7433                         btrfs_tree_lock(eb);
7434                         btrfs_set_lock_blocking(eb);
7435                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7436
7437                         ret = btrfs_lookup_extent_info(trans, root,
7438                                                        eb->start, level, 1,
7439                                                        &wc->refs[level],
7440                                                        &wc->flags[level]);
7441                         if (ret < 0) {
7442                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7443                                 path->locks[level] = 0;
7444                                 return ret;
7445                         }
7446                         BUG_ON(wc->refs[level] == 0);
7447                         if (wc->refs[level] == 1) {
7448                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7449                                 path->locks[level] = 0;
7450                                 return 1;
7451                         }
7452                 }
7453         }
7454
7455         /* wc->stage == DROP_REFERENCE */
7456         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7457
7458         if (wc->refs[level] == 1) {
7459                 if (level == 0) {
7460                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7461                                 ret = btrfs_dec_ref(trans, root, eb, 1,
7462                                                     wc->for_reloc);
7463                         else
7464                                 ret = btrfs_dec_ref(trans, root, eb, 0,
7465                                                     wc->for_reloc);
7466                         BUG_ON(ret); /* -ENOMEM */
7467                 }
7468                 /* make block locked assertion in clean_tree_block happy */
7469                 if (!path->locks[level] &&
7470                     btrfs_header_generation(eb) == trans->transid) {
7471                         btrfs_tree_lock(eb);
7472                         btrfs_set_lock_blocking(eb);
7473                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7474                 }
7475                 clean_tree_block(trans, root, eb);
7476         }
7477
7478         if (eb == root->node) {
7479                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7480                         parent = eb->start;
7481                 else
7482                         BUG_ON(root->root_key.objectid !=
7483                                btrfs_header_owner(eb));
7484         } else {
7485                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7486                         parent = path->nodes[level + 1]->start;
7487                 else
7488                         BUG_ON(root->root_key.objectid !=
7489                                btrfs_header_owner(path->nodes[level + 1]));
7490         }
7491
7492         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7493 out:
7494         wc->refs[level] = 0;
7495         wc->flags[level] = 0;
7496         return 0;
7497 }
7498
7499 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7500                                    struct btrfs_root *root,
7501                                    struct btrfs_path *path,
7502                                    struct walk_control *wc)
7503 {
7504         int level = wc->level;
7505         int lookup_info = 1;
7506         int ret;
7507
7508         while (level >= 0) {
7509                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7510                 if (ret > 0)
7511                         break;
7512
7513                 if (level == 0)
7514                         break;
7515
7516                 if (path->slots[level] >=
7517                     btrfs_header_nritems(path->nodes[level]))
7518                         break;
7519
7520                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7521                 if (ret > 0) {
7522                         path->slots[level]++;
7523                         continue;
7524                 } else if (ret < 0)
7525                         return ret;
7526                 level = wc->level;
7527         }
7528         return 0;
7529 }
7530
7531 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7532                                  struct btrfs_root *root,
7533                                  struct btrfs_path *path,
7534                                  struct walk_control *wc, int max_level)
7535 {
7536         int level = wc->level;
7537         int ret;
7538
7539         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7540         while (level < max_level && path->nodes[level]) {
7541                 wc->level = level;
7542                 if (path->slots[level] + 1 <
7543                     btrfs_header_nritems(path->nodes[level])) {
7544                         path->slots[level]++;
7545                         return 0;
7546                 } else {
7547                         ret = walk_up_proc(trans, root, path, wc);
7548                         if (ret > 0)
7549                                 return 0;
7550
7551                         if (path->locks[level]) {
7552                                 btrfs_tree_unlock_rw(path->nodes[level],
7553                                                      path->locks[level]);
7554                                 path->locks[level] = 0;
7555                         }
7556                         free_extent_buffer(path->nodes[level]);
7557                         path->nodes[level] = NULL;
7558                         level++;
7559                 }
7560         }
7561         return 1;
7562 }
7563
7564 /*
7565  * drop a subvolume tree.
7566  *
7567  * this function traverses the tree freeing any blocks that only
7568  * referenced by the tree.
7569  *
7570  * when a shared tree block is found. this function decreases its
7571  * reference count by one. if update_ref is true, this function
7572  * also make sure backrefs for the shared block and all lower level
7573  * blocks are properly updated.
7574  *
7575  * If called with for_reloc == 0, may exit early with -EAGAIN
7576  */
7577 int btrfs_drop_snapshot(struct btrfs_root *root,
7578                          struct btrfs_block_rsv *block_rsv, int update_ref,
7579                          int for_reloc)
7580 {
7581         struct btrfs_path *path;
7582         struct btrfs_trans_handle *trans;
7583         struct btrfs_root *tree_root = root->fs_info->tree_root;
7584         struct btrfs_root_item *root_item = &root->root_item;
7585         struct walk_control *wc;
7586         struct btrfs_key key;
7587         int err = 0;
7588         int ret;
7589         int level;
7590         bool root_dropped = false;
7591
7592         path = btrfs_alloc_path();
7593         if (!path) {
7594                 err = -ENOMEM;
7595                 goto out;
7596         }
7597
7598         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7599         if (!wc) {
7600                 btrfs_free_path(path);
7601                 err = -ENOMEM;
7602                 goto out;
7603         }
7604
7605         trans = btrfs_start_transaction(tree_root, 0);
7606         if (IS_ERR(trans)) {
7607                 err = PTR_ERR(trans);
7608                 goto out_free;
7609         }
7610
7611         if (block_rsv)
7612                 trans->block_rsv = block_rsv;
7613
7614         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7615                 level = btrfs_header_level(root->node);
7616                 path->nodes[level] = btrfs_lock_root_node(root);
7617                 btrfs_set_lock_blocking(path->nodes[level]);
7618                 path->slots[level] = 0;
7619                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7620                 memset(&wc->update_progress, 0,
7621                        sizeof(wc->update_progress));
7622         } else {
7623                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7624                 memcpy(&wc->update_progress, &key,
7625                        sizeof(wc->update_progress));
7626
7627                 level = root_item->drop_level;
7628                 BUG_ON(level == 0);
7629                 path->lowest_level = level;
7630                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7631                 path->lowest_level = 0;
7632                 if (ret < 0) {
7633                         err = ret;
7634                         goto out_end_trans;
7635                 }
7636                 WARN_ON(ret > 0);
7637
7638                 /*
7639                  * unlock our path, this is safe because only this
7640                  * function is allowed to delete this snapshot
7641                  */
7642                 btrfs_unlock_up_safe(path, 0);
7643
7644                 level = btrfs_header_level(root->node);
7645                 while (1) {
7646                         btrfs_tree_lock(path->nodes[level]);
7647                         btrfs_set_lock_blocking(path->nodes[level]);
7648                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7649
7650                         ret = btrfs_lookup_extent_info(trans, root,
7651                                                 path->nodes[level]->start,
7652                                                 level, 1, &wc->refs[level],
7653                                                 &wc->flags[level]);
7654                         if (ret < 0) {
7655                                 err = ret;
7656                                 goto out_end_trans;
7657                         }
7658                         BUG_ON(wc->refs[level] == 0);
7659
7660                         if (level == root_item->drop_level)
7661                                 break;
7662
7663                         btrfs_tree_unlock(path->nodes[level]);
7664                         path->locks[level] = 0;
7665                         WARN_ON(wc->refs[level] != 1);
7666                         level--;
7667                 }
7668         }
7669
7670         wc->level = level;
7671         wc->shared_level = -1;
7672         wc->stage = DROP_REFERENCE;
7673         wc->update_ref = update_ref;
7674         wc->keep_locks = 0;
7675         wc->for_reloc = for_reloc;
7676         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7677
7678         while (1) {
7679
7680                 ret = walk_down_tree(trans, root, path, wc);
7681                 if (ret < 0) {
7682                         err = ret;
7683                         break;
7684                 }
7685
7686                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7687                 if (ret < 0) {
7688                         err = ret;
7689                         break;
7690                 }
7691
7692                 if (ret > 0) {
7693                         BUG_ON(wc->stage != DROP_REFERENCE);
7694                         break;
7695                 }
7696
7697                 if (wc->stage == DROP_REFERENCE) {
7698                         level = wc->level;
7699                         btrfs_node_key(path->nodes[level],
7700                                        &root_item->drop_progress,
7701                                        path->slots[level]);
7702                         root_item->drop_level = level;
7703                 }
7704
7705                 BUG_ON(wc->level == 0);
7706                 if (btrfs_should_end_transaction(trans, tree_root) ||
7707                     (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7708                         ret = btrfs_update_root(trans, tree_root,
7709                                                 &root->root_key,
7710                                                 root_item);
7711                         if (ret) {
7712                                 btrfs_abort_transaction(trans, tree_root, ret);
7713                                 err = ret;
7714                                 goto out_end_trans;
7715                         }
7716
7717                         btrfs_end_transaction_throttle(trans, tree_root);
7718                         if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7719                                 pr_debug("btrfs: drop snapshot early exit\n");
7720                                 err = -EAGAIN;
7721                                 goto out_free;
7722                         }
7723
7724                         trans = btrfs_start_transaction(tree_root, 0);
7725                         if (IS_ERR(trans)) {
7726                                 err = PTR_ERR(trans);
7727                                 goto out_free;
7728                         }
7729                         if (block_rsv)
7730                                 trans->block_rsv = block_rsv;
7731                 }
7732         }
7733         btrfs_release_path(path);
7734         if (err)
7735                 goto out_end_trans;
7736
7737         ret = btrfs_del_root(trans, tree_root, &root->root_key);
7738         if (ret) {
7739                 btrfs_abort_transaction(trans, tree_root, ret);
7740                 goto out_end_trans;
7741         }
7742
7743         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7744                 ret = btrfs_find_root(tree_root, &root->root_key, path,
7745                                       NULL, NULL);
7746                 if (ret < 0) {
7747                         btrfs_abort_transaction(trans, tree_root, ret);
7748                         err = ret;
7749                         goto out_end_trans;
7750                 } else if (ret > 0) {
7751                         /* if we fail to delete the orphan item this time
7752                          * around, it'll get picked up the next time.
7753                          *
7754                          * The most common failure here is just -ENOENT.
7755                          */
7756                         btrfs_del_orphan_item(trans, tree_root,
7757                                               root->root_key.objectid);
7758                 }
7759         }
7760
7761         if (root->in_radix) {
7762                 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7763         } else {
7764                 free_extent_buffer(root->node);
7765                 free_extent_buffer(root->commit_root);
7766                 btrfs_put_fs_root(root);
7767         }
7768         root_dropped = true;
7769 out_end_trans:
7770         btrfs_end_transaction_throttle(trans, tree_root);
7771 out_free:
7772         kfree(wc);
7773         btrfs_free_path(path);
7774 out:
7775         /*
7776          * So if we need to stop dropping the snapshot for whatever reason we
7777          * need to make sure to add it back to the dead root list so that we
7778          * keep trying to do the work later.  This also cleans up roots if we
7779          * don't have it in the radix (like when we recover after a power fail
7780          * or unmount) so we don't leak memory.
7781          */
7782         if (!for_reloc && root_dropped == false)
7783                 btrfs_add_dead_root(root);
7784         if (err)
7785                 btrfs_std_error(root->fs_info, err);
7786         return err;
7787 }
7788
7789 /*
7790  * drop subtree rooted at tree block 'node'.
7791  *
7792  * NOTE: this function will unlock and release tree block 'node'
7793  * only used by relocation code
7794  */
7795 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7796                         struct btrfs_root *root,
7797                         struct extent_buffer *node,
7798                         struct extent_buffer *parent)
7799 {
7800         struct btrfs_path *path;
7801         struct walk_control *wc;
7802         int level;
7803         int parent_level;
7804         int ret = 0;
7805         int wret;
7806
7807         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7808
7809         path = btrfs_alloc_path();
7810         if (!path)
7811                 return -ENOMEM;
7812
7813         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7814         if (!wc) {
7815                 btrfs_free_path(path);
7816                 return -ENOMEM;
7817         }
7818
7819         btrfs_assert_tree_locked(parent);
7820         parent_level = btrfs_header_level(parent);
7821         extent_buffer_get(parent);
7822         path->nodes[parent_level] = parent;
7823         path->slots[parent_level] = btrfs_header_nritems(parent);
7824
7825         btrfs_assert_tree_locked(node);
7826         level = btrfs_header_level(node);
7827         path->nodes[level] = node;
7828         path->slots[level] = 0;
7829         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7830
7831         wc->refs[parent_level] = 1;
7832         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7833         wc->level = level;
7834         wc->shared_level = -1;
7835         wc->stage = DROP_REFERENCE;
7836         wc->update_ref = 0;
7837         wc->keep_locks = 1;
7838         wc->for_reloc = 1;
7839         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7840
7841         while (1) {
7842                 wret = walk_down_tree(trans, root, path, wc);
7843                 if (wret < 0) {
7844                         ret = wret;
7845                         break;
7846                 }
7847
7848                 wret = walk_up_tree(trans, root, path, wc, parent_level);
7849                 if (wret < 0)
7850                         ret = wret;
7851                 if (wret != 0)
7852                         break;
7853         }
7854
7855         kfree(wc);
7856         btrfs_free_path(path);
7857         return ret;
7858 }
7859
7860 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7861 {
7862         u64 num_devices;
7863         u64 stripped;
7864
7865         /*
7866          * if restripe for this chunk_type is on pick target profile and
7867          * return, otherwise do the usual balance
7868          */
7869         stripped = get_restripe_target(root->fs_info, flags);
7870         if (stripped)
7871                 return extended_to_chunk(stripped);
7872
7873         /*
7874          * we add in the count of missing devices because we want
7875          * to make sure that any RAID levels on a degraded FS
7876          * continue to be honored.
7877          */
7878         num_devices = root->fs_info->fs_devices->rw_devices +
7879                 root->fs_info->fs_devices->missing_devices;
7880
7881         stripped = BTRFS_BLOCK_GROUP_RAID0 |
7882                 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7883                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7884
7885         if (num_devices == 1) {
7886                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7887                 stripped = flags & ~stripped;
7888
7889                 /* turn raid0 into single device chunks */
7890                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7891                         return stripped;
7892
7893                 /* turn mirroring into duplication */
7894                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7895                              BTRFS_BLOCK_GROUP_RAID10))
7896                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7897         } else {
7898                 /* they already had raid on here, just return */
7899                 if (flags & stripped)
7900                         return flags;
7901
7902                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7903                 stripped = flags & ~stripped;
7904
7905                 /* switch duplicated blocks with raid1 */
7906                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7907                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7908
7909                 /* this is drive concat, leave it alone */
7910         }
7911
7912         return flags;
7913 }
7914
7915 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7916 {
7917         struct btrfs_space_info *sinfo = cache->space_info;
7918         u64 num_bytes;
7919         u64 min_allocable_bytes;
7920         int ret = -ENOSPC;
7921
7922
7923         /*
7924          * We need some metadata space and system metadata space for
7925          * allocating chunks in some corner cases until we force to set
7926          * it to be readonly.
7927          */
7928         if ((sinfo->flags &
7929              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7930             !force)
7931                 min_allocable_bytes = 1 * 1024 * 1024;
7932         else
7933                 min_allocable_bytes = 0;
7934
7935         spin_lock(&sinfo->lock);
7936         spin_lock(&cache->lock);
7937
7938         if (cache->ro) {
7939                 ret = 0;
7940                 goto out;
7941         }
7942
7943         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7944                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7945
7946         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7947             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7948             min_allocable_bytes <= sinfo->total_bytes) {
7949                 sinfo->bytes_readonly += num_bytes;
7950                 cache->ro = 1;
7951                 ret = 0;
7952         }
7953 out:
7954         spin_unlock(&cache->lock);
7955         spin_unlock(&sinfo->lock);
7956         return ret;
7957 }
7958
7959 int btrfs_set_block_group_ro(struct btrfs_root *root,
7960                              struct btrfs_block_group_cache *cache)
7961
7962 {
7963         struct btrfs_trans_handle *trans;
7964         u64 alloc_flags;
7965         int ret;
7966
7967         BUG_ON(cache->ro);
7968
7969         trans = btrfs_join_transaction(root);
7970         if (IS_ERR(trans))
7971                 return PTR_ERR(trans);
7972
7973         alloc_flags = update_block_group_flags(root, cache->flags);
7974         if (alloc_flags != cache->flags) {
7975                 ret = do_chunk_alloc(trans, root, alloc_flags,
7976                                      CHUNK_ALLOC_FORCE);
7977                 if (ret < 0)
7978                         goto out;
7979         }
7980
7981         ret = set_block_group_ro(cache, 0);
7982         if (!ret)
7983                 goto out;
7984         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7985         ret = do_chunk_alloc(trans, root, alloc_flags,
7986                              CHUNK_ALLOC_FORCE);
7987         if (ret < 0)
7988                 goto out;
7989         ret = set_block_group_ro(cache, 0);
7990 out:
7991         btrfs_end_transaction(trans, root);
7992         return ret;
7993 }
7994
7995 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7996                             struct btrfs_root *root, u64 type)
7997 {
7998         u64 alloc_flags = get_alloc_profile(root, type);
7999         return do_chunk_alloc(trans, root, alloc_flags,
8000                               CHUNK_ALLOC_FORCE);
8001 }
8002
8003 /*
8004  * helper to account the unused space of all the readonly block group in the
8005  * list. takes mirrors into account.
8006  */
8007 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8008 {
8009         struct btrfs_block_group_cache *block_group;
8010         u64 free_bytes = 0;
8011         int factor;
8012
8013         list_for_each_entry(block_group, groups_list, list) {
8014                 spin_lock(&block_group->lock);
8015
8016                 if (!block_group->ro) {
8017                         spin_unlock(&block_group->lock);
8018                         continue;
8019                 }
8020
8021                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8022                                           BTRFS_BLOCK_GROUP_RAID10 |
8023                                           BTRFS_BLOCK_GROUP_DUP))
8024                         factor = 2;
8025                 else
8026                         factor = 1;
8027
8028                 free_bytes += (block_group->key.offset -
8029                                btrfs_block_group_used(&block_group->item)) *
8030                                factor;
8031
8032                 spin_unlock(&block_group->lock);
8033         }
8034
8035         return free_bytes;
8036 }
8037
8038 /*
8039  * helper to account the unused space of all the readonly block group in the
8040  * space_info. takes mirrors into account.
8041  */
8042 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8043 {
8044         int i;
8045         u64 free_bytes = 0;
8046
8047         spin_lock(&sinfo->lock);
8048
8049         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8050                 if (!list_empty(&sinfo->block_groups[i]))
8051                         free_bytes += __btrfs_get_ro_block_group_free_space(
8052                                                 &sinfo->block_groups[i]);
8053
8054         spin_unlock(&sinfo->lock);
8055
8056         return free_bytes;
8057 }
8058
8059 void btrfs_set_block_group_rw(struct btrfs_root *root,
8060                               struct btrfs_block_group_cache *cache)
8061 {
8062         struct btrfs_space_info *sinfo = cache->space_info;
8063         u64 num_bytes;
8064
8065         BUG_ON(!cache->ro);
8066
8067         spin_lock(&sinfo->lock);
8068         spin_lock(&cache->lock);
8069         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8070                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8071         sinfo->bytes_readonly -= num_bytes;
8072         cache->ro = 0;
8073         spin_unlock(&cache->lock);
8074         spin_unlock(&sinfo->lock);
8075 }
8076
8077 /*
8078  * checks to see if its even possible to relocate this block group.
8079  *
8080  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8081  * ok to go ahead and try.
8082  */
8083 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8084 {
8085         struct btrfs_block_group_cache *block_group;
8086         struct btrfs_space_info *space_info;
8087         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8088         struct btrfs_device *device;
8089         struct btrfs_trans_handle *trans;
8090         u64 min_free;
8091         u64 dev_min = 1;
8092         u64 dev_nr = 0;
8093         u64 target;
8094         int index;
8095         int full = 0;
8096         int ret = 0;
8097
8098         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8099
8100         /* odd, couldn't find the block group, leave it alone */
8101         if (!block_group)
8102                 return -1;
8103
8104         min_free = btrfs_block_group_used(&block_group->item);
8105
8106         /* no bytes used, we're good */
8107         if (!min_free)
8108                 goto out;
8109
8110         space_info = block_group->space_info;
8111         spin_lock(&space_info->lock);
8112
8113         full = space_info->full;
8114
8115         /*
8116          * if this is the last block group we have in this space, we can't
8117          * relocate it unless we're able to allocate a new chunk below.
8118          *
8119          * Otherwise, we need to make sure we have room in the space to handle
8120          * all of the extents from this block group.  If we can, we're good
8121          */
8122         if ((space_info->total_bytes != block_group->key.offset) &&
8123             (space_info->bytes_used + space_info->bytes_reserved +
8124              space_info->bytes_pinned + space_info->bytes_readonly +
8125              min_free < space_info->total_bytes)) {
8126                 spin_unlock(&space_info->lock);
8127                 goto out;
8128         }
8129         spin_unlock(&space_info->lock);
8130
8131         /*
8132          * ok we don't have enough space, but maybe we have free space on our
8133          * devices to allocate new chunks for relocation, so loop through our
8134          * alloc devices and guess if we have enough space.  if this block
8135          * group is going to be restriped, run checks against the target
8136          * profile instead of the current one.
8137          */
8138         ret = -1;
8139
8140         /*
8141          * index:
8142          *      0: raid10
8143          *      1: raid1
8144          *      2: dup
8145          *      3: raid0
8146          *      4: single
8147          */
8148         target = get_restripe_target(root->fs_info, block_group->flags);
8149         if (target) {
8150                 index = __get_raid_index(extended_to_chunk(target));
8151         } else {
8152                 /*
8153                  * this is just a balance, so if we were marked as full
8154                  * we know there is no space for a new chunk
8155                  */
8156                 if (full)
8157                         goto out;
8158
8159                 index = get_block_group_index(block_group);
8160         }
8161
8162         if (index == BTRFS_RAID_RAID10) {
8163                 dev_min = 4;
8164                 /* Divide by 2 */
8165                 min_free >>= 1;
8166         } else if (index == BTRFS_RAID_RAID1) {
8167                 dev_min = 2;
8168         } else if (index == BTRFS_RAID_DUP) {
8169                 /* Multiply by 2 */
8170                 min_free <<= 1;
8171         } else if (index == BTRFS_RAID_RAID0) {
8172                 dev_min = fs_devices->rw_devices;
8173                 do_div(min_free, dev_min);
8174         }
8175
8176         /* We need to do this so that we can look at pending chunks */
8177         trans = btrfs_join_transaction(root);
8178         if (IS_ERR(trans)) {
8179                 ret = PTR_ERR(trans);
8180                 goto out;
8181         }
8182
8183         mutex_lock(&root->fs_info->chunk_mutex);
8184         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8185                 u64 dev_offset;
8186
8187                 /*
8188                  * check to make sure we can actually find a chunk with enough
8189                  * space to fit our block group in.
8190                  */
8191                 if (device->total_bytes > device->bytes_used + min_free &&
8192                     !device->is_tgtdev_for_dev_replace) {
8193                         ret = find_free_dev_extent(trans, device, min_free,
8194                                                    &dev_offset, NULL);
8195                         if (!ret)
8196                                 dev_nr++;
8197
8198                         if (dev_nr >= dev_min)
8199                                 break;
8200
8201                         ret = -1;
8202                 }
8203         }
8204         mutex_unlock(&root->fs_info->chunk_mutex);
8205         btrfs_end_transaction(trans, root);
8206 out:
8207         btrfs_put_block_group(block_group);
8208         return ret;
8209 }
8210
8211 static int find_first_block_group(struct btrfs_root *root,
8212                 struct btrfs_path *path, struct btrfs_key *key)
8213 {
8214         int ret = 0;
8215         struct btrfs_key found_key;
8216         struct extent_buffer *leaf;
8217         int slot;
8218
8219         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8220         if (ret < 0)
8221                 goto out;
8222
8223         while (1) {
8224                 slot = path->slots[0];
8225                 leaf = path->nodes[0];
8226                 if (slot >= btrfs_header_nritems(leaf)) {
8227                         ret = btrfs_next_leaf(root, path);
8228                         if (ret == 0)
8229                                 continue;
8230                         if (ret < 0)
8231                                 goto out;
8232                         break;
8233                 }
8234                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8235
8236                 if (found_key.objectid >= key->objectid &&
8237                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8238                         ret = 0;
8239                         goto out;
8240                 }
8241                 path->slots[0]++;
8242         }
8243 out:
8244         return ret;
8245 }
8246
8247 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8248 {
8249         struct btrfs_block_group_cache *block_group;
8250         u64 last = 0;
8251
8252         while (1) {
8253                 struct inode *inode;
8254
8255                 block_group = btrfs_lookup_first_block_group(info, last);
8256                 while (block_group) {
8257                         spin_lock(&block_group->lock);
8258                         if (block_group->iref)
8259                                 break;
8260                         spin_unlock(&block_group->lock);
8261                         block_group = next_block_group(info->tree_root,
8262                                                        block_group);
8263                 }
8264                 if (!block_group) {
8265                         if (last == 0)
8266                                 break;
8267                         last = 0;
8268                         continue;
8269                 }
8270
8271                 inode = block_group->inode;
8272                 block_group->iref = 0;
8273                 block_group->inode = NULL;
8274                 spin_unlock(&block_group->lock);
8275                 iput(inode);
8276                 last = block_group->key.objectid + block_group->key.offset;
8277                 btrfs_put_block_group(block_group);
8278         }
8279 }
8280
8281 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8282 {
8283         struct btrfs_block_group_cache *block_group;
8284         struct btrfs_space_info *space_info;
8285         struct btrfs_caching_control *caching_ctl;
8286         struct rb_node *n;
8287
8288         down_write(&info->extent_commit_sem);
8289         while (!list_empty(&info->caching_block_groups)) {
8290                 caching_ctl = list_entry(info->caching_block_groups.next,
8291                                          struct btrfs_caching_control, list);
8292                 list_del(&caching_ctl->list);
8293                 put_caching_control(caching_ctl);
8294         }
8295         up_write(&info->extent_commit_sem);
8296
8297         spin_lock(&info->block_group_cache_lock);
8298         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8299                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8300                                        cache_node);
8301                 rb_erase(&block_group->cache_node,
8302                          &info->block_group_cache_tree);
8303                 spin_unlock(&info->block_group_cache_lock);
8304
8305                 down_write(&block_group->space_info->groups_sem);
8306                 list_del(&block_group->list);
8307                 up_write(&block_group->space_info->groups_sem);
8308
8309                 if (block_group->cached == BTRFS_CACHE_STARTED)
8310                         wait_block_group_cache_done(block_group);
8311
8312                 /*
8313                  * We haven't cached this block group, which means we could
8314                  * possibly have excluded extents on this block group.
8315                  */
8316                 if (block_group->cached == BTRFS_CACHE_NO ||
8317                     block_group->cached == BTRFS_CACHE_ERROR)
8318                         free_excluded_extents(info->extent_root, block_group);
8319
8320                 btrfs_remove_free_space_cache(block_group);
8321                 btrfs_put_block_group(block_group);
8322
8323                 spin_lock(&info->block_group_cache_lock);
8324         }
8325         spin_unlock(&info->block_group_cache_lock);
8326
8327         /* now that all the block groups are freed, go through and
8328          * free all the space_info structs.  This is only called during
8329          * the final stages of unmount, and so we know nobody is
8330          * using them.  We call synchronize_rcu() once before we start,
8331          * just to be on the safe side.
8332          */
8333         synchronize_rcu();
8334
8335         release_global_block_rsv(info);
8336
8337         while (!list_empty(&info->space_info)) {
8338                 space_info = list_entry(info->space_info.next,
8339                                         struct btrfs_space_info,
8340                                         list);
8341                 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8342                         if (WARN_ON(space_info->bytes_pinned > 0 ||
8343                             space_info->bytes_reserved > 0 ||
8344                             space_info->bytes_may_use > 0)) {
8345                                 dump_space_info(space_info, 0, 0);
8346                         }
8347                 }
8348                 percpu_counter_destroy(&space_info->total_bytes_pinned);
8349                 list_del(&space_info->list);
8350                 kfree(space_info);
8351         }
8352         return 0;
8353 }
8354
8355 static void __link_block_group(struct btrfs_space_info *space_info,
8356                                struct btrfs_block_group_cache *cache)
8357 {
8358         int index = get_block_group_index(cache);
8359
8360         down_write(&space_info->groups_sem);
8361         list_add_tail(&cache->list, &space_info->block_groups[index]);
8362         up_write(&space_info->groups_sem);
8363 }
8364
8365 int btrfs_read_block_groups(struct btrfs_root *root)
8366 {
8367         struct btrfs_path *path;
8368         int ret;
8369         struct btrfs_block_group_cache *cache;
8370         struct btrfs_fs_info *info = root->fs_info;
8371         struct btrfs_space_info *space_info;
8372         struct btrfs_key key;
8373         struct btrfs_key found_key;
8374         struct extent_buffer *leaf;
8375         int need_clear = 0;
8376         u64 cache_gen;
8377
8378         root = info->extent_root;
8379         key.objectid = 0;
8380         key.offset = 0;
8381         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8382         path = btrfs_alloc_path();
8383         if (!path)
8384                 return -ENOMEM;
8385         path->reada = 1;
8386
8387         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8388         if (btrfs_test_opt(root, SPACE_CACHE) &&
8389             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8390                 need_clear = 1;
8391         if (btrfs_test_opt(root, CLEAR_CACHE))
8392                 need_clear = 1;
8393
8394         while (1) {
8395                 ret = find_first_block_group(root, path, &key);
8396                 if (ret > 0)
8397                         break;
8398                 if (ret != 0)
8399                         goto error;
8400                 leaf = path->nodes[0];
8401                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8402                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8403                 if (!cache) {
8404                         ret = -ENOMEM;
8405                         goto error;
8406                 }
8407                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8408                                                 GFP_NOFS);
8409                 if (!cache->free_space_ctl) {
8410                         kfree(cache);
8411                         ret = -ENOMEM;
8412                         goto error;
8413                 }
8414
8415                 atomic_set(&cache->count, 1);
8416                 spin_lock_init(&cache->lock);
8417                 cache->fs_info = info;
8418                 INIT_LIST_HEAD(&cache->list);
8419                 INIT_LIST_HEAD(&cache->cluster_list);
8420
8421                 if (need_clear) {
8422                         /*
8423                          * When we mount with old space cache, we need to
8424                          * set BTRFS_DC_CLEAR and set dirty flag.
8425                          *
8426                          * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8427                          *    truncate the old free space cache inode and
8428                          *    setup a new one.
8429                          * b) Setting 'dirty flag' makes sure that we flush
8430                          *    the new space cache info onto disk.
8431                          */
8432                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8433                         if (btrfs_test_opt(root, SPACE_CACHE))
8434                                 cache->dirty = 1;
8435                 }
8436
8437                 read_extent_buffer(leaf, &cache->item,
8438                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8439                                    sizeof(cache->item));
8440                 memcpy(&cache->key, &found_key, sizeof(found_key));
8441
8442                 key.objectid = found_key.objectid + found_key.offset;
8443                 btrfs_release_path(path);
8444                 cache->flags = btrfs_block_group_flags(&cache->item);
8445                 cache->sectorsize = root->sectorsize;
8446                 cache->full_stripe_len = btrfs_full_stripe_len(root,
8447                                                &root->fs_info->mapping_tree,
8448                                                found_key.objectid);
8449                 btrfs_init_free_space_ctl(cache);
8450
8451                 /*
8452                  * We need to exclude the super stripes now so that the space
8453                  * info has super bytes accounted for, otherwise we'll think
8454                  * we have more space than we actually do.
8455                  */
8456                 ret = exclude_super_stripes(root, cache);
8457                 if (ret) {
8458                         /*
8459                          * We may have excluded something, so call this just in
8460                          * case.
8461                          */
8462                         free_excluded_extents(root, cache);
8463                         kfree(cache->free_space_ctl);
8464                         kfree(cache);
8465                         goto error;
8466                 }
8467
8468                 /*
8469                  * check for two cases, either we are full, and therefore
8470                  * don't need to bother with the caching work since we won't
8471                  * find any space, or we are empty, and we can just add all
8472                  * the space in and be done with it.  This saves us _alot_ of
8473                  * time, particularly in the full case.
8474                  */
8475                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8476                         cache->last_byte_to_unpin = (u64)-1;
8477                         cache->cached = BTRFS_CACHE_FINISHED;
8478                         free_excluded_extents(root, cache);
8479                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8480                         cache->last_byte_to_unpin = (u64)-1;
8481                         cache->cached = BTRFS_CACHE_FINISHED;
8482                         add_new_free_space(cache, root->fs_info,
8483                                            found_key.objectid,
8484                                            found_key.objectid +
8485                                            found_key.offset);
8486                         free_excluded_extents(root, cache);
8487                 }
8488
8489                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8490                 if (ret) {
8491                         btrfs_remove_free_space_cache(cache);
8492                         btrfs_put_block_group(cache);
8493                         goto error;
8494                 }
8495
8496                 ret = update_space_info(info, cache->flags, found_key.offset,
8497                                         btrfs_block_group_used(&cache->item),
8498                                         &space_info);
8499                 if (ret) {
8500                         btrfs_remove_free_space_cache(cache);
8501                         spin_lock(&info->block_group_cache_lock);
8502                         rb_erase(&cache->cache_node,
8503                                  &info->block_group_cache_tree);
8504                         spin_unlock(&info->block_group_cache_lock);
8505                         btrfs_put_block_group(cache);
8506                         goto error;
8507                 }
8508
8509                 cache->space_info = space_info;
8510                 spin_lock(&cache->space_info->lock);
8511                 cache->space_info->bytes_readonly += cache->bytes_super;
8512                 spin_unlock(&cache->space_info->lock);
8513
8514                 __link_block_group(space_info, cache);
8515
8516                 set_avail_alloc_bits(root->fs_info, cache->flags);
8517                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8518                         set_block_group_ro(cache, 1);
8519         }
8520
8521         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8522                 if (!(get_alloc_profile(root, space_info->flags) &
8523                       (BTRFS_BLOCK_GROUP_RAID10 |
8524                        BTRFS_BLOCK_GROUP_RAID1 |
8525                        BTRFS_BLOCK_GROUP_RAID5 |
8526                        BTRFS_BLOCK_GROUP_RAID6 |
8527                        BTRFS_BLOCK_GROUP_DUP)))
8528                         continue;
8529                 /*
8530                  * avoid allocating from un-mirrored block group if there are
8531                  * mirrored block groups.
8532                  */
8533                 list_for_each_entry(cache,
8534                                 &space_info->block_groups[BTRFS_RAID_RAID0],
8535                                 list)
8536                         set_block_group_ro(cache, 1);
8537                 list_for_each_entry(cache,
8538                                 &space_info->block_groups[BTRFS_RAID_SINGLE],
8539                                 list)
8540                         set_block_group_ro(cache, 1);
8541         }
8542
8543         init_global_block_rsv(info);
8544         ret = 0;
8545 error:
8546         btrfs_free_path(path);
8547         return ret;
8548 }
8549
8550 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8551                                        struct btrfs_root *root)
8552 {
8553         struct btrfs_block_group_cache *block_group, *tmp;
8554         struct btrfs_root *extent_root = root->fs_info->extent_root;
8555         struct btrfs_block_group_item item;
8556         struct btrfs_key key;
8557         int ret = 0;
8558
8559         list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8560                                  new_bg_list) {
8561                 list_del_init(&block_group->new_bg_list);
8562
8563                 if (ret)
8564                         continue;
8565
8566                 spin_lock(&block_group->lock);
8567                 memcpy(&item, &block_group->item, sizeof(item));
8568                 memcpy(&key, &block_group->key, sizeof(key));
8569                 spin_unlock(&block_group->lock);
8570
8571                 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8572                                         sizeof(item));
8573                 if (ret)
8574                         btrfs_abort_transaction(trans, extent_root, ret);
8575                 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8576                                                key.objectid, key.offset);
8577                 if (ret)
8578                         btrfs_abort_transaction(trans, extent_root, ret);
8579         }
8580 }
8581
8582 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8583                            struct btrfs_root *root, u64 bytes_used,
8584                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8585                            u64 size)
8586 {
8587         int ret;
8588         struct btrfs_root *extent_root;
8589         struct btrfs_block_group_cache *cache;
8590
8591         extent_root = root->fs_info->extent_root;
8592
8593         root->fs_info->last_trans_log_full_commit = trans->transid;
8594
8595         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8596         if (!cache)
8597                 return -ENOMEM;
8598         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8599                                         GFP_NOFS);
8600         if (!cache->free_space_ctl) {
8601                 kfree(cache);
8602                 return -ENOMEM;
8603         }
8604
8605         cache->key.objectid = chunk_offset;
8606         cache->key.offset = size;
8607         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8608         cache->sectorsize = root->sectorsize;
8609         cache->fs_info = root->fs_info;
8610         cache->full_stripe_len = btrfs_full_stripe_len(root,
8611                                                &root->fs_info->mapping_tree,
8612                                                chunk_offset);
8613
8614         atomic_set(&cache->count, 1);
8615         spin_lock_init(&cache->lock);
8616         INIT_LIST_HEAD(&cache->list);
8617         INIT_LIST_HEAD(&cache->cluster_list);
8618         INIT_LIST_HEAD(&cache->new_bg_list);
8619
8620         btrfs_init_free_space_ctl(cache);
8621
8622         btrfs_set_block_group_used(&cache->item, bytes_used);
8623         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8624         cache->flags = type;
8625         btrfs_set_block_group_flags(&cache->item, type);
8626
8627         cache->last_byte_to_unpin = (u64)-1;
8628         cache->cached = BTRFS_CACHE_FINISHED;
8629         ret = exclude_super_stripes(root, cache);
8630         if (ret) {
8631                 /*
8632                  * We may have excluded something, so call this just in
8633                  * case.
8634                  */
8635                 free_excluded_extents(root, cache);
8636                 kfree(cache->free_space_ctl);
8637                 kfree(cache);
8638                 return ret;
8639         }
8640
8641         add_new_free_space(cache, root->fs_info, chunk_offset,
8642                            chunk_offset + size);
8643
8644         free_excluded_extents(root, cache);
8645
8646         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8647         if (ret) {
8648                 btrfs_remove_free_space_cache(cache);
8649                 btrfs_put_block_group(cache);
8650                 return ret;
8651         }
8652
8653         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8654                                 &cache->space_info);
8655         if (ret) {
8656                 btrfs_remove_free_space_cache(cache);
8657                 spin_lock(&root->fs_info->block_group_cache_lock);
8658                 rb_erase(&cache->cache_node,
8659                          &root->fs_info->block_group_cache_tree);
8660                 spin_unlock(&root->fs_info->block_group_cache_lock);
8661                 btrfs_put_block_group(cache);
8662                 return ret;
8663         }
8664         update_global_block_rsv(root->fs_info);
8665
8666         spin_lock(&cache->space_info->lock);
8667         cache->space_info->bytes_readonly += cache->bytes_super;
8668         spin_unlock(&cache->space_info->lock);
8669
8670         __link_block_group(cache->space_info, cache);
8671
8672         list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8673
8674         set_avail_alloc_bits(extent_root->fs_info, type);
8675
8676         return 0;
8677 }
8678
8679 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8680 {
8681         u64 extra_flags = chunk_to_extended(flags) &
8682                                 BTRFS_EXTENDED_PROFILE_MASK;
8683
8684         write_seqlock(&fs_info->profiles_lock);
8685         if (flags & BTRFS_BLOCK_GROUP_DATA)
8686                 fs_info->avail_data_alloc_bits &= ~extra_flags;
8687         if (flags & BTRFS_BLOCK_GROUP_METADATA)
8688                 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8689         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8690                 fs_info->avail_system_alloc_bits &= ~extra_flags;
8691         write_sequnlock(&fs_info->profiles_lock);
8692 }
8693
8694 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8695                              struct btrfs_root *root, u64 group_start)
8696 {
8697         struct btrfs_path *path;
8698         struct btrfs_block_group_cache *block_group;
8699         struct btrfs_free_cluster *cluster;
8700         struct btrfs_root *tree_root = root->fs_info->tree_root;
8701         struct btrfs_key key;
8702         struct inode *inode;
8703         int ret;
8704         int index;
8705         int factor;
8706
8707         root = root->fs_info->extent_root;
8708
8709         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8710         BUG_ON(!block_group);
8711         BUG_ON(!block_group->ro);
8712
8713         /*
8714          * Free the reserved super bytes from this block group before
8715          * remove it.
8716          */
8717         free_excluded_extents(root, block_group);
8718
8719         memcpy(&key, &block_group->key, sizeof(key));
8720         index = get_block_group_index(block_group);
8721         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8722                                   BTRFS_BLOCK_GROUP_RAID1 |
8723                                   BTRFS_BLOCK_GROUP_RAID10))
8724                 factor = 2;
8725         else
8726                 factor = 1;
8727
8728         /* make sure this block group isn't part of an allocation cluster */
8729         cluster = &root->fs_info->data_alloc_cluster;
8730         spin_lock(&cluster->refill_lock);
8731         btrfs_return_cluster_to_free_space(block_group, cluster);
8732         spin_unlock(&cluster->refill_lock);
8733
8734         /*
8735          * make sure this block group isn't part of a metadata
8736          * allocation cluster
8737          */
8738         cluster = &root->fs_info->meta_alloc_cluster;
8739         spin_lock(&cluster->refill_lock);
8740         btrfs_return_cluster_to_free_space(block_group, cluster);
8741         spin_unlock(&cluster->refill_lock);
8742
8743         path = btrfs_alloc_path();
8744         if (!path) {
8745                 ret = -ENOMEM;
8746                 goto out;
8747         }
8748
8749         inode = lookup_free_space_inode(tree_root, block_group, path);
8750         if (!IS_ERR(inode)) {
8751                 ret = btrfs_orphan_add(trans, inode);
8752                 if (ret) {
8753                         btrfs_add_delayed_iput(inode);
8754                         goto out;
8755                 }
8756                 clear_nlink(inode);
8757                 /* One for the block groups ref */
8758                 spin_lock(&block_group->lock);
8759                 if (block_group->iref) {
8760                         block_group->iref = 0;
8761                         block_group->inode = NULL;
8762                         spin_unlock(&block_group->lock);
8763                         iput(inode);
8764                 } else {
8765                         spin_unlock(&block_group->lock);
8766                 }
8767                 /* One for our lookup ref */
8768                 btrfs_add_delayed_iput(inode);
8769         }
8770
8771         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8772         key.offset = block_group->key.objectid;
8773         key.type = 0;
8774
8775         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8776         if (ret < 0)
8777                 goto out;
8778         if (ret > 0)
8779                 btrfs_release_path(path);
8780         if (ret == 0) {
8781                 ret = btrfs_del_item(trans, tree_root, path);
8782                 if (ret)
8783                         goto out;
8784                 btrfs_release_path(path);
8785         }
8786
8787         spin_lock(&root->fs_info->block_group_cache_lock);
8788         rb_erase(&block_group->cache_node,
8789                  &root->fs_info->block_group_cache_tree);
8790
8791         if (root->fs_info->first_logical_byte == block_group->key.objectid)
8792                 root->fs_info->first_logical_byte = (u64)-1;
8793         spin_unlock(&root->fs_info->block_group_cache_lock);
8794
8795         down_write(&block_group->space_info->groups_sem);
8796         /*
8797          * we must use list_del_init so people can check to see if they
8798          * are still on the list after taking the semaphore
8799          */
8800         list_del_init(&block_group->list);
8801         if (list_empty(&block_group->space_info->block_groups[index]))
8802                 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8803         up_write(&block_group->space_info->groups_sem);
8804
8805         if (block_group->cached == BTRFS_CACHE_STARTED)
8806                 wait_block_group_cache_done(block_group);
8807
8808         btrfs_remove_free_space_cache(block_group);
8809
8810         spin_lock(&block_group->space_info->lock);
8811         block_group->space_info->total_bytes -= block_group->key.offset;
8812         block_group->space_info->bytes_readonly -= block_group->key.offset;
8813         block_group->space_info->disk_total -= block_group->key.offset * factor;
8814         spin_unlock(&block_group->space_info->lock);
8815
8816         memcpy(&key, &block_group->key, sizeof(key));
8817
8818         btrfs_clear_space_info_full(root->fs_info);
8819
8820         btrfs_put_block_group(block_group);
8821         btrfs_put_block_group(block_group);
8822
8823         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8824         if (ret > 0)
8825                 ret = -EIO;
8826         if (ret < 0)
8827                 goto out;
8828
8829         ret = btrfs_del_item(trans, root, path);
8830 out:
8831         btrfs_free_path(path);
8832         return ret;
8833 }
8834
8835 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8836 {
8837         struct btrfs_space_info *space_info;
8838         struct btrfs_super_block *disk_super;
8839         u64 features;
8840         u64 flags;
8841         int mixed = 0;
8842         int ret;
8843
8844         disk_super = fs_info->super_copy;
8845         if (!btrfs_super_root(disk_super))
8846                 return 1;
8847
8848         features = btrfs_super_incompat_flags(disk_super);
8849         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8850                 mixed = 1;
8851
8852         flags = BTRFS_BLOCK_GROUP_SYSTEM;
8853         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8854         if (ret)
8855                 goto out;
8856
8857         if (mixed) {
8858                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8859                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8860         } else {
8861                 flags = BTRFS_BLOCK_GROUP_METADATA;
8862                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8863                 if (ret)
8864                         goto out;
8865
8866                 flags = BTRFS_BLOCK_GROUP_DATA;
8867                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8868         }
8869 out:
8870         return ret;
8871 }
8872
8873 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8874 {
8875         return unpin_extent_range(root, start, end);
8876 }
8877
8878 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8879                                u64 num_bytes, u64 *actual_bytes)
8880 {
8881         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8882 }
8883
8884 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8885 {
8886         struct btrfs_fs_info *fs_info = root->fs_info;
8887         struct btrfs_block_group_cache *cache = NULL;
8888         u64 group_trimmed;
8889         u64 start;
8890         u64 end;
8891         u64 trimmed = 0;
8892         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8893         int ret = 0;
8894
8895         /*
8896          * try to trim all FS space, our block group may start from non-zero.
8897          */
8898         if (range->len == total_bytes)
8899                 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8900         else
8901                 cache = btrfs_lookup_block_group(fs_info, range->start);
8902
8903         while (cache) {
8904                 if (cache->key.objectid >= (range->start + range->len)) {
8905                         btrfs_put_block_group(cache);
8906                         break;
8907                 }
8908
8909                 start = max(range->start, cache->key.objectid);
8910                 end = min(range->start + range->len,
8911                                 cache->key.objectid + cache->key.offset);
8912
8913                 if (end - start >= range->minlen) {
8914                         if (!block_group_cache_done(cache)) {
8915                                 ret = cache_block_group(cache, 0);
8916                                 if (ret) {
8917                                         btrfs_put_block_group(cache);
8918                                         break;
8919                                 }
8920                                 ret = wait_block_group_cache_done(cache);
8921                                 if (ret) {
8922                                         btrfs_put_block_group(cache);
8923                                         break;
8924                                 }
8925                         }
8926                         ret = btrfs_trim_block_group(cache,
8927                                                      &group_trimmed,
8928                                                      start,
8929                                                      end,
8930                                                      range->minlen);
8931
8932                         trimmed += group_trimmed;
8933                         if (ret) {
8934                                 btrfs_put_block_group(cache);
8935                                 break;
8936                         }
8937                 }
8938
8939                 cache = next_block_group(fs_info->tree_root, cache);
8940         }
8941
8942         range->len = trimmed;
8943         return ret;
8944 }