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 "compat.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
 109 static noinline int
 110 block_group_cache_done(struct btrfs_block_group_cache *cache)
 111 {
 112         smp_mb();
 113         return cache->cached == BTRFS_CACHE_FINISHED;
 114 }
 115
 116 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 117 {
 118         return (cache->flags & bits) == bits;
 119 }
 120
 121 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 122 {
 123         atomic_inc(&cache->count);
 124 }
 125
 126 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 127 {
 128         if (atomic_dec_and_test(&cache->count)) {
 129                 WARN_ON(cache->pinned > 0);
 130                 WARN_ON(cache->reserved > 0);
 131                 kfree(cache->free_space_ctl);
 132                 kfree(cache);
 133         }
 134 }
 135
 136 /*
 137  * this adds the block group to the fs_info rb tree for the block group
 138  * cache
 139  */
 140 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 141                                 struct btrfs_block_group_cache *block_group)
 142 {
 143         struct rb_node **p;
 144         struct rb_node *parent = NULL;
 145         struct btrfs_block_group_cache *cache;
 146
 147         spin_lock(&info->block_group_cache_lock);
 148         p = &info->block_group_cache_tree.rb_node;
 149
 150         while (*p) {
 151                 parent = *p;
 152                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 153                                  cache_node);
 154                 if (block_group->key.objectid < cache->key.objectid) {
 155                         p = &(*p)->rb_left;
 156                 } else if (block_group->key.objectid > cache->key.objectid) {
 157                         p = &(*p)->rb_right;
 158                 } else {
 159                         spin_unlock(&info->block_group_cache_lock);
 160                         return -EEXIST;
 161                 }
 162         }
 163
 164         rb_link_node(&block_group->cache_node, parent, p);
 165         rb_insert_color(&block_group->cache_node,
 166                         &info->block_group_cache_tree);
 167
 168         if (info->first_logical_byte > block_group->key.objectid)
 169                 info->first_logical_byte = block_group->key.objectid;
 170
 171         spin_unlock(&info->block_group_cache_lock);
 172
 173         return 0;
 174 }
 175
 176 /*
 177  * This will return the block group at or after bytenr if contains is 0, else
 178  * it will return the block group that contains the bytenr
 179  */
 180 static struct btrfs_block_group_cache *
 181 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 182                               int contains)
 183 {
 184         struct btrfs_block_group_cache *cache, *ret = NULL;
 185         struct rb_node *n;
 186         u64 end, start;
 187
 188         spin_lock(&info->block_group_cache_lock);
 189         n = info->block_group_cache_tree.rb_node;
 190
 191         while (n) {
 192                 cache = rb_entry(n, struct btrfs_block_group_cache,
 193                                  cache_node);
 194                 end = cache->key.objectid + cache->key.offset - 1;
 195                 start = cache->key.objectid;
 196
 197                 if (bytenr < start) {
 198                         if (!contains && (!ret || start < ret->key.objectid))
 199                                 ret = cache;
 200                         n = n->rb_left;
 201                 } else if (bytenr > start) {
 202                         if (contains && bytenr <= end) {
 203                                 ret = cache;
 204                                 break;
 205                         }
 206                         n = n->rb_right;
 207                 } else {
 208                         ret = cache;
 209                         break;
 210                 }
 211         }
 212         if (ret) {
 213                 btrfs_get_block_group(ret);
 214                 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
 215                         info->first_logical_byte = ret->key.objectid;
 216         }
 217         spin_unlock(&info->block_group_cache_lock);
 218
 219         return ret;
 220 }
 221
 222 static int add_excluded_extent(struct btrfs_root *root,
 223                                u64 start, u64 num_bytes)
 224 {
 225         u64 end = start + num_bytes - 1;
 226         set_extent_bits(&root->fs_info->freed_extents[0],
 227                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 228         set_extent_bits(&root->fs_info->freed_extents[1],
 229                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 230         return 0;
 231 }
 232
 233 static void free_excluded_extents(struct btrfs_root *root,
 234                                   struct btrfs_block_group_cache *cache)
 235 {
 236         u64 start, end;
 237
 238         start = cache->key.objectid;
 239         end = start + cache->key.offset - 1;
 240
 241         clear_extent_bits(&root->fs_info->freed_extents[0],
 242                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 243         clear_extent_bits(&root->fs_info->freed_extents[1],
 244                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 245 }
 246
 247 static int exclude_super_stripes(struct btrfs_root *root,
 248                                  struct btrfs_block_group_cache *cache)
 249 {
 250         u64 bytenr;
 251         u64 *logical;
 252         int stripe_len;
 253         int i, nr, ret;
 254
 255         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 256                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 257                 cache->bytes_super += stripe_len;
 258                 ret = add_excluded_extent(root, cache->key.objectid,
 259                                           stripe_len);
 260                 if (ret)
 261                         return ret;
 262         }
 263
 264         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 265                 bytenr = btrfs_sb_offset(i);
 266                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 267                                        cache->key.objectid, bytenr,
 268                                        0, &logical, &nr, &stripe_len);
 269                 if (ret)
 270                         return ret;
 271
 272                 while (nr--) {
 273                         cache->bytes_super += stripe_len;
 274                         ret = add_excluded_extent(root, logical[nr],
 275                                                   stripe_len);
 276                         if (ret) {
 277                                 kfree(logical);
 278                                 return ret;
 279                         }
 280                 }
 281
 282                 kfree(logical);
 283         }
 284         return 0;
 285 }
 286
 287 static struct btrfs_caching_control *
 288 get_caching_control(struct btrfs_block_group_cache *cache)
 289 {
 290         struct btrfs_caching_control *ctl;
 291
 292         spin_lock(&cache->lock);
 293         if (cache->cached != BTRFS_CACHE_STARTED) {
 294                 spin_unlock(&cache->lock);
 295                 return NULL;
 296         }
 297
 298         /* We're loading it the fast way, so we don't have a caching_ctl. */
 299         if (!cache->caching_ctl) {
 300                 spin_unlock(&cache->lock);
 301                 return NULL;
 302         }
 303
 304         ctl = cache->caching_ctl;
 305         atomic_inc(&ctl->count);
 306         spin_unlock(&cache->lock);
 307         return ctl;
 308 }
 309
 310 static void put_caching_control(struct btrfs_caching_control *ctl)
 311 {
 312         if (atomic_dec_and_test(&ctl->count))
 313                 kfree(ctl);
 314 }
 315
 316 /*
 317  * this is only called by cache_block_group, since we could have freed extents
 318  * we need to check the pinned_extents for any extents that can't be used yet
 319  * since their free space will be released as soon as the transaction commits.
 320  */
 321 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 322                               struct btrfs_fs_info *info, u64 start, u64 end)
 323 {
 324         u64 extent_start, extent_end, size, total_added = 0;
 325         int ret;
 326
 327         while (start < end) {
 328                 ret = find_first_extent_bit(info->pinned_extents, start,
 329                                             &extent_start, &extent_end,
 330                                             EXTENT_DIRTY | EXTENT_UPTODATE,
 331                                             NULL);
 332                 if (ret)
 333                         break;
 334
 335                 if (extent_start <= start) {
 336                         start = extent_end + 1;
 337                 } else if (extent_start > start && extent_start < end) {
 338                         size = extent_start - start;
 339                         total_added += size;
 340                         ret = btrfs_add_free_space(block_group, start,
 341                                                    size);
 342                         BUG_ON(ret); /* -ENOMEM or logic error */
 343                         start = extent_end + 1;
 344                 } else {
 345                         break;
 346                 }
 347         }
 348
 349         if (start < end) {
 350                 size = end - start;
 351                 total_added += size;
 352                 ret = btrfs_add_free_space(block_group, start, size);
 353                 BUG_ON(ret); /* -ENOMEM or logic error */
 354         }
 355
 356         return total_added;
 357 }
 358
 359 static noinline void caching_thread(struct btrfs_work *work)
 360 {
 361         struct btrfs_block_group_cache *block_group;
 362         struct btrfs_fs_info *fs_info;
 363         struct btrfs_caching_control *caching_ctl;
 364         struct btrfs_root *extent_root;
 365         struct btrfs_path *path;
 366         struct extent_buffer *leaf;
 367         struct btrfs_key key;
 368         u64 total_found = 0;
 369         u64 last = 0;
 370         u32 nritems;
 371         int ret = 0;
 372
 373         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 374         block_group = caching_ctl->block_group;
 375         fs_info = block_group->fs_info;
 376         extent_root = fs_info->extent_root;
 377
 378         path = btrfs_alloc_path();
 379         if (!path)
 380                 goto out;
 381
 382         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 383
 384         /*
 385          * We don't want to deadlock with somebody trying to allocate a new
 386          * extent for the extent root while also trying to search the extent
 387          * root to add free space.  So we skip locking and search the commit
 388          * root, since its read-only
 389          */
 390         path->skip_locking = 1;
 391         path->search_commit_root = 1;
 392         path->reada = 1;
 393
 394         key.objectid = last;
 395         key.offset = 0;
 396         key.type = BTRFS_EXTENT_ITEM_KEY;
 397 again:
 398         mutex_lock(&caching_ctl->mutex);
 399         /* need to make sure the commit_root doesn't disappear */
 400         down_read(&fs_info->extent_commit_sem);
 401
 402         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 403         if (ret < 0)
 404                 goto err;
 405
 406         leaf = path->nodes[0];
 407         nritems = btrfs_header_nritems(leaf);
 408
 409         while (1) {
 410                 if (btrfs_fs_closing(fs_info) > 1) {
 411                         last = (u64)-1;
 412                         break;
 413                 }
 414
 415                 if (path->slots[0] < nritems) {
 416                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 417                 } else {
 418                         ret = find_next_key(path, 0, &key);
 419                         if (ret)
 420                                 break;
 421
 422                         if (need_resched() ||
 423                             btrfs_next_leaf(extent_root, path)) {
 424                                 caching_ctl->progress = last;
 425                                 btrfs_release_path(path);
 426                                 up_read(&fs_info->extent_commit_sem);
 427                                 mutex_unlock(&caching_ctl->mutex);
 428                                 cond_resched();
 429                                 goto again;
 430                         }
 431                         leaf = path->nodes[0];
 432                         nritems = btrfs_header_nritems(leaf);
 433                         continue;
 434                 }
 435
 436                 if (key.objectid < block_group->key.objectid) {
 437                         path->slots[0]++;
 438                         continue;
 439                 }
 440
 441                 if (key.objectid >= block_group->key.objectid +
 442                     block_group->key.offset)
 443                         break;
 444
 445                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
 446                     key.type == BTRFS_METADATA_ITEM_KEY) {
 447                         total_found += add_new_free_space(block_group,
 448                                                           fs_info, last,
 449                                                           key.objectid);
 450                         if (key.type == BTRFS_METADATA_ITEM_KEY)
 451                                 last = key.objectid +
 452                                         fs_info->tree_root->leafsize;
 453                         else
 454                                 last = key.objectid + key.offset;
 455
 456                         if (total_found > (1024 * 1024 * 2)) {
 457                                 total_found = 0;
 458                                 wake_up(&caching_ctl->wait);
 459                         }
 460                 }
 461                 path->slots[0]++;
 462         }
 463         ret = 0;
 464
 465         total_found += add_new_free_space(block_group, fs_info, last,
 466                                           block_group->key.objectid +
 467                                           block_group->key.offset);
 468         caching_ctl->progress = (u64)-1;
 469
 470         spin_lock(&block_group->lock);
 471         block_group->caching_ctl = NULL;
 472         block_group->cached = BTRFS_CACHE_FINISHED;
 473         spin_unlock(&block_group->lock);
 474
 475 err:
 476         btrfs_free_path(path);
 477         up_read(&fs_info->extent_commit_sem);
 478
 479         free_excluded_extents(extent_root, block_group);
 480
 481         mutex_unlock(&caching_ctl->mutex);
 482 out:
 483         wake_up(&caching_ctl->wait);
 484
 485         put_caching_control(caching_ctl);
 486         btrfs_put_block_group(block_group);
 487 }
 488
 489 static int cache_block_group(struct btrfs_block_group_cache *cache,
 490                              int load_cache_only)
 491 {
 492         DEFINE_WAIT(wait);
 493         struct btrfs_fs_info *fs_info = cache->fs_info;
 494         struct btrfs_caching_control *caching_ctl;
 495         int ret = 0;
 496
 497         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 498         if (!caching_ctl)
 499                 return -ENOMEM;
 500
 501         INIT_LIST_HEAD(&caching_ctl->list);
 502         mutex_init(&caching_ctl->mutex);
 503         init_waitqueue_head(&caching_ctl->wait);
 504         caching_ctl->block_group = cache;
 505         caching_ctl->progress = cache->key.objectid;
 506         atomic_set(&caching_ctl->count, 1);
 507         caching_ctl->work.func = caching_thread;
 508
 509         spin_lock(&cache->lock);
 510         /*
 511          * This should be a rare occasion, but this could happen I think in the
 512          * case where one thread starts to load the space cache info, and then
 513          * some other thread starts a transaction commit which tries to do an
 514          * allocation while the other thread is still loading the space cache
 515          * info.  The previous loop should have kept us from choosing this block
 516          * group, but if we've moved to the state where we will wait on caching
 517          * block groups we need to first check if we're doing a fast load here,
 518          * so we can wait for it to finish, otherwise we could end up allocating
 519          * from a block group who's cache gets evicted for one reason or
 520          * another.
 521          */
 522         while (cache->cached == BTRFS_CACHE_FAST) {
 523                 struct btrfs_caching_control *ctl;
 524
 525                 ctl = cache->caching_ctl;
 526                 atomic_inc(&ctl->count);
 527                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
 528                 spin_unlock(&cache->lock);
 529
 530                 schedule();
 531
 532                 finish_wait(&ctl->wait, &wait);
 533                 put_caching_control(ctl);
 534                 spin_lock(&cache->lock);
 535         }
 536
 537         if (cache->cached != BTRFS_CACHE_NO) {
 538                 spin_unlock(&cache->lock);
 539                 kfree(caching_ctl);
 540                 return 0;
 541         }
 542         WARN_ON(cache->caching_ctl);
 543         cache->caching_ctl = caching_ctl;
 544         cache->cached = BTRFS_CACHE_FAST;
 545         spin_unlock(&cache->lock);
 546
 547         if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
 548                 ret = load_free_space_cache(fs_info, cache);
 549
 550                 spin_lock(&cache->lock);
 551                 if (ret == 1) {
 552                         cache->caching_ctl = NULL;
 553                         cache->cached = BTRFS_CACHE_FINISHED;
 554                         cache->last_byte_to_unpin = (u64)-1;
 555                 } else {
 556                         if (load_cache_only) {
 557                                 cache->caching_ctl = NULL;
 558                                 cache->cached = BTRFS_CACHE_NO;
 559                         } else {
 560                                 cache->cached = BTRFS_CACHE_STARTED;
 561                         }
 562                 }
 563                 spin_unlock(&cache->lock);
 564                 wake_up(&caching_ctl->wait);
 565                 if (ret == 1) {
 566                         put_caching_control(caching_ctl);
 567                         free_excluded_extents(fs_info->extent_root, cache);
 568                         return 0;
 569                 }
 570         } else {
 571                 /*
 572                  * We are not going to do the fast caching, set cached to the
 573                  * appropriate value and wakeup any waiters.
 574                  */
 575                 spin_lock(&cache->lock);
 576                 if (load_cache_only) {
 577                         cache->caching_ctl = NULL;
 578                         cache->cached = BTRFS_CACHE_NO;
 579                 } else {
 580                         cache->cached = BTRFS_CACHE_STARTED;
 581                 }
 582                 spin_unlock(&cache->lock);
 583                 wake_up(&caching_ctl->wait);
 584         }
 585
 586         if (load_cache_only) {
 587                 put_caching_control(caching_ctl);
 588                 return 0;
 589         }
 590
 591         down_write(&fs_info->extent_commit_sem);
 592         atomic_inc(&caching_ctl->count);
 593         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 594         up_write(&fs_info->extent_commit_sem);
 595
 596         btrfs_get_block_group(cache);
 597
 598         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 599
 600         return ret;
 601 }
 602
 603 /*
 604  * return the block group that starts at or after bytenr
 605  */
 606 static struct btrfs_block_group_cache *
 607 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 608 {
 609         struct btrfs_block_group_cache *cache;
 610
 611         cache = block_group_cache_tree_search(info, bytenr, 0);
 612
 613         return cache;
 614 }
 615
 616 /*
 617  * return the block group that contains the given bytenr
 618  */
 619 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 620                                                  struct btrfs_fs_info *info,
 621                                                  u64 bytenr)
 622 {
 623         struct btrfs_block_group_cache *cache;
 624
 625         cache = block_group_cache_tree_search(info, bytenr, 1);
 626
 627         return cache;
 628 }
 629
 630 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 631                                                   u64 flags)
 632 {
 633         struct list_head *head = &info->space_info;
 634         struct btrfs_space_info *found;
 635
 636         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 637
 638         rcu_read_lock();
 639         list_for_each_entry_rcu(found, head, list) {
 640                 if (found->flags & flags) {
 641                         rcu_read_unlock();
 642                         return found;
 643                 }
 644         }
 645         rcu_read_unlock();
 646         return NULL;
 647 }
 648
 649 /*
 650  * after adding space to the filesystem, we need to clear the full flags
 651  * on all the space infos.
 652  */
 653 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 654 {
 655         struct list_head *head = &info->space_info;
 656         struct btrfs_space_info *found;
 657
 658         rcu_read_lock();
 659         list_for_each_entry_rcu(found, head, list)
 660                 found->full = 0;
 661         rcu_read_unlock();
 662 }
 663
 664 u64 btrfs_find_block_group(struct btrfs_root *root,
 665                            u64 search_start, u64 search_hint, int owner)
 666 {
 667         struct btrfs_block_group_cache *cache;
 668         u64 used;
 669         u64 last = max(search_hint, search_start);
 670         u64 group_start = 0;
 671         int full_search = 0;
 672         int factor = 9;
 673         int wrapped = 0;
 674 again:
 675         while (1) {
 676                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 677                 if (!cache)
 678                         break;
 679
 680                 spin_lock(&cache->lock);
 681                 last = cache->key.objectid + cache->key.offset;
 682                 used = btrfs_block_group_used(&cache->item);
 683
 684                 if ((full_search || !cache->ro) &&
 685                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 686                         if (used + cache->pinned + cache->reserved <
 687                             div_factor(cache->key.offset, factor)) {
 688                                 group_start = cache->key.objectid;
 689                                 spin_unlock(&cache->lock);
 690                                 btrfs_put_block_group(cache);
 691                                 goto found;
 692                         }
 693                 }
 694                 spin_unlock(&cache->lock);
 695                 btrfs_put_block_group(cache);
 696                 cond_resched();
 697         }
 698         if (!wrapped) {
 699                 last = search_start;
 700                 wrapped = 1;
 701                 goto again;
 702         }
 703         if (!full_search && factor < 10) {
 704                 last = search_start;
 705                 full_search = 1;
 706                 factor = 10;
 707                 goto again;
 708         }
 709 found:
 710         return group_start;
 711 }
 712
 713 /* simple helper to search for an existing extent at a given offset */
 714 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 715 {
 716         int ret;
 717         struct btrfs_key key;
 718         struct btrfs_path *path;
 719
 720         path = btrfs_alloc_path();
 721         if (!path)
 722                 return -ENOMEM;
 723
 724         key.objectid = start;
 725         key.offset = len;
 726         key.type = BTRFS_EXTENT_ITEM_KEY;
 727         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 728                                 0, 0);
 729         if (ret > 0) {
 730                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 731                 if (key.objectid == start &&
 732                     key.type == BTRFS_METADATA_ITEM_KEY)
 733                         ret = 0;
 734         }
 735         btrfs_free_path(path);
 736         return ret;
 737 }
 738
 739 /*
 740  * helper function to lookup reference count and flags of a tree block.
 741  *
 742  * the head node for delayed ref is used to store the sum of all the
 743  * reference count modifications queued up in the rbtree. the head
 744  * node may also store the extent flags to set. This way you can check
 745  * to see what the reference count and extent flags would be if all of
 746  * the delayed refs are not processed.
 747  */
 748 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 749                              struct btrfs_root *root, u64 bytenr,
 750                              u64 offset, int metadata, u64 *refs, u64 *flags)
 751 {
 752         struct btrfs_delayed_ref_head *head;
 753         struct btrfs_delayed_ref_root *delayed_refs;
 754         struct btrfs_path *path;
 755         struct btrfs_extent_item *ei;
 756         struct extent_buffer *leaf;
 757         struct btrfs_key key;
 758         u32 item_size;
 759         u64 num_refs;
 760         u64 extent_flags;
 761         int ret;
 762
 763         /*
 764          * If we don't have skinny metadata, don't bother doing anything
 765          * different
 766          */
 767         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
 768                 offset = root->leafsize;
 769                 metadata = 0;
 770         }
 771
 772         path = btrfs_alloc_path();
 773         if (!path)
 774                 return -ENOMEM;
 775
 776         if (metadata) {
 777                 key.objectid = bytenr;
 778                 key.type = BTRFS_METADATA_ITEM_KEY;
 779                 key.offset = offset;
 780         } else {
 781                 key.objectid = bytenr;
 782                 key.type = BTRFS_EXTENT_ITEM_KEY;
 783                 key.offset = offset;
 784         }
 785
 786         if (!trans) {
 787                 path->skip_locking = 1;
 788                 path->search_commit_root = 1;
 789         }
 790 again:
 791         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 792                                 &key, path, 0, 0);
 793         if (ret < 0)
 794                 goto out_free;
 795
 796         if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
 797                 key.type = BTRFS_EXTENT_ITEM_KEY;
 798                 key.offset = root->leafsize;
 799                 btrfs_release_path(path);
 800                 goto again;
 801         }
 802
 803         if (ret == 0) {
 804                 leaf = path->nodes[0];
 805                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 806                 if (item_size >= sizeof(*ei)) {
 807                         ei = btrfs_item_ptr(leaf, path->slots[0],
 808                                             struct btrfs_extent_item);
 809                         num_refs = btrfs_extent_refs(leaf, ei);
 810                         extent_flags = btrfs_extent_flags(leaf, ei);
 811                 } else {
 812 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 813                         struct btrfs_extent_item_v0 *ei0;
 814                         BUG_ON(item_size != sizeof(*ei0));
 815                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 816                                              struct btrfs_extent_item_v0);
 817                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 818                         /* FIXME: this isn't correct for data */
 819                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 820 #else
 821                         BUG();
 822 #endif
 823                 }
 824                 BUG_ON(num_refs == 0);
 825         } else {
 826                 num_refs = 0;
 827                 extent_flags = 0;
 828                 ret = 0;
 829         }
 830
 831         if (!trans)
 832                 goto out;
 833
 834         delayed_refs = &trans->transaction->delayed_refs;
 835         spin_lock(&delayed_refs->lock);
 836         head = btrfs_find_delayed_ref_head(trans, bytenr);
 837         if (head) {
 838                 if (!mutex_trylock(&head->mutex)) {
 839                         atomic_inc(&head->node.refs);
 840                         spin_unlock(&delayed_refs->lock);
 841
 842                         btrfs_release_path(path);
 843
 844                         /*
 845                          * Mutex was contended, block until it's released and try
 846                          * again
 847                          */
 848                         mutex_lock(&head->mutex);
 849                         mutex_unlock(&head->mutex);
 850                         btrfs_put_delayed_ref(&head->node);
 851                         goto again;
 852                 }
 853                 if (head->extent_op && head->extent_op->update_flags)
 854                         extent_flags |= head->extent_op->flags_to_set;
 855                 else
 856                         BUG_ON(num_refs == 0);
 857
 858                 num_refs += head->node.ref_mod;
 859                 mutex_unlock(&head->mutex);
 860         }
 861         spin_unlock(&delayed_refs->lock);
 862 out:
 863         WARN_ON(num_refs == 0);
 864         if (refs)
 865                 *refs = num_refs;
 866         if (flags)
 867                 *flags = extent_flags;
 868 out_free:
 869         btrfs_free_path(path);
 870         return ret;
 871 }
 872
 873 /*
 874  * Back reference rules.  Back refs have three main goals:
 875  *
 876  * 1) differentiate between all holders of references to an extent so that
 877  *    when a reference is dropped we can make sure it was a valid reference
 878  *    before freeing the extent.
 879  *
 880  * 2) Provide enough information to quickly find the holders of an extent
 881  *    if we notice a given block is corrupted or bad.
 882  *
 883  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 884  *    maintenance.  This is actually the same as #2, but with a slightly
 885  *    different use case.
 886  *
 887  * There are two kinds of back refs. The implicit back refs is optimized
 888  * for pointers in non-shared tree blocks. For a given pointer in a block,
 889  * back refs of this kind provide information about the block's owner tree
 890  * and the pointer's key. These information allow us to find the block by
 891  * b-tree searching. The full back refs is for pointers in tree blocks not
 892  * referenced by their owner trees. The location of tree block is recorded
 893  * in the back refs. Actually the full back refs is generic, and can be
 894  * used in all cases the implicit back refs is used. The major shortcoming
 895  * of the full back refs is its overhead. Every time a tree block gets
 896  * COWed, we have to update back refs entry for all pointers in it.
 897  *
 898  * For a newly allocated tree block, we use implicit back refs for
 899  * pointers in it. This means most tree related operations only involve
 900  * implicit back refs. For a tree block created in old transaction, the
 901  * only way to drop a reference to it is COW it. So we can detect the
 902  * event that tree block loses its owner tree's reference and do the
 903  * back refs conversion.
 904  *
 905  * When a tree block is COW'd through a tree, there are four cases:
 906  *
 907  * The reference count of the block is one and the tree is the block's
 908  * owner tree. Nothing to do in this case.
 909  *
 910  * The reference count of the block is one and the tree is not the
 911  * block's owner tree. In this case, full back refs is used for pointers
 912  * in the block. Remove these full back refs, add implicit back refs for
 913  * every pointers in the new block.
 914  *
 915  * The reference count of the block is greater than one and the tree is
 916  * the block's owner tree. In this case, implicit back refs is used for
 917  * pointers in the block. Add full back refs for every pointers in the
 918  * block, increase lower level extents' reference counts. The original
 919  * implicit back refs are entailed to the new block.
 920  *
 921  * The reference count of the block is greater than one and the tree is
 922  * not the block's owner tree. Add implicit back refs for every pointer in
 923  * the new block, increase lower level extents' reference count.
 924  *
 925  * Back Reference Key composing:
 926  *
 927  * The key objectid corresponds to the first byte in the extent,
 928  * The key type is used to differentiate between types of back refs.
 929  * There are different meanings of the key offset for different types
 930  * of back refs.
 931  *
 932  * File extents can be referenced by:
 933  *
 934  * - multiple snapshots, subvolumes, or different generations in one subvol
 935  * - different files inside a single subvolume
 936  * - different offsets inside a file (bookend extents in file.c)
 937  *
 938  * The extent ref structure for the implicit back refs has fields for:
 939  *
 940  * - Objectid of the subvolume root
 941  * - objectid of the file holding the reference
 942  * - original offset in the file
 943  * - how many bookend extents
 944  *
 945  * The key offset for the implicit back refs is hash of the first
 946  * three fields.
 947  *
 948  * The extent ref structure for the full back refs has field for:
 949  *
 950  * - number of pointers in the tree leaf
 951  *
 952  * The key offset for the implicit back refs is the first byte of
 953  * the tree leaf
 954  *
 955  * When a file extent is allocated, The implicit back refs is used.
 956  * the fields are filled in:
 957  *
 958  *     (root_key.objectid, inode objectid, offset in file, 1)
 959  *
 960  * When a file extent is removed file truncation, we find the
 961  * corresponding implicit back refs and check the following fields:
 962  *
 963  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 964  *
 965  * Btree extents can be referenced by:
 966  *
 967  * - Different subvolumes
 968  *
 969  * Both the implicit back refs and the full back refs for tree blocks
 970  * only consist of key. The key offset for the implicit back refs is
 971  * objectid of block's owner tree. The key offset for the full back refs
 972  * is the first byte of parent block.
 973  *
 974  * When implicit back refs is used, information about the lowest key and
 975  * level of the tree block are required. These information are stored in
 976  * tree block info structure.
 977  */
 978
 979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 980 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 981                                   struct btrfs_root *root,
 982                                   struct btrfs_path *path,
 983                                   u64 owner, u32 extra_size)
 984 {
 985         struct btrfs_extent_item *item;
 986         struct btrfs_extent_item_v0 *ei0;
 987         struct btrfs_extent_ref_v0 *ref0;
 988         struct btrfs_tree_block_info *bi;
 989         struct extent_buffer *leaf;
 990         struct btrfs_key key;
 991         struct btrfs_key found_key;
 992         u32 new_size = sizeof(*item);
 993         u64 refs;
 994         int ret;
 995
 996         leaf = path->nodes[0];
 997         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 998
 999         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1000         ei0 = btrfs_item_ptr(leaf, path->slots[0],
1001                              struct btrfs_extent_item_v0);
1002         refs = btrfs_extent_refs_v0(leaf, ei0);
1003
1004         if (owner == (u64)-1) {
1005                 while (1) {
1006                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1007                                 ret = btrfs_next_leaf(root, path);
1008                                 if (ret < 0)
1009                                         return ret;
1010                                 BUG_ON(ret > 0); /* Corruption */
1011                                 leaf = path->nodes[0];
1012                         }
1013                         btrfs_item_key_to_cpu(leaf, &found_key,
1014                                               path->slots[0]);
1015                         BUG_ON(key.objectid != found_key.objectid);
1016                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1017                                 path->slots[0]++;
1018                                 continue;
1019                         }
1020                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1021                                               struct btrfs_extent_ref_v0);
1022                         owner = btrfs_ref_objectid_v0(leaf, ref0);
1023                         break;
1024                 }
1025         }
1026         btrfs_release_path(path);
1027
1028         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1029                 new_size += sizeof(*bi);
1030
1031         new_size -= sizeof(*ei0);
1032         ret = btrfs_search_slot(trans, root, &key, path,
1033                                 new_size + extra_size, 1);
1034         if (ret < 0)
1035                 return ret;
1036         BUG_ON(ret); /* Corruption */
1037
1038         btrfs_extend_item(trans, root, path, new_size);
1039
1040         leaf = path->nodes[0];
1041         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1042         btrfs_set_extent_refs(leaf, item, refs);
1043         /* FIXME: get real generation */
1044         btrfs_set_extent_generation(leaf, item, 0);
1045         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1046                 btrfs_set_extent_flags(leaf, item,
1047                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1048                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1049                 bi = (struct btrfs_tree_block_info *)(item + 1);
1050                 /* FIXME: get first key of the block */
1051                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1052                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1053         } else {
1054                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1055         }
1056         btrfs_mark_buffer_dirty(leaf);
1057         return 0;
1058 }
1059 #endif
1060
1061 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1062 {
1063         u32 high_crc = ~(u32)0;
1064         u32 low_crc = ~(u32)0;
1065         __le64 lenum;
1066
1067         lenum = cpu_to_le64(root_objectid);
1068         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1069         lenum = cpu_to_le64(owner);
1070         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1071         lenum = cpu_to_le64(offset);
1072         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1073
1074         return ((u64)high_crc << 31) ^ (u64)low_crc;
1075 }
1076
1077 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1078                                      struct btrfs_extent_data_ref *ref)
1079 {
1080         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1081                                     btrfs_extent_data_ref_objectid(leaf, ref),
1082                                     btrfs_extent_data_ref_offset(leaf, ref));
1083 }
1084
1085 static int match_extent_data_ref(struct extent_buffer *leaf,
1086                                  struct btrfs_extent_data_ref *ref,
1087                                  u64 root_objectid, u64 owner, u64 offset)
1088 {
1089         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1090             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1091             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1092                 return 0;
1093         return 1;
1094 }
1095
1096 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1097                                            struct btrfs_root *root,
1098                                            struct btrfs_path *path,
1099                                            u64 bytenr, u64 parent,
1100                                            u64 root_objectid,
1101                                            u64 owner, u64 offset)
1102 {
1103         struct btrfs_key key;
1104         struct btrfs_extent_data_ref *ref;
1105         struct extent_buffer *leaf;
1106         u32 nritems;
1107         int ret;
1108         int recow;
1109         int err = -ENOENT;
1110
1111         key.objectid = bytenr;
1112         if (parent) {
1113                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1114                 key.offset = parent;
1115         } else {
1116                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1117                 key.offset = hash_extent_data_ref(root_objectid,
1118                                                   owner, offset);
1119         }
1120 again:
1121         recow = 0;
1122         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1123         if (ret < 0) {
1124                 err = ret;
1125                 goto fail;
1126         }
1127
1128         if (parent) {
1129                 if (!ret)
1130                         return 0;
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1133                 btrfs_release_path(path);
1134                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1135                 if (ret < 0) {
1136                         err = ret;
1137                         goto fail;
1138                 }
1139                 if (!ret)
1140                         return 0;
1141 #endif
1142                 goto fail;
1143         }
1144
1145         leaf = path->nodes[0];
1146         nritems = btrfs_header_nritems(leaf);
1147         while (1) {
1148                 if (path->slots[0] >= nritems) {
1149                         ret = btrfs_next_leaf(root, path);
1150                         if (ret < 0)
1151                                 err = ret;
1152                         if (ret)
1153                                 goto fail;
1154
1155                         leaf = path->nodes[0];
1156                         nritems = btrfs_header_nritems(leaf);
1157                         recow = 1;
1158                 }
1159
1160                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1161                 if (key.objectid != bytenr ||
1162                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1163                         goto fail;
1164
1165                 ref = btrfs_item_ptr(leaf, path->slots[0],
1166                                      struct btrfs_extent_data_ref);
1167
1168                 if (match_extent_data_ref(leaf, ref, root_objectid,
1169                                           owner, offset)) {
1170                         if (recow) {
1171                                 btrfs_release_path(path);
1172                                 goto again;
1173                         }
1174                         err = 0;
1175                         break;
1176                 }
1177                 path->slots[0]++;
1178         }
1179 fail:
1180         return err;
1181 }
1182
1183 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1184                                            struct btrfs_root *root,
1185                                            struct btrfs_path *path,
1186                                            u64 bytenr, u64 parent,
1187                                            u64 root_objectid, u64 owner,
1188                                            u64 offset, int refs_to_add)
1189 {
1190         struct btrfs_key key;
1191         struct extent_buffer *leaf;
1192         u32 size;
1193         u32 num_refs;
1194         int ret;
1195
1196         key.objectid = bytenr;
1197         if (parent) {
1198                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1199                 key.offset = parent;
1200                 size = sizeof(struct btrfs_shared_data_ref);
1201         } else {
1202                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1203                 key.offset = hash_extent_data_ref(root_objectid,
1204                                                   owner, offset);
1205                 size = sizeof(struct btrfs_extent_data_ref);
1206         }
1207
1208         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1209         if (ret && ret != -EEXIST)
1210                 goto fail;
1211
1212         leaf = path->nodes[0];
1213         if (parent) {
1214                 struct btrfs_shared_data_ref *ref;
1215                 ref = btrfs_item_ptr(leaf, path->slots[0],
1216                                      struct btrfs_shared_data_ref);
1217                 if (ret == 0) {
1218                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1219                 } else {
1220                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1221                         num_refs += refs_to_add;
1222                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1223                 }
1224         } else {
1225                 struct btrfs_extent_data_ref *ref;
1226                 while (ret == -EEXIST) {
1227                         ref = btrfs_item_ptr(leaf, path->slots[0],
1228                                              struct btrfs_extent_data_ref);
1229                         if (match_extent_data_ref(leaf, ref, root_objectid,
1230                                                   owner, offset))
1231                                 break;
1232                         btrfs_release_path(path);
1233                         key.offset++;
1234                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1235                                                       size);
1236                         if (ret && ret != -EEXIST)
1237                                 goto fail;
1238
1239                         leaf = path->nodes[0];
1240                 }
1241                 ref = btrfs_item_ptr(leaf, path->slots[0],
1242                                      struct btrfs_extent_data_ref);
1243                 if (ret == 0) {
1244                         btrfs_set_extent_data_ref_root(leaf, ref,
1245                                                        root_objectid);
1246                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1247                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1248                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1249                 } else {
1250                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1251                         num_refs += refs_to_add;
1252                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1253                 }
1254         }
1255         btrfs_mark_buffer_dirty(leaf);
1256         ret = 0;
1257 fail:
1258         btrfs_release_path(path);
1259         return ret;
1260 }
1261
1262 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1263                                            struct btrfs_root *root,
1264                                            struct btrfs_path *path,
1265                                            int refs_to_drop)
1266 {
1267         struct btrfs_key key;
1268         struct btrfs_extent_data_ref *ref1 = NULL;
1269         struct btrfs_shared_data_ref *ref2 = NULL;
1270         struct extent_buffer *leaf;
1271         u32 num_refs = 0;
1272         int ret = 0;
1273
1274         leaf = path->nodes[0];
1275         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1276
1277         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1278                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1279                                       struct btrfs_extent_data_ref);
1280                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1281         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1282                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1283                                       struct btrfs_shared_data_ref);
1284                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1286         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1287                 struct btrfs_extent_ref_v0 *ref0;
1288                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1289                                       struct btrfs_extent_ref_v0);
1290                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1291 #endif
1292         } else {
1293                 BUG();
1294         }
1295
1296         BUG_ON(num_refs < refs_to_drop);
1297         num_refs -= refs_to_drop;
1298
1299         if (num_refs == 0) {
1300                 ret = btrfs_del_item(trans, root, path);
1301         } else {
1302                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1303                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1304                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1305                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1307                 else {
1308                         struct btrfs_extent_ref_v0 *ref0;
1309                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310                                         struct btrfs_extent_ref_v0);
1311                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1312                 }
1313 #endif
1314                 btrfs_mark_buffer_dirty(leaf);
1315         }
1316         return ret;
1317 }
1318
1319 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1320                                           struct btrfs_path *path,
1321                                           struct btrfs_extent_inline_ref *iref)
1322 {
1323         struct btrfs_key key;
1324         struct extent_buffer *leaf;
1325         struct btrfs_extent_data_ref *ref1;
1326         struct btrfs_shared_data_ref *ref2;
1327         u32 num_refs = 0;
1328
1329         leaf = path->nodes[0];
1330         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1331         if (iref) {
1332                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1333                     BTRFS_EXTENT_DATA_REF_KEY) {
1334                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1335                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1336                 } else {
1337                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1338                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1339                 }
1340         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342                                       struct btrfs_extent_data_ref);
1343                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346                                       struct btrfs_shared_data_ref);
1347                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350                 struct btrfs_extent_ref_v0 *ref0;
1351                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352                                       struct btrfs_extent_ref_v0);
1353                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1354 #endif
1355         } else {
1356                 WARN_ON(1);
1357         }
1358         return num_refs;
1359 }
1360
1361 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1362                                           struct btrfs_root *root,
1363                                           struct btrfs_path *path,
1364                                           u64 bytenr, u64 parent,
1365                                           u64 root_objectid)
1366 {
1367         struct btrfs_key key;
1368         int ret;
1369
1370         key.objectid = bytenr;
1371         if (parent) {
1372                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1373                 key.offset = parent;
1374         } else {
1375                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1376                 key.offset = root_objectid;
1377         }
1378
1379         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1380         if (ret > 0)
1381                 ret = -ENOENT;
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383         if (ret == -ENOENT && parent) {
1384                 btrfs_release_path(path);
1385                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1386                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1387                 if (ret > 0)
1388                         ret = -ENOENT;
1389         }
1390 #endif
1391         return ret;
1392 }
1393
1394 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1395                                           struct btrfs_root *root,
1396                                           struct btrfs_path *path,
1397                                           u64 bytenr, u64 parent,
1398                                           u64 root_objectid)
1399 {
1400         struct btrfs_key key;
1401         int ret;
1402
1403         key.objectid = bytenr;
1404         if (parent) {
1405                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1406                 key.offset = parent;
1407         } else {
1408                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1409                 key.offset = root_objectid;
1410         }
1411
1412         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1413         btrfs_release_path(path);
1414         return ret;
1415 }
1416
1417 static inline int extent_ref_type(u64 parent, u64 owner)
1418 {
1419         int type;
1420         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1421                 if (parent > 0)
1422                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1423                 else
1424                         type = BTRFS_TREE_BLOCK_REF_KEY;
1425         } else {
1426                 if (parent > 0)
1427                         type = BTRFS_SHARED_DATA_REF_KEY;
1428                 else
1429                         type = BTRFS_EXTENT_DATA_REF_KEY;
1430         }
1431         return type;
1432 }
1433
1434 static int find_next_key(struct btrfs_path *path, int level,
1435                          struct btrfs_key *key)
1436
1437 {
1438         for (; level < BTRFS_MAX_LEVEL; level++) {
1439                 if (!path->nodes[level])
1440                         break;
1441                 if (path->slots[level] + 1 >=
1442                     btrfs_header_nritems(path->nodes[level]))
1443                         continue;
1444                 if (level == 0)
1445                         btrfs_item_key_to_cpu(path->nodes[level], key,
1446                                               path->slots[level] + 1);
1447                 else
1448                         btrfs_node_key_to_cpu(path->nodes[level], key,
1449                                               path->slots[level] + 1);
1450                 return 0;
1451         }
1452         return 1;
1453 }
1454
1455 /*
1456  * look for inline back ref. if back ref is found, *ref_ret is set
1457  * to the address of inline back ref, and 0 is returned.
1458  *
1459  * if back ref isn't found, *ref_ret is set to the address where it
1460  * should be inserted, and -ENOENT is returned.
1461  *
1462  * if insert is true and there are too many inline back refs, the path
1463  * points to the extent item, and -EAGAIN is returned.
1464  *
1465  * NOTE: inline back refs are ordered in the same way that back ref
1466  *       items in the tree are ordered.
1467  */
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1470                                  struct btrfs_root *root,
1471                                  struct btrfs_path *path,
1472                                  struct btrfs_extent_inline_ref **ref_ret,
1473                                  u64 bytenr, u64 num_bytes,
1474                                  u64 parent, u64 root_objectid,
1475                                  u64 owner, u64 offset, int insert)
1476 {
1477         struct btrfs_key key;
1478         struct extent_buffer *leaf;
1479         struct btrfs_extent_item *ei;
1480         struct btrfs_extent_inline_ref *iref;
1481         u64 flags;
1482         u64 item_size;
1483         unsigned long ptr;
1484         unsigned long end;
1485         int extra_size;
1486         int type;
1487         int want;
1488         int ret;
1489         int err = 0;
1490         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1491                                                  SKINNY_METADATA);
1492
1493         key.objectid = bytenr;
1494         key.type = BTRFS_EXTENT_ITEM_KEY;
1495         key.offset = num_bytes;
1496
1497         want = extent_ref_type(parent, owner);
1498         if (insert) {
1499                 extra_size = btrfs_extent_inline_ref_size(want);
1500                 path->keep_locks = 1;
1501         } else
1502                 extra_size = -1;
1503
1504         /*
1505          * Owner is our parent level, so we can just add one to get the level
1506          * for the block we are interested in.
1507          */
1508         if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1509                 key.type = BTRFS_METADATA_ITEM_KEY;
1510                 key.offset = owner;
1511         }
1512
1513 again:
1514         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1515         if (ret < 0) {
1516                 err = ret;
1517                 goto out;
1518         }
1519
1520         /*
1521          * We may be a newly converted file system which still has the old fat
1522          * extent entries for metadata, so try and see if we have one of those.
1523          */
1524         if (ret > 0 && skinny_metadata) {
1525                 skinny_metadata = false;
1526                 if (path->slots[0]) {
1527                         path->slots[0]--;
1528                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1529                                               path->slots[0]);
1530                         if (key.objectid == bytenr &&
1531                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1532                             key.offset == num_bytes)
1533                                 ret = 0;
1534                 }
1535                 if (ret) {
1536                         key.type = BTRFS_EXTENT_ITEM_KEY;
1537                         key.offset = num_bytes;
1538                         btrfs_release_path(path);
1539                         goto again;
1540                 }
1541         }
1542
1543         if (ret && !insert) {
1544                 err = -ENOENT;
1545                 goto out;
1546         } else if (ret) {
1547                 err = -EIO;
1548                 WARN_ON(1);
1549                 goto out;
1550         }
1551
1552         leaf = path->nodes[0];
1553         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555         if (item_size < sizeof(*ei)) {
1556                 if (!insert) {
1557                         err = -ENOENT;
1558                         goto out;
1559                 }
1560                 ret = convert_extent_item_v0(trans, root, path, owner,
1561                                              extra_size);
1562                 if (ret < 0) {
1563                         err = ret;
1564                         goto out;
1565                 }
1566                 leaf = path->nodes[0];
1567                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1568         }
1569 #endif
1570         BUG_ON(item_size < sizeof(*ei));
1571
1572         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573         flags = btrfs_extent_flags(leaf, ei);
1574
1575         ptr = (unsigned long)(ei + 1);
1576         end = (unsigned long)ei + item_size;
1577
1578         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1579                 ptr += sizeof(struct btrfs_tree_block_info);
1580                 BUG_ON(ptr > end);
1581         }
1582
1583         err = -ENOENT;
1584         while (1) {
1585                 if (ptr >= end) {
1586                         WARN_ON(ptr > end);
1587                         break;
1588                 }
1589                 iref = (struct btrfs_extent_inline_ref *)ptr;
1590                 type = btrfs_extent_inline_ref_type(leaf, iref);
1591                 if (want < type)
1592                         break;
1593                 if (want > type) {
1594                         ptr += btrfs_extent_inline_ref_size(type);
1595                         continue;
1596                 }
1597
1598                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1599                         struct btrfs_extent_data_ref *dref;
1600                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1601                         if (match_extent_data_ref(leaf, dref, root_objectid,
1602                                                   owner, offset)) {
1603                                 err = 0;
1604                                 break;
1605                         }
1606                         if (hash_extent_data_ref_item(leaf, dref) <
1607                             hash_extent_data_ref(root_objectid, owner, offset))
1608                                 break;
1609                 } else {
1610                         u64 ref_offset;
1611                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1612                         if (parent > 0) {
1613                                 if (parent == ref_offset) {
1614                                         err = 0;
1615                                         break;
1616                                 }
1617                                 if (ref_offset < parent)
1618                                         break;
1619                         } else {
1620                                 if (root_objectid == ref_offset) {
1621                                         err = 0;
1622                                         break;
1623                                 }
1624                                 if (ref_offset < root_objectid)
1625                                         break;
1626                         }
1627                 }
1628                 ptr += btrfs_extent_inline_ref_size(type);
1629         }
1630         if (err == -ENOENT && insert) {
1631                 if (item_size + extra_size >=
1632                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1633                         err = -EAGAIN;
1634                         goto out;
1635                 }
1636                 /*
1637                  * To add new inline back ref, we have to make sure
1638                  * there is no corresponding back ref item.
1639                  * For simplicity, we just do not add new inline back
1640                  * ref if there is any kind of item for this block
1641                  */
1642                 if (find_next_key(path, 0, &key) == 0 &&
1643                     key.objectid == bytenr &&
1644                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1645                         err = -EAGAIN;
1646                         goto out;
1647                 }
1648         }
1649         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1650 out:
1651         if (insert) {
1652                 path->keep_locks = 0;
1653                 btrfs_unlock_up_safe(path, 1);
1654         }
1655         return err;
1656 }
1657
1658 /*
1659  * helper to add new inline back ref
1660  */
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1663                                  struct btrfs_root *root,
1664                                  struct btrfs_path *path,
1665                                  struct btrfs_extent_inline_ref *iref,
1666                                  u64 parent, u64 root_objectid,
1667                                  u64 owner, u64 offset, int refs_to_add,
1668                                  struct btrfs_delayed_extent_op *extent_op)
1669 {
1670         struct extent_buffer *leaf;
1671         struct btrfs_extent_item *ei;
1672         unsigned long ptr;
1673         unsigned long end;
1674         unsigned long item_offset;
1675         u64 refs;
1676         int size;
1677         int type;
1678
1679         leaf = path->nodes[0];
1680         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1681         item_offset = (unsigned long)iref - (unsigned long)ei;
1682
1683         type = extent_ref_type(parent, owner);
1684         size = btrfs_extent_inline_ref_size(type);
1685
1686         btrfs_extend_item(trans, root, path, size);
1687
1688         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1689         refs = btrfs_extent_refs(leaf, ei);
1690         refs += refs_to_add;
1691         btrfs_set_extent_refs(leaf, ei, refs);
1692         if (extent_op)
1693                 __run_delayed_extent_op(extent_op, leaf, ei);
1694
1695         ptr = (unsigned long)ei + item_offset;
1696         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1697         if (ptr < end - size)
1698                 memmove_extent_buffer(leaf, ptr + size, ptr,
1699                                       end - size - ptr);
1700
1701         iref = (struct btrfs_extent_inline_ref *)ptr;
1702         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1703         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1704                 struct btrfs_extent_data_ref *dref;
1705                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1706                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1707                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1708                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1709                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1710         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1711                 struct btrfs_shared_data_ref *sref;
1712                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1713                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1714                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1715         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1716                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1717         } else {
1718                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1719         }
1720         btrfs_mark_buffer_dirty(leaf);
1721 }
1722
1723 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1724                                  struct btrfs_root *root,
1725                                  struct btrfs_path *path,
1726                                  struct btrfs_extent_inline_ref **ref_ret,
1727                                  u64 bytenr, u64 num_bytes, u64 parent,
1728                                  u64 root_objectid, u64 owner, u64 offset)
1729 {
1730         int ret;
1731
1732         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1733                                            bytenr, num_bytes, parent,
1734                                            root_objectid, owner, offset, 0);
1735         if (ret != -ENOENT)
1736                 return ret;
1737
1738         btrfs_release_path(path);
1739         *ref_ret = NULL;
1740
1741         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1742                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1743                                             root_objectid);
1744         } else {
1745                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1746                                              root_objectid, owner, offset);
1747         }
1748         return ret;
1749 }
1750
1751 /*
1752  * helper to update/remove inline back ref
1753  */
1754 static noinline_for_stack
1755 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1756                                   struct btrfs_root *root,
1757                                   struct btrfs_path *path,
1758                                   struct btrfs_extent_inline_ref *iref,
1759                                   int refs_to_mod,
1760                                   struct btrfs_delayed_extent_op *extent_op)
1761 {
1762         struct extent_buffer *leaf;
1763         struct btrfs_extent_item *ei;
1764         struct btrfs_extent_data_ref *dref = NULL;
1765         struct btrfs_shared_data_ref *sref = NULL;
1766         unsigned long ptr;
1767         unsigned long end;
1768         u32 item_size;
1769         int size;
1770         int type;
1771         u64 refs;
1772
1773         leaf = path->nodes[0];
1774         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1775         refs = btrfs_extent_refs(leaf, ei);
1776         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1777         refs += refs_to_mod;
1778         btrfs_set_extent_refs(leaf, ei, refs);
1779         if (extent_op)
1780                 __run_delayed_extent_op(extent_op, leaf, ei);
1781
1782         type = btrfs_extent_inline_ref_type(leaf, iref);
1783
1784         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1785                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1786                 refs = btrfs_extent_data_ref_count(leaf, dref);
1787         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1789                 refs = btrfs_shared_data_ref_count(leaf, sref);
1790         } else {
1791                 refs = 1;
1792                 BUG_ON(refs_to_mod != -1);
1793         }
1794
1795         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1796         refs += refs_to_mod;
1797
1798         if (refs > 0) {
1799                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1800                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1801                 else
1802                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1803         } else {
1804                 size =  btrfs_extent_inline_ref_size(type);
1805                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1806                 ptr = (unsigned long)iref;
1807                 end = (unsigned long)ei + item_size;
1808                 if (ptr + size < end)
1809                         memmove_extent_buffer(leaf, ptr, ptr + size,
1810                                               end - ptr - size);
1811                 item_size -= size;
1812                 btrfs_truncate_item(trans, root, path, item_size, 1);
1813         }
1814         btrfs_mark_buffer_dirty(leaf);
1815 }
1816
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1819                                  struct btrfs_root *root,
1820                                  struct btrfs_path *path,
1821                                  u64 bytenr, u64 num_bytes, u64 parent,
1822                                  u64 root_objectid, u64 owner,
1823                                  u64 offset, int refs_to_add,
1824                                  struct btrfs_delayed_extent_op *extent_op)
1825 {
1826         struct btrfs_extent_inline_ref *iref;
1827         int ret;
1828
1829         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1830                                            bytenr, num_bytes, parent,
1831                                            root_objectid, owner, offset, 1);
1832         if (ret == 0) {
1833                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1834                 update_inline_extent_backref(trans, root, path, iref,
1835                                              refs_to_add, extent_op);
1836         } else if (ret == -ENOENT) {
1837                 setup_inline_extent_backref(trans, root, path, iref, parent,
1838                                             root_objectid, owner, offset,
1839                                             refs_to_add, extent_op);
1840                 ret = 0;
1841         }
1842         return ret;
1843 }
1844
1845 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1846                                  struct btrfs_root *root,
1847                                  struct btrfs_path *path,
1848                                  u64 bytenr, u64 parent, u64 root_objectid,
1849                                  u64 owner, u64 offset, int refs_to_add)
1850 {
1851         int ret;
1852         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1853                 BUG_ON(refs_to_add != 1);
1854                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1855                                             parent, root_objectid);
1856         } else {
1857                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1858                                              parent, root_objectid,
1859                                              owner, offset, refs_to_add);
1860         }
1861         return ret;
1862 }
1863
1864 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1865                                  struct btrfs_root *root,
1866                                  struct btrfs_path *path,
1867                                  struct btrfs_extent_inline_ref *iref,
1868                                  int refs_to_drop, int is_data)
1869 {
1870         int ret = 0;
1871
1872         BUG_ON(!is_data && refs_to_drop != 1);
1873         if (iref) {
1874                 update_inline_extent_backref(trans, root, path, iref,
1875                                              -refs_to_drop, NULL);
1876         } else if (is_data) {
1877                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1878         } else {
1879                 ret = btrfs_del_item(trans, root, path);
1880         }
1881         return ret;
1882 }
1883
1884 static int btrfs_issue_discard(struct block_device *bdev,
1885                                 u64 start, u64 len)
1886 {
1887         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1888 }
1889
1890 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1891                                 u64 num_bytes, u64 *actual_bytes)
1892 {
1893         int ret;
1894         u64 discarded_bytes = 0;
1895         struct btrfs_bio *bbio = NULL;
1896
1897
1898         /* Tell the block device(s) that the sectors can be discarded */
1899         ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1900                               bytenr, &num_bytes, &bbio, 0);
1901         /* Error condition is -ENOMEM */
1902         if (!ret) {
1903                 struct btrfs_bio_stripe *stripe = bbio->stripes;
1904                 int i;
1905
1906
1907                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1908                         if (!stripe->dev->can_discard)
1909                                 continue;
1910
1911                         ret = btrfs_issue_discard(stripe->dev->bdev,
1912                                                   stripe->physical,
1913                                                   stripe->length);
1914                         if (!ret)
1915                                 discarded_bytes += stripe->length;
1916                         else if (ret != -EOPNOTSUPP)
1917                                 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1918
1919                         /*
1920                          * Just in case we get back EOPNOTSUPP for some reason,
1921                          * just ignore the return value so we don't screw up
1922                          * people calling discard_extent.
1923                          */
1924                         ret = 0;
1925                 }
1926                 kfree(bbio);
1927         }
1928
1929         if (actual_bytes)
1930                 *actual_bytes = discarded_bytes;
1931
1932
1933         if (ret == -EOPNOTSUPP)
1934                 ret = 0;
1935         return ret;
1936 }
1937
1938 /* Can return -ENOMEM */
1939 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1940                          struct btrfs_root *root,
1941                          u64 bytenr, u64 num_bytes, u64 parent,
1942                          u64 root_objectid, u64 owner, u64 offset, int for_cow)
1943 {
1944         int ret;
1945         struct btrfs_fs_info *fs_info = root->fs_info;
1946
1947         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1948                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1949
1950         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1951                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1952                                         num_bytes,
1953                                         parent, root_objectid, (int)owner,
1954                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1955         } else {
1956                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1957                                         num_bytes,
1958                                         parent, root_objectid, owner, offset,
1959                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1960         }
1961         return ret;
1962 }
1963
1964 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1965                                   struct btrfs_root *root,
1966                                   u64 bytenr, u64 num_bytes,
1967                                   u64 parent, u64 root_objectid,
1968                                   u64 owner, u64 offset, int refs_to_add,
1969                                   struct btrfs_delayed_extent_op *extent_op)
1970 {
1971         struct btrfs_path *path;
1972         struct extent_buffer *leaf;
1973         struct btrfs_extent_item *item;
1974         u64 refs;
1975         int ret;
1976         int err = 0;
1977
1978         path = btrfs_alloc_path();
1979         if (!path)
1980                 return -ENOMEM;
1981
1982         path->reada = 1;
1983         path->leave_spinning = 1;
1984         /* this will setup the path even if it fails to insert the back ref */
1985         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1986                                            path, bytenr, num_bytes, parent,
1987                                            root_objectid, owner, offset,
1988                                            refs_to_add, extent_op);
1989         if (ret == 0)
1990                 goto out;
1991
1992         if (ret != -EAGAIN) {
1993                 err = ret;
1994                 goto out;
1995         }
1996
1997         leaf = path->nodes[0];
1998         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1999         refs = btrfs_extent_refs(leaf, item);
2000         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2001         if (extent_op)
2002                 __run_delayed_extent_op(extent_op, leaf, item);
2003
2004         btrfs_mark_buffer_dirty(leaf);
2005         btrfs_release_path(path);
2006
2007         path->reada = 1;
2008         path->leave_spinning = 1;
2009
2010         /* now insert the actual backref */
2011         ret = insert_extent_backref(trans, root->fs_info->extent_root,
2012                                     path, bytenr, parent, root_objectid,
2013                                     owner, offset, refs_to_add);
2014         if (ret)
2015                 btrfs_abort_transaction(trans, root, ret);
2016 out:
2017         btrfs_free_path(path);
2018         return err;
2019 }
2020
2021 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2022                                 struct btrfs_root *root,
2023                                 struct btrfs_delayed_ref_node *node,
2024                                 struct btrfs_delayed_extent_op *extent_op,
2025                                 int insert_reserved)
2026 {
2027         int ret = 0;
2028         struct btrfs_delayed_data_ref *ref;
2029         struct btrfs_key ins;
2030         u64 parent = 0;
2031         u64 ref_root = 0;
2032         u64 flags = 0;
2033
2034         ins.objectid = node->bytenr;
2035         ins.offset = node->num_bytes;
2036         ins.type = BTRFS_EXTENT_ITEM_KEY;
2037
2038         ref = btrfs_delayed_node_to_data_ref(node);
2039         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 = (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2100                         node->type == BTRFS_SHARED_BLOCK_REF_KEY);
2101
2102         if (trans->aborted)
2103                 return 0;
2104
2105         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2106                 metadata = 0;
2107
2108         path = btrfs_alloc_path();
2109         if (!path)
2110                 return -ENOMEM;
2111
2112         key.objectid = node->bytenr;
2113
2114         if (metadata) {
2115                 struct btrfs_delayed_tree_ref *tree_ref;
2116
2117                 tree_ref = btrfs_delayed_node_to_tree_ref(node);
2118                 key.type = BTRFS_METADATA_ITEM_KEY;
2119                 key.offset = tree_ref->level;
2120         } else {
2121                 key.type = BTRFS_EXTENT_ITEM_KEY;
2122                 key.offset = node->num_bytes;
2123         }
2124
2125 again:
2126         path->reada = 1;
2127         path->leave_spinning = 1;
2128         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2129                                 path, 0, 1);
2130         if (ret < 0) {
2131                 err = ret;
2132                 goto out;
2133         }
2134         if (ret > 0) {
2135                 if (metadata) {
2136                         btrfs_release_path(path);
2137                         metadata = 0;
2138
2139                         key.offset = node->num_bytes;
2140                         key.type = BTRFS_EXTENT_ITEM_KEY;
2141                         goto again;
2142                 }
2143                 err = -EIO;
2144                 goto out;
2145         }
2146
2147         leaf = path->nodes[0];
2148         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2149 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2150         if (item_size < sizeof(*ei)) {
2151                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2152                                              path, (u64)-1, 0);
2153                 if (ret < 0) {
2154                         err = ret;
2155                         goto out;
2156                 }
2157                 leaf = path->nodes[0];
2158                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2159         }
2160 #endif
2161         BUG_ON(item_size < sizeof(*ei));
2162         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2163         __run_delayed_extent_op(extent_op, leaf, ei);
2164
2165         btrfs_mark_buffer_dirty(leaf);
2166 out:
2167         btrfs_free_path(path);
2168         return err;
2169 }
2170
2171 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2172                                 struct btrfs_root *root,
2173                                 struct btrfs_delayed_ref_node *node,
2174                                 struct btrfs_delayed_extent_op *extent_op,
2175                                 int insert_reserved)
2176 {
2177         int ret = 0;
2178         struct btrfs_delayed_tree_ref *ref;
2179         struct btrfs_key ins;
2180         u64 parent = 0;
2181         u64 ref_root = 0;
2182         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2183                                                  SKINNY_METADATA);
2184
2185         ref = btrfs_delayed_node_to_tree_ref(node);
2186         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2187                 parent = ref->parent;
2188         else
2189                 ref_root = ref->root;
2190
2191         ins.objectid = node->bytenr;
2192         if (skinny_metadata) {
2193                 ins.offset = ref->level;
2194                 ins.type = BTRFS_METADATA_ITEM_KEY;
2195         } else {
2196                 ins.offset = node->num_bytes;
2197                 ins.type = BTRFS_EXTENT_ITEM_KEY;
2198         }
2199
2200         BUG_ON(node->ref_mod != 1);
2201         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2202                 BUG_ON(!extent_op || !extent_op->update_flags);
2203                 ret = alloc_reserved_tree_block(trans, root,
2204                                                 parent, ref_root,
2205                                                 extent_op->flags_to_set,
2206                                                 &extent_op->key,
2207                                                 ref->level, &ins);
2208         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2209                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2210                                              node->num_bytes, parent, ref_root,
2211                                              ref->level, 0, 1, extent_op);
2212         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2213                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2214                                           node->num_bytes, parent, ref_root,
2215                                           ref->level, 0, 1, extent_op);
2216         } else {
2217                 BUG();
2218         }
2219         return ret;
2220 }
2221
2222 /* helper function to actually process a single delayed ref entry */
2223 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2224                                struct btrfs_root *root,
2225                                struct btrfs_delayed_ref_node *node,
2226                                struct btrfs_delayed_extent_op *extent_op,
2227                                int insert_reserved)
2228 {
2229         int ret = 0;
2230
2231         if (trans->aborted)
2232                 return 0;
2233
2234         if (btrfs_delayed_ref_is_head(node)) {
2235                 struct btrfs_delayed_ref_head *head;
2236                 /*
2237                  * we've hit the end of the chain and we were supposed
2238                  * to insert this extent into the tree.  But, it got
2239                  * deleted before we ever needed to insert it, so all
2240                  * we have to do is clean up the accounting
2241                  */
2242                 BUG_ON(extent_op);
2243                 head = btrfs_delayed_node_to_head(node);
2244                 if (insert_reserved) {
2245                         btrfs_pin_extent(root, node->bytenr,
2246                                          node->num_bytes, 1);
2247                         if (head->is_data) {
2248                                 ret = btrfs_del_csums(trans, root,
2249                                                       node->bytenr,
2250                                                       node->num_bytes);
2251                         }
2252                 }
2253                 return ret;
2254         }
2255
2256         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2257             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2258                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2259                                            insert_reserved);
2260         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2261                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2262                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2263                                            insert_reserved);
2264         else
2265                 BUG();
2266         return ret;
2267 }
2268
2269 static noinline struct btrfs_delayed_ref_node *
2270 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2271 {
2272         struct rb_node *node;
2273         struct btrfs_delayed_ref_node *ref;
2274         int action = BTRFS_ADD_DELAYED_REF;
2275 again:
2276         /*
2277          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2278          * this prevents ref count from going down to zero when
2279          * there still are pending delayed ref.
2280          */
2281         node = rb_prev(&head->node.rb_node);
2282         while (1) {
2283                 if (!node)
2284                         break;
2285                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2286                                 rb_node);
2287                 if (ref->bytenr != head->node.bytenr)
2288                         break;
2289                 if (ref->action == action)
2290                         return ref;
2291                 node = rb_prev(node);
2292         }
2293         if (action == BTRFS_ADD_DELAYED_REF) {
2294                 action = BTRFS_DROP_DELAYED_REF;
2295                 goto again;
2296         }
2297         return NULL;
2298 }
2299
2300 /*
2301  * Returns 0 on success or if called with an already aborted transaction.
2302  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2303  */
2304 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2305                                        struct btrfs_root *root,
2306                                        struct list_head *cluster)
2307 {
2308         struct btrfs_delayed_ref_root *delayed_refs;
2309         struct btrfs_delayed_ref_node *ref;
2310         struct btrfs_delayed_ref_head *locked_ref = NULL;
2311         struct btrfs_delayed_extent_op *extent_op;
2312         struct btrfs_fs_info *fs_info = root->fs_info;
2313         int ret;
2314         int count = 0;
2315         int must_insert_reserved = 0;
2316
2317         delayed_refs = &trans->transaction->delayed_refs;
2318         while (1) {
2319                 if (!locked_ref) {
2320                         /* pick a new head ref from the cluster list */
2321                         if (list_empty(cluster))
2322                                 break;
2323
2324                         locked_ref = list_entry(cluster->next,
2325                                      struct btrfs_delayed_ref_head, cluster);
2326
2327                         /* grab the lock that says we are going to process
2328                          * all the refs for this head */
2329                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2330
2331                         /*
2332                          * we may have dropped the spin lock to get the head
2333                          * mutex lock, and that might have given someone else
2334                          * time to free the head.  If that's true, it has been
2335                          * removed from our list and we can move on.
2336                          */
2337                         if (ret == -EAGAIN) {
2338                                 locked_ref = NULL;
2339                                 count++;
2340                                 continue;
2341                         }
2342                 }
2343
2344                 /*
2345                  * We need to try and merge add/drops of the same ref since we
2346                  * can run into issues with relocate dropping the implicit ref
2347                  * and then it being added back again before the drop can
2348                  * finish.  If we merged anything we need to re-loop so we can
2349                  * get a good ref.
2350                  */
2351                 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2352                                          locked_ref);
2353
2354                 /*
2355                  * locked_ref is the head node, so we have to go one
2356                  * node back for any delayed ref updates
2357                  */
2358                 ref = select_delayed_ref(locked_ref);
2359
2360                 if (ref && ref->seq &&
2361                     btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2362                         /*
2363                          * there are still refs with lower seq numbers in the
2364                          * process of being added. Don't run this ref yet.
2365                          */
2366                         list_del_init(&locked_ref->cluster);
2367                         btrfs_delayed_ref_unlock(locked_ref);
2368                         locked_ref = NULL;
2369                         delayed_refs->num_heads_ready++;
2370                         spin_unlock(&delayed_refs->lock);
2371                         cond_resched();
2372                         spin_lock(&delayed_refs->lock);
2373                         continue;
2374                 }
2375
2376                 /*
2377                  * record the must insert reserved flag before we
2378                  * drop the spin lock.
2379                  */
2380                 must_insert_reserved = locked_ref->must_insert_reserved;
2381                 locked_ref->must_insert_reserved = 0;
2382
2383                 extent_op = locked_ref->extent_op;
2384                 locked_ref->extent_op = NULL;
2385
2386                 if (!ref) {
2387                         /* All delayed refs have been processed, Go ahead
2388                          * and send the head node to run_one_delayed_ref,
2389                          * so that any accounting fixes can happen
2390                          */
2391                         ref = &locked_ref->node;
2392
2393                         if (extent_op && must_insert_reserved) {
2394                                 btrfs_free_delayed_extent_op(extent_op);
2395                                 extent_op = NULL;
2396                         }
2397
2398                         if (extent_op) {
2399                                 spin_unlock(&delayed_refs->lock);
2400
2401                                 ret = run_delayed_extent_op(trans, root,
2402                                                             ref, extent_op);
2403                                 btrfs_free_delayed_extent_op(extent_op);
2404
2405                                 if (ret) {
2406                                         printk(KERN_DEBUG
2407                                                "btrfs: run_delayed_extent_op "
2408                                                "returned %d\n", ret);
2409                                         spin_lock(&delayed_refs->lock);
2410                                         btrfs_delayed_ref_unlock(locked_ref);
2411                                         return ret;
2412                                 }
2413
2414                                 goto next;
2415                         }
2416                 }
2417
2418                 ref->in_tree = 0;
2419                 rb_erase(&ref->rb_node, &delayed_refs->root);
2420                 delayed_refs->num_entries--;
2421                 if (!btrfs_delayed_ref_is_head(ref)) {
2422                         /*
2423                          * when we play the delayed ref, also correct the
2424                          * ref_mod on head
2425                          */
2426                         switch (ref->action) {
2427                         case BTRFS_ADD_DELAYED_REF:
2428                         case BTRFS_ADD_DELAYED_EXTENT:
2429                                 locked_ref->node.ref_mod -= ref->ref_mod;
2430                                 break;
2431                         case BTRFS_DROP_DELAYED_REF:
2432                                 locked_ref->node.ref_mod += ref->ref_mod;
2433                                 break;
2434                         default:
2435                                 WARN_ON(1);
2436                         }
2437                 }
2438                 spin_unlock(&delayed_refs->lock);
2439
2440                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2441                                           must_insert_reserved);
2442
2443                 btrfs_free_delayed_extent_op(extent_op);
2444                 if (ret) {
2445                         btrfs_delayed_ref_unlock(locked_ref);
2446                         btrfs_put_delayed_ref(ref);
2447                         printk(KERN_DEBUG
2448                                "btrfs: run_one_delayed_ref returned %d\n", ret);
2449                         spin_lock(&delayed_refs->lock);
2450                         return ret;
2451                 }
2452
2453                 /*
2454                  * If this node is a head, that means all the refs in this head
2455                  * have been dealt with, and we will pick the next head to deal
2456                  * with, so we must unlock the head and drop it from the cluster
2457                  * list before we release it.
2458                  */
2459                 if (btrfs_delayed_ref_is_head(ref)) {
2460                         list_del_init(&locked_ref->cluster);
2461                         btrfs_delayed_ref_unlock(locked_ref);
2462                         locked_ref = NULL;
2463                 }
2464                 btrfs_put_delayed_ref(ref);
2465                 count++;
2466 next:
2467                 cond_resched();
2468                 spin_lock(&delayed_refs->lock);
2469         }
2470         return count;
2471 }
2472
2473 #ifdef SCRAMBLE_DELAYED_REFS
2474 /*
2475  * Normally delayed refs get processed in ascending bytenr order. This
2476  * correlates in most cases to the order added. To expose dependencies on this
2477  * order, we start to process the tree in the middle instead of the beginning
2478  */
2479 static u64 find_middle(struct rb_root *root)
2480 {
2481         struct rb_node *n = root->rb_node;
2482         struct btrfs_delayed_ref_node *entry;
2483         int alt = 1;
2484         u64 middle;
2485         u64 first = 0, last = 0;
2486
2487         n = rb_first(root);
2488         if (n) {
2489                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2490                 first = entry->bytenr;
2491         }
2492         n = rb_last(root);
2493         if (n) {
2494                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2495                 last = entry->bytenr;
2496         }
2497         n = root->rb_node;
2498
2499         while (n) {
2500                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2501                 WARN_ON(!entry->in_tree);
2502
2503                 middle = entry->bytenr;
2504
2505                 if (alt)
2506                         n = n->rb_left;
2507                 else
2508                         n = n->rb_right;
2509
2510                 alt = 1 - alt;
2511         }
2512         return middle;
2513 }
2514 #endif
2515
2516 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2517                                          struct btrfs_fs_info *fs_info)
2518 {
2519         struct qgroup_update *qgroup_update;
2520         int ret = 0;
2521
2522         if (list_empty(&trans->qgroup_ref_list) !=
2523             !trans->delayed_ref_elem.seq) {
2524                 /* list without seq or seq without list */
2525                 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2526                         list_empty(&trans->qgroup_ref_list) ? "" : " not",
2527                         trans->delayed_ref_elem.seq);
2528                 BUG();
2529         }
2530
2531         if (!trans->delayed_ref_elem.seq)
2532                 return 0;
2533
2534         while (!list_empty(&trans->qgroup_ref_list)) {
2535                 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2536                                                  struct qgroup_update, list);
2537                 list_del(&qgroup_update->list);
2538                 if (!ret)
2539                         ret = btrfs_qgroup_account_ref(
2540                                         trans, fs_info, qgroup_update->node,
2541                                         qgroup_update->extent_op);
2542                 kfree(qgroup_update);
2543         }
2544
2545         btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2546
2547         return ret;
2548 }
2549
2550 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2551                       int count)
2552 {
2553         int val = atomic_read(&delayed_refs->ref_seq);
2554
2555         if (val < seq || val >= seq + count)
2556                 return 1;
2557         return 0;
2558 }
2559
2560 /*
2561  * this starts processing the delayed reference count updates and
2562  * extent insertions we have queued up so far.  count can be
2563  * 0, which means to process everything in the tree at the start
2564  * of the run (but not newly added entries), or it can be some target
2565  * number you'd like to process.
2566  *
2567  * Returns 0 on success or if called with an aborted transaction
2568  * Returns <0 on error and aborts the transaction
2569  */
2570 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2571                            struct btrfs_root *root, unsigned long count)
2572 {
2573         struct rb_node *node;
2574         struct btrfs_delayed_ref_root *delayed_refs;
2575         struct btrfs_delayed_ref_node *ref;
2576         struct list_head cluster;
2577         int ret;
2578         u64 delayed_start;
2579         int run_all = count == (unsigned long)-1;
2580         int run_most = 0;
2581         int loops;
2582
2583         /* We'll clean this up in btrfs_cleanup_transaction */
2584         if (trans->aborted)
2585                 return 0;
2586
2587         if (root == root->fs_info->extent_root)
2588                 root = root->fs_info->tree_root;
2589
2590         btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2591
2592         delayed_refs = &trans->transaction->delayed_refs;
2593         INIT_LIST_HEAD(&cluster);
2594         if (count == 0) {
2595                 count = delayed_refs->num_entries * 2;
2596                 run_most = 1;
2597         }
2598
2599         if (!run_all && !run_most) {
2600                 int old;
2601                 int seq = atomic_read(&delayed_refs->ref_seq);
2602
2603 progress:
2604                 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2605                 if (old) {
2606                         DEFINE_WAIT(__wait);
2607                         if (delayed_refs->num_entries < 16348)
2608                                 return 0;
2609
2610                         prepare_to_wait(&delayed_refs->wait, &__wait,
2611                                         TASK_UNINTERRUPTIBLE);
2612
2613                         old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2614                         if (old) {
2615                                 schedule();
2616                                 finish_wait(&delayed_refs->wait, &__wait);
2617
2618                                 if (!refs_newer(delayed_refs, seq, 256))
2619                                         goto progress;
2620                                 else
2621                                         return 0;
2622                         } else {
2623                                 finish_wait(&delayed_refs->wait, &__wait);
2624                                 goto again;
2625                         }
2626                 }
2627
2628         } else {
2629                 atomic_inc(&delayed_refs->procs_running_refs);
2630         }
2631
2632 again:
2633         loops = 0;
2634         spin_lock(&delayed_refs->lock);
2635
2636 #ifdef SCRAMBLE_DELAYED_REFS
2637         delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2638 #endif
2639
2640         while (1) {
2641                 if (!(run_all || run_most) &&
2642                     delayed_refs->num_heads_ready < 64)
2643                         break;
2644
2645                 /*
2646                  * go find something we can process in the rbtree.  We start at
2647                  * the beginning of the tree, and then build a cluster
2648                  * of refs to process starting at the first one we are able to
2649                  * lock
2650                  */
2651                 delayed_start = delayed_refs->run_delayed_start;
2652                 ret = btrfs_find_ref_cluster(trans, &cluster,
2653                                              delayed_refs->run_delayed_start);
2654                 if (ret)
2655                         break;
2656
2657                 ret = run_clustered_refs(trans, root, &cluster);
2658                 if (ret < 0) {
2659                         btrfs_release_ref_cluster(&cluster);
2660                         spin_unlock(&delayed_refs->lock);
2661                         btrfs_abort_transaction(trans, root, ret);
2662                         atomic_dec(&delayed_refs->procs_running_refs);
2663                         return ret;
2664                 }
2665
2666                 atomic_add(ret, &delayed_refs->ref_seq);
2667
2668                 count -= min_t(unsigned long, ret, count);
2669
2670                 if (count == 0)
2671                         break;
2672
2673                 if (delayed_start >= delayed_refs->run_delayed_start) {
2674                         if (loops == 0) {
2675                                 /*
2676                                  * btrfs_find_ref_cluster looped. let's do one
2677                                  * more cycle. if we don't run any delayed ref
2678                                  * during that cycle (because we can't because
2679                                  * all of them are blocked), bail out.
2680                                  */
2681                                 loops = 1;
2682                         } else {
2683                                 /*
2684                                  * no runnable refs left, stop trying
2685                                  */
2686                                 BUG_ON(run_all);
2687                                 break;
2688                         }
2689                 }
2690                 if (ret) {
2691                         /* refs were run, let's reset staleness detection */
2692                         loops = 0;
2693                 }
2694         }
2695
2696         if (run_all) {
2697                 if (!list_empty(&trans->new_bgs)) {
2698                         spin_unlock(&delayed_refs->lock);
2699                         btrfs_create_pending_block_groups(trans, root);
2700                         spin_lock(&delayed_refs->lock);
2701                 }
2702
2703                 node = rb_first(&delayed_refs->root);
2704                 if (!node)
2705                         goto out;
2706                 count = (unsigned long)-1;
2707
2708                 while (node) {
2709                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2710                                        rb_node);
2711                         if (btrfs_delayed_ref_is_head(ref)) {
2712                                 struct btrfs_delayed_ref_head *head;
2713
2714                                 head = btrfs_delayed_node_to_head(ref);
2715                                 atomic_inc(&ref->refs);
2716
2717                                 spin_unlock(&delayed_refs->lock);
2718                                 /*
2719                                  * Mutex was contended, block until it's
2720                                  * released and try again
2721                                  */
2722                                 mutex_lock(&head->mutex);
2723                                 mutex_unlock(&head->mutex);
2724
2725                                 btrfs_put_delayed_ref(ref);
2726                                 cond_resched();
2727                                 goto again;
2728                         }
2729                         node = rb_next(node);
2730                 }
2731                 spin_unlock(&delayed_refs->lock);
2732                 schedule_timeout(1);
2733                 goto again;
2734         }
2735 out:
2736         atomic_dec(&delayed_refs->procs_running_refs);
2737         smp_mb();
2738         if (waitqueue_active(&delayed_refs->wait))
2739                 wake_up(&delayed_refs->wait);
2740
2741         spin_unlock(&delayed_refs->lock);
2742         assert_qgroups_uptodate(trans);
2743         return 0;
2744 }
2745
2746 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2747                                 struct btrfs_root *root,
2748                                 u64 bytenr, u64 num_bytes, u64 flags,
2749                                 int is_data)
2750 {
2751         struct btrfs_delayed_extent_op *extent_op;
2752         int ret;
2753
2754         extent_op = btrfs_alloc_delayed_extent_op();
2755         if (!extent_op)
2756                 return -ENOMEM;
2757
2758         extent_op->flags_to_set = flags;
2759         extent_op->update_flags = 1;
2760         extent_op->update_key = 0;
2761         extent_op->is_data = is_data ? 1 : 0;
2762
2763         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2764                                           num_bytes, extent_op);
2765         if (ret)
2766                 btrfs_free_delayed_extent_op(extent_op);
2767         return ret;
2768 }
2769
2770 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2771                                       struct btrfs_root *root,
2772                                       struct btrfs_path *path,
2773                                       u64 objectid, u64 offset, u64 bytenr)
2774 {
2775         struct btrfs_delayed_ref_head *head;
2776         struct btrfs_delayed_ref_node *ref;
2777         struct btrfs_delayed_data_ref *data_ref;
2778         struct btrfs_delayed_ref_root *delayed_refs;
2779         struct rb_node *node;
2780         int ret = 0;
2781
2782         ret = -ENOENT;
2783         delayed_refs = &trans->transaction->delayed_refs;
2784         spin_lock(&delayed_refs->lock);
2785         head = btrfs_find_delayed_ref_head(trans, bytenr);
2786         if (!head)
2787                 goto out;
2788
2789         if (!mutex_trylock(&head->mutex)) {
2790                 atomic_inc(&head->node.refs);
2791                 spin_unlock(&delayed_refs->lock);
2792
2793                 btrfs_release_path(path);
2794
2795                 /*
2796                  * Mutex was contended, block until it's released and let
2797                  * caller try again
2798                  */
2799                 mutex_lock(&head->mutex);
2800                 mutex_unlock(&head->mutex);
2801                 btrfs_put_delayed_ref(&head->node);
2802                 return -EAGAIN;
2803         }
2804
2805         node = rb_prev(&head->node.rb_node);
2806         if (!node)
2807                 goto out_unlock;
2808
2809         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2810
2811         if (ref->bytenr != bytenr)
2812                 goto out_unlock;
2813
2814         ret = 1;
2815         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2816                 goto out_unlock;
2817
2818         data_ref = btrfs_delayed_node_to_data_ref(ref);
2819
2820         node = rb_prev(node);
2821         if (node) {
2822                 int seq = ref->seq;
2823
2824                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2825                 if (ref->bytenr == bytenr && ref->seq == seq)
2826                         goto out_unlock;
2827         }
2828
2829         if (data_ref->root != root->root_key.objectid ||
2830             data_ref->objectid != objectid || data_ref->offset != offset)
2831                 goto out_unlock;
2832
2833         ret = 0;
2834 out_unlock:
2835         mutex_unlock(&head->mutex);
2836 out:
2837         spin_unlock(&delayed_refs->lock);
2838         return ret;
2839 }
2840
2841 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2842                                         struct btrfs_root *root,
2843                                         struct btrfs_path *path,
2844                                         u64 objectid, u64 offset, u64 bytenr)
2845 {
2846         struct btrfs_root *extent_root = root->fs_info->extent_root;
2847         struct extent_buffer *leaf;
2848         struct btrfs_extent_data_ref *ref;
2849         struct btrfs_extent_inline_ref *iref;
2850         struct btrfs_extent_item *ei;
2851         struct btrfs_key key;
2852         u32 item_size;
2853         int ret;
2854
2855         key.objectid = bytenr;
2856         key.offset = (u64)-1;
2857         key.type = BTRFS_EXTENT_ITEM_KEY;
2858
2859         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2860         if (ret < 0)
2861                 goto out;
2862         BUG_ON(ret == 0); /* Corruption */
2863
2864         ret = -ENOENT;
2865         if (path->slots[0] == 0)
2866                 goto out;
2867
2868         path->slots[0]--;
2869         leaf = path->nodes[0];
2870         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2871
2872         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2873                 goto out;
2874
2875         ret = 1;
2876         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2877 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2878         if (item_size < sizeof(*ei)) {
2879                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2880                 goto out;
2881         }
2882 #endif
2883         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2884
2885         if (item_size != sizeof(*ei) +
2886             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2887                 goto out;
2888
2889         if (btrfs_extent_generation(leaf, ei) <=
2890             btrfs_root_last_snapshot(&root->root_item))
2891                 goto out;
2892
2893         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2894         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2895             BTRFS_EXTENT_DATA_REF_KEY)
2896                 goto out;
2897
2898         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2899         if (btrfs_extent_refs(leaf, ei) !=
2900             btrfs_extent_data_ref_count(leaf, ref) ||
2901             btrfs_extent_data_ref_root(leaf, ref) !=
2902             root->root_key.objectid ||
2903             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2904             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2905                 goto out;
2906
2907         ret = 0;
2908 out:
2909         return ret;
2910 }
2911
2912 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2913                           struct btrfs_root *root,
2914                           u64 objectid, u64 offset, u64 bytenr)
2915 {
2916         struct btrfs_path *path;
2917         int ret;
2918         int ret2;
2919
2920         path = btrfs_alloc_path();
2921         if (!path)
2922                 return -ENOENT;
2923
2924         do {
2925                 ret = check_committed_ref(trans, root, path, objectid,
2926                                           offset, bytenr);
2927                 if (ret && ret != -ENOENT)
2928                         goto out;
2929
2930                 ret2 = check_delayed_ref(trans, root, path, objectid,
2931                                          offset, bytenr);
2932         } while (ret2 == -EAGAIN);
2933
2934         if (ret2 && ret2 != -ENOENT) {
2935                 ret = ret2;
2936                 goto out;
2937         }
2938
2939         if (ret != -ENOENT || ret2 != -ENOENT)
2940                 ret = 0;
2941 out:
2942         btrfs_free_path(path);
2943         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2944                 WARN_ON(ret > 0);
2945         return ret;
2946 }
2947
2948 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2949                            struct btrfs_root *root,
2950                            struct extent_buffer *buf,
2951                            int full_backref, int inc, int for_cow)
2952 {
2953         u64 bytenr;
2954         u64 num_bytes;
2955         u64 parent;
2956         u64 ref_root;
2957         u32 nritems;
2958         struct btrfs_key key;
2959         struct btrfs_file_extent_item *fi;
2960         int i;
2961         int level;
2962         int ret = 0;
2963         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2964                             u64, u64, u64, u64, u64, u64, int);
2965
2966         ref_root = btrfs_header_owner(buf);
2967         nritems = btrfs_header_nritems(buf);
2968         level = btrfs_header_level(buf);
2969
2970         if (!root->ref_cows && level == 0)
2971                 return 0;
2972
2973         if (inc)
2974                 process_func = btrfs_inc_extent_ref;
2975         else
2976                 process_func = btrfs_free_extent;
2977
2978         if (full_backref)
2979                 parent = buf->start;
2980         else
2981                 parent = 0;
2982
2983         for (i = 0; i < nritems; i++) {
2984                 if (level == 0) {
2985                         btrfs_item_key_to_cpu(buf, &key, i);
2986                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2987                                 continue;
2988                         fi = btrfs_item_ptr(buf, i,
2989                                             struct btrfs_file_extent_item);
2990                         if (btrfs_file_extent_type(buf, fi) ==
2991                             BTRFS_FILE_EXTENT_INLINE)
2992                                 continue;
2993                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2994                         if (bytenr == 0)
2995                                 continue;
2996
2997                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2998                         key.offset -= btrfs_file_extent_offset(buf, fi);
2999                         ret = process_func(trans, root, bytenr, num_bytes,
3000                                            parent, ref_root, key.objectid,
3001                                            key.offset, for_cow);
3002                         if (ret)
3003                                 goto fail;
3004                 } else {
3005                         bytenr = btrfs_node_blockptr(buf, i);
3006                         num_bytes = btrfs_level_size(root, level - 1);
3007                         ret = process_func(trans, root, bytenr, num_bytes,
3008                                            parent, ref_root, level - 1, 0,
3009                                            for_cow);
3010                         if (ret)
3011                                 goto fail;
3012                 }
3013         }
3014         return 0;
3015 fail:
3016         return ret;
3017 }
3018
3019 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3020                   struct extent_buffer *buf, int full_backref, int for_cow)
3021 {
3022         return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3023 }
3024
3025 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3026                   struct extent_buffer *buf, int full_backref, int for_cow)
3027 {
3028         return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3029 }
3030
3031 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3032                                  struct btrfs_root *root,
3033                                  struct btrfs_path *path,
3034                                  struct btrfs_block_group_cache *cache)
3035 {
3036         int ret;
3037         struct btrfs_root *extent_root = root->fs_info->extent_root;
3038         unsigned long bi;
3039         struct extent_buffer *leaf;
3040
3041         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3042         if (ret < 0)
3043                 goto fail;
3044         BUG_ON(ret); /* Corruption */
3045
3046         leaf = path->nodes[0];
3047         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3048         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3049         btrfs_mark_buffer_dirty(leaf);
3050         btrfs_release_path(path);
3051 fail:
3052         if (ret) {
3053                 btrfs_abort_transaction(trans, root, ret);
3054                 return ret;
3055         }
3056         return 0;
3057
3058 }
3059
3060 static struct btrfs_block_group_cache *
3061 next_block_group(struct btrfs_root *root,
3062                  struct btrfs_block_group_cache *cache)
3063 {
3064         struct rb_node *node;
3065         spin_lock(&root->fs_info->block_group_cache_lock);
3066         node = rb_next(&cache->cache_node);
3067         btrfs_put_block_group(cache);
3068         if (node) {
3069                 cache = rb_entry(node, struct btrfs_block_group_cache,
3070                                  cache_node);
3071                 btrfs_get_block_group(cache);
3072         } else
3073                 cache = NULL;
3074         spin_unlock(&root->fs_info->block_group_cache_lock);
3075         return cache;
3076 }
3077
3078 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3079                             struct btrfs_trans_handle *trans,
3080                             struct btrfs_path *path)
3081 {
3082         struct btrfs_root *root = block_group->fs_info->tree_root;
3083         struct inode *inode = NULL;
3084         u64 alloc_hint = 0;
3085         int dcs = BTRFS_DC_ERROR;
3086         int num_pages = 0;
3087         int retries = 0;
3088         int ret = 0;
3089
3090         /*
3091          * If this block group is smaller than 100 megs don't bother caching the
3092          * block group.
3093          */
3094         if (block_group->key.offset < (100 * 1024 * 1024)) {
3095                 spin_lock(&block_group->lock);
3096                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3097                 spin_unlock(&block_group->lock);
3098                 return 0;
3099         }
3100
3101 again:
3102         inode = lookup_free_space_inode(root, block_group, path);
3103         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3104                 ret = PTR_ERR(inode);
3105                 btrfs_release_path(path);
3106                 goto out;
3107         }
3108
3109         if (IS_ERR(inode)) {
3110                 BUG_ON(retries);
3111                 retries++;
3112
3113                 if (block_group->ro)
3114                         goto out_free;
3115
3116                 ret = create_free_space_inode(root, trans, block_group, path);
3117                 if (ret)
3118                         goto out_free;
3119                 goto again;
3120         }
3121
3122         /* We've already setup this transaction, go ahead and exit */
3123         if (block_group->cache_generation == trans->transid &&
3124             i_size_read(inode)) {
3125                 dcs = BTRFS_DC_SETUP;
3126                 goto out_put;
3127         }
3128
3129         /*
3130          * We want to set the generation to 0, that way if anything goes wrong
3131          * from here on out we know not to trust this cache when we load up next
3132          * time.
3133          */
3134         BTRFS_I(inode)->generation = 0;
3135         ret = btrfs_update_inode(trans, root, inode);
3136         WARN_ON(ret);
3137
3138         if (i_size_read(inode) > 0) {
3139                 ret = btrfs_truncate_free_space_cache(root, trans, path,
3140                                                       inode);
3141                 if (ret)
3142                         goto out_put;
3143         }
3144
3145         spin_lock(&block_group->lock);
3146         if (block_group->cached != BTRFS_CACHE_FINISHED ||
3147             !btrfs_test_opt(root, SPACE_CACHE)) {
3148                 /*
3149                  * don't bother trying to write stuff out _if_
3150                  * a) we're not cached,
3151                  * b) we're with nospace_cache mount option.
3152                  */
3153                 dcs = BTRFS_DC_WRITTEN;
3154                 spin_unlock(&block_group->lock);
3155                 goto out_put;
3156         }
3157         spin_unlock(&block_group->lock);
3158
3159         /*
3160          * Try to preallocate enough space based on how big the block group is.
3161          * Keep in mind this has to include any pinned space which could end up
3162          * taking up quite a bit since it's not folded into the other space
3163          * cache.
3164          */
3165         num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3166         if (!num_pages)
3167                 num_pages = 1;
3168
3169         num_pages *= 16;
3170         num_pages *= PAGE_CACHE_SIZE;
3171
3172         ret = btrfs_check_data_free_space(inode, num_pages);
3173         if (ret)
3174                 goto out_put;
3175
3176         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3177                                               num_pages, num_pages,
3178                                               &alloc_hint);
3179         if (!ret)
3180                 dcs = BTRFS_DC_SETUP;
3181         btrfs_free_reserved_data_space(inode, num_pages);
3182
3183 out_put:
3184         iput(inode);
3185 out_free:
3186         btrfs_release_path(path);
3187 out:
3188         spin_lock(&block_group->lock);
3189         if (!ret && dcs == BTRFS_DC_SETUP)
3190                 block_group->cache_generation = trans->transid;
3191         block_group->disk_cache_state = dcs;
3192         spin_unlock(&block_group->lock);
3193
3194         return ret;
3195 }
3196
3197 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3198                                    struct btrfs_root *root)
3199 {
3200         struct btrfs_block_group_cache *cache;
3201         int err = 0;
3202         struct btrfs_path *path;
3203         u64 last = 0;
3204
3205         path = btrfs_alloc_path();
3206         if (!path)
3207                 return -ENOMEM;
3208
3209 again:
3210         while (1) {
3211                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3212                 while (cache) {
3213                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3214                                 break;
3215                         cache = next_block_group(root, cache);
3216                 }
3217                 if (!cache) {
3218                         if (last == 0)
3219                                 break;
3220                         last = 0;
3221                         continue;
3222                 }
3223                 err = cache_save_setup(cache, trans, path);
3224                 last = cache->key.objectid + cache->key.offset;
3225                 btrfs_put_block_group(cache);
3226         }
3227
3228         while (1) {
3229                 if (last == 0) {
3230                         err = btrfs_run_delayed_refs(trans, root,
3231                                                      (unsigned long)-1);
3232                         if (err) /* File system offline */
3233                                 goto out;
3234                 }
3235
3236                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3237                 while (cache) {
3238                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3239                                 btrfs_put_block_group(cache);
3240                                 goto again;
3241                         }
3242
3243                         if (cache->dirty)
3244                                 break;
3245                         cache = next_block_group(root, cache);
3246                 }
3247                 if (!cache) {
3248                         if (last == 0)
3249                                 break;
3250                         last = 0;
3251                         continue;
3252                 }
3253
3254                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3255                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3256                 cache->dirty = 0;
3257                 last = cache->key.objectid + cache->key.offset;
3258
3259                 err = write_one_cache_group(trans, root, path, cache);
3260                 if (err) /* File system offline */
3261                         goto out;
3262
3263                 btrfs_put_block_group(cache);
3264         }
3265
3266         while (1) {
3267                 /*
3268                  * I don't think this is needed since we're just marking our
3269                  * preallocated extent as written, but just in case it can't
3270                  * hurt.
3271                  */
3272                 if (last == 0) {
3273                         err = btrfs_run_delayed_refs(trans, root,
3274                                                      (unsigned long)-1);
3275                         if (err) /* File system offline */
3276                                 goto out;
3277                 }
3278
3279                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3280                 while (cache) {
3281                         /*
3282                          * Really this shouldn't happen, but it could if we
3283                          * couldn't write the entire preallocated extent and
3284                          * splitting the extent resulted in a new block.
3285                          */
3286                         if (cache->dirty) {
3287                                 btrfs_put_block_group(cache);
3288                                 goto again;
3289                         }
3290                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
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
3301                 err = btrfs_write_out_cache(root, trans, cache, path);
3302
3303                 /*
3304                  * If we didn't have an error then the cache state is still
3305                  * NEED_WRITE, so we can set it to WRITTEN.
3306                  */
3307                 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3308                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
3309                 last = cache->key.objectid + cache->key.offset;
3310                 btrfs_put_block_group(cache);
3311         }
3312 out:
3313
3314         btrfs_free_path(path);
3315         return err;
3316 }
3317
3318 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3319 {
3320         struct btrfs_block_group_cache *block_group;
3321         int readonly = 0;
3322
3323         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3324         if (!block_group || block_group->ro)
3325                 readonly = 1;
3326         if (block_group)
3327                 btrfs_put_block_group(block_group);
3328         return readonly;
3329 }
3330
3331 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3332                              u64 total_bytes, u64 bytes_used,
3333                              struct btrfs_space_info **space_info)
3334 {
3335         struct btrfs_space_info *found;
3336         int i;
3337         int factor;
3338
3339         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3340                      BTRFS_BLOCK_GROUP_RAID10))
3341                 factor = 2;
3342         else
3343                 factor = 1;
3344
3345         found = __find_space_info(info, flags);
3346         if (found) {
3347                 spin_lock(&found->lock);
3348                 found->total_bytes += total_bytes;
3349                 found->disk_total += total_bytes * factor;
3350                 found->bytes_used += bytes_used;
3351                 found->disk_used += bytes_used * factor;
3352                 found->full = 0;
3353                 spin_unlock(&found->lock);
3354                 *space_info = found;
3355                 return 0;
3356         }
3357         found = kzalloc(sizeof(*found), GFP_NOFS);
3358         if (!found)
3359                 return -ENOMEM;
3360
3361         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3362                 INIT_LIST_HEAD(&found->block_groups[i]);
3363         init_rwsem(&found->groups_sem);
3364         spin_lock_init(&found->lock);
3365         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3366         found->total_bytes = total_bytes;
3367         found->disk_total = total_bytes * factor;
3368         found->bytes_used = bytes_used;
3369         found->disk_used = bytes_used * factor;
3370         found->bytes_pinned = 0;
3371         found->bytes_reserved = 0;
3372         found->bytes_readonly = 0;
3373         found->bytes_may_use = 0;
3374         found->full = 0;
3375         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3376         found->chunk_alloc = 0;
3377         found->flush = 0;
3378         init_waitqueue_head(&found->wait);
3379         *space_info = found;
3380         list_add_rcu(&found->list, &info->space_info);
3381         if (flags & BTRFS_BLOCK_GROUP_DATA)
3382                 info->data_sinfo = found;
3383         return 0;
3384 }
3385
3386 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3387 {
3388         u64 extra_flags = chunk_to_extended(flags) &
3389                                 BTRFS_EXTENDED_PROFILE_MASK;
3390
3391         write_seqlock(&fs_info->profiles_lock);
3392         if (flags & BTRFS_BLOCK_GROUP_DATA)
3393                 fs_info->avail_data_alloc_bits |= extra_flags;
3394         if (flags & BTRFS_BLOCK_GROUP_METADATA)
3395                 fs_info->avail_metadata_alloc_bits |= extra_flags;
3396         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3397                 fs_info->avail_system_alloc_bits |= extra_flags;
3398         write_sequnlock(&fs_info->profiles_lock);
3399 }
3400
3401 /*
3402  * returns target flags in extended format or 0 if restripe for this
3403  * chunk_type is not in progress
3404  *
3405  * should be called with either volume_mutex or balance_lock held
3406  */
3407 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3408 {
3409         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3410         u64 target = 0;
3411
3412         if (!bctl)
3413                 return 0;
3414
3415         if (flags & BTRFS_BLOCK_GROUP_DATA &&
3416             bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3417                 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3418         } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3419                    bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3420                 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3421         } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3422                    bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3423                 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3424         }
3425
3426         return target;
3427 }
3428
3429 /*
3430  * @flags: available profiles in extended format (see ctree.h)
3431  *
3432  * Returns reduced profile in chunk format.  If profile changing is in
3433  * progress (either running or paused) picks the target profile (if it's
3434  * already available), otherwise falls back to plain reducing.
3435  */
3436 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3437 {
3438         /*
3439          * we add in the count of missing devices because we want
3440          * to make sure that any RAID levels on a degraded FS
3441          * continue to be honored.
3442          */
3443         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3444                 root->fs_info->fs_devices->missing_devices;
3445         u64 target;
3446         u64 tmp;
3447
3448         /*
3449          * see if restripe for this chunk_type is in progress, if so
3450          * try to reduce to the target profile
3451          */
3452         spin_lock(&root->fs_info->balance_lock);
3453         target = get_restripe_target(root->fs_info, flags);
3454         if (target) {
3455                 /* pick target profile only if it's already available */
3456                 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3457                         spin_unlock(&root->fs_info->balance_lock);
3458                         return extended_to_chunk(target);
3459                 }
3460         }
3461         spin_unlock(&root->fs_info->balance_lock);
3462
3463         /* First, mask out the RAID levels which aren't possible */
3464         if (num_devices == 1)
3465                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3466                            BTRFS_BLOCK_GROUP_RAID5);
3467         if (num_devices < 3)
3468                 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3469         if (num_devices < 4)
3470                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3471
3472         tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3473                        BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3474                        BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3475         flags &= ~tmp;
3476
3477         if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3478                 tmp = BTRFS_BLOCK_GROUP_RAID6;
3479         else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3480                 tmp = BTRFS_BLOCK_GROUP_RAID5;
3481         else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3482                 tmp = BTRFS_BLOCK_GROUP_RAID10;
3483         else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3484                 tmp = BTRFS_BLOCK_GROUP_RAID1;
3485         else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3486                 tmp = BTRFS_BLOCK_GROUP_RAID0;
3487
3488         return extended_to_chunk(flags | tmp);
3489 }
3490
3491 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3492 {
3493         unsigned seq;
3494
3495         do {
3496                 seq = read_seqbegin(&root->fs_info->profiles_lock);
3497
3498                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3499                         flags |= root->fs_info->avail_data_alloc_bits;
3500                 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3501                         flags |= root->fs_info->avail_system_alloc_bits;
3502                 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3503                         flags |= root->fs_info->avail_metadata_alloc_bits;
3504         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3505
3506         return btrfs_reduce_alloc_profile(root, flags);
3507 }
3508
3509 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3510 {
3511         u64 flags;
3512         u64 ret;
3513
3514         if (data)
3515                 flags = BTRFS_BLOCK_GROUP_DATA;
3516         else if (root == root->fs_info->chunk_root)
3517                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3518         else
3519                 flags = BTRFS_BLOCK_GROUP_METADATA;
3520
3521         ret = get_alloc_profile(root, flags);
3522         return ret;
3523 }
3524
3525 /*
3526  * This will check the space that the inode allocates from to make sure we have
3527  * enough space for bytes.
3528  */
3529 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3530 {
3531         struct btrfs_space_info *data_sinfo;
3532         struct btrfs_root *root = BTRFS_I(inode)->root;
3533         struct btrfs_fs_info *fs_info = root->fs_info;
3534         u64 used;
3535         int ret = 0, committed = 0, alloc_chunk = 1;
3536
3537         /* make sure bytes are sectorsize aligned */
3538         bytes = ALIGN(bytes, root->sectorsize);
3539
3540         if (root == root->fs_info->tree_root ||
3541             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3542                 alloc_chunk = 0;
3543                 committed = 1;
3544         }
3545
3546         data_sinfo = fs_info->data_sinfo;
3547         if (!data_sinfo)
3548                 goto alloc;
3549
3550 again:
3551         /* make sure we have enough space to handle the data first */
3552         spin_lock(&data_sinfo->lock);
3553         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3554                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3555                 data_sinfo->bytes_may_use;
3556
3557         if (used + bytes > data_sinfo->total_bytes) {
3558                 struct btrfs_trans_handle *trans;
3559
3560                 /*
3561                  * if we don't have enough free bytes in this space then we need
3562                  * to alloc a new chunk.
3563                  */
3564                 if (!data_sinfo->full && alloc_chunk) {
3565                         u64 alloc_target;
3566
3567                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3568                         spin_unlock(&data_sinfo->lock);
3569 alloc:
3570                         alloc_target = btrfs_get_alloc_profile(root, 1);
3571                         trans = btrfs_join_transaction(root);
3572                         if (IS_ERR(trans))
3573                                 return PTR_ERR(trans);
3574
3575                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3576                                              alloc_target,
3577                                              CHUNK_ALLOC_NO_FORCE);
3578                         btrfs_end_transaction(trans, root);
3579                         if (ret < 0) {
3580                                 if (ret != -ENOSPC)
3581                                         return ret;
3582                                 else
3583                                         goto commit_trans;
3584                         }
3585
3586                         if (!data_sinfo)
3587                                 data_sinfo = fs_info->data_sinfo;
3588
3589                         goto again;
3590                 }
3591
3592                 /*
3593                  * If we have less pinned bytes than we want to allocate then
3594                  * don't bother committing the transaction, it won't help us.
3595                  */
3596                 if (data_sinfo->bytes_pinned < bytes)
3597                         committed = 1;
3598                 spin_unlock(&data_sinfo->lock);
3599
3600                 /* commit the current transaction and try again */
3601 commit_trans:
3602                 if (!committed &&
3603                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3604                         committed = 1;
3605                         trans = btrfs_join_transaction(root);
3606                         if (IS_ERR(trans))
3607                                 return PTR_ERR(trans);
3608                         ret = btrfs_commit_transaction(trans, root);
3609                         if (ret)
3610                                 return ret;
3611                         goto again;
3612                 }
3613
3614                 return -ENOSPC;
3615         }
3616         data_sinfo->bytes_may_use += bytes;
3617         trace_btrfs_space_reservation(root->fs_info, "space_info",
3618                                       data_sinfo->flags, bytes, 1);
3619         spin_unlock(&data_sinfo->lock);
3620
3621         return 0;
3622 }
3623
3624 /*
3625  * Called if we need to clear a data reservation for this inode.
3626  */
3627 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3628 {
3629         struct btrfs_root *root = BTRFS_I(inode)->root;
3630         struct btrfs_space_info *data_sinfo;
3631
3632         /* make sure bytes are sectorsize aligned */
3633         bytes = ALIGN(bytes, root->sectorsize);
3634
3635         data_sinfo = root->fs_info->data_sinfo;
3636         spin_lock(&data_sinfo->lock);
3637         data_sinfo->bytes_may_use -= bytes;
3638         trace_btrfs_space_reservation(root->fs_info, "space_info",
3639                                       data_sinfo->flags, bytes, 0);
3640         spin_unlock(&data_sinfo->lock);
3641 }
3642
3643 static void force_metadata_allocation(struct btrfs_fs_info *info)
3644 {
3645         struct list_head *head = &info->space_info;
3646         struct btrfs_space_info *found;
3647
3648         rcu_read_lock();
3649         list_for_each_entry_rcu(found, head, list) {
3650                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3651                         found->force_alloc = CHUNK_ALLOC_FORCE;
3652         }
3653         rcu_read_unlock();
3654 }
3655
3656 static int should_alloc_chunk(struct btrfs_root *root,
3657                               struct btrfs_space_info *sinfo, int force)
3658 {
3659         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3660         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3661         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3662         u64 thresh;
3663
3664         if (force == CHUNK_ALLOC_FORCE)
3665                 return 1;
3666
3667         /*
3668          * We need to take into account the global rsv because for all intents
3669          * and purposes it's used space.  Don't worry about locking the
3670          * global_rsv, it doesn't change except when the transaction commits.
3671          */
3672         if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3673                 num_allocated += global_rsv->size;
3674
3675         /*
3676          * in limited mode, we want to have some free space up to
3677          * about 1% of the FS size.
3678          */
3679         if (force == CHUNK_ALLOC_LIMITED) {
3680                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3681                 thresh = max_t(u64, 64 * 1024 * 1024,
3682                                div_factor_fine(thresh, 1));
3683
3684                 if (num_bytes - num_allocated < thresh)
3685                         return 1;
3686         }
3687
3688         if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3689                 return 0;
3690         return 1;
3691 }
3692
3693 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3694 {
3695         u64 num_dev;
3696
3697         if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3698                     BTRFS_BLOCK_GROUP_RAID0 |
3699                     BTRFS_BLOCK_GROUP_RAID5 |
3700                     BTRFS_BLOCK_GROUP_RAID6))
3701                 num_dev = root->fs_info->fs_devices->rw_devices;
3702         else if (type & BTRFS_BLOCK_GROUP_RAID1)
3703                 num_dev = 2;
3704         else
3705                 num_dev = 1;    /* DUP or single */
3706
3707         /* metadata for updaing devices and chunk tree */
3708         return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3709 }
3710
3711 static void check_system_chunk(struct btrfs_trans_handle *trans,
3712                                struct btrfs_root *root, u64 type)
3713 {
3714         struct btrfs_space_info *info;
3715         u64 left;
3716         u64 thresh;
3717
3718         info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3719         spin_lock(&info->lock);
3720         left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3721                 info->bytes_reserved - info->bytes_readonly;
3722         spin_unlock(&info->lock);
3723
3724         thresh = get_system_chunk_thresh(root, type);
3725         if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3726                 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3727                        left, thresh, type);
3728                 dump_space_info(info, 0, 0);
3729         }
3730
3731         if (left < thresh) {
3732                 u64 flags;
3733
3734                 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3735                 btrfs_alloc_chunk(trans, root, flags);
3736         }
3737 }
3738
3739 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3740                           struct btrfs_root *extent_root, u64 flags, int force)
3741 {
3742         struct btrfs_space_info *space_info;
3743         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3744         int wait_for_alloc = 0;
3745         int ret = 0;
3746
3747         /* Don't re-enter if we're already allocating a chunk */
3748         if (trans->allocating_chunk)
3749                 return -ENOSPC;
3750
3751         space_info = __find_space_info(extent_root->fs_info, flags);
3752         if (!space_info) {
3753                 ret = update_space_info(extent_root->fs_info, flags,
3754                                         0, 0, &space_info);
3755                 BUG_ON(ret); /* -ENOMEM */
3756         }
3757         BUG_ON(!space_info); /* Logic error */
3758
3759 again:
3760         spin_lock(&space_info->lock);
3761         if (force < space_info->force_alloc)
3762                 force = space_info->force_alloc;
3763         if (space_info->full) {
3764                 spin_unlock(&space_info->lock);
3765                 return 0;
3766         }
3767
3768         if (!should_alloc_chunk(extent_root, space_info, force)) {
3769                 spin_unlock(&space_info->lock);
3770                 return 0;
3771         } else if (space_info->chunk_alloc) {
3772                 wait_for_alloc = 1;
3773         } else {
3774                 space_info->chunk_alloc = 1;
3775         }
3776
3777         spin_unlock(&space_info->lock);
3778
3779         mutex_lock(&fs_info->chunk_mutex);
3780
3781         /*
3782          * The chunk_mutex is held throughout the entirety of a chunk
3783          * allocation, so once we've acquired the chunk_mutex we know that the
3784          * other guy is done and we need to recheck and see if we should
3785          * allocate.
3786          */
3787         if (wait_for_alloc) {
3788                 mutex_unlock(&fs_info->chunk_mutex);
3789                 wait_for_alloc = 0;
3790                 goto again;
3791         }
3792
3793         trans->allocating_chunk = true;
3794
3795         /*
3796          * If we have mixed data/metadata chunks we want to make sure we keep
3797          * allocating mixed chunks instead of individual chunks.
3798          */
3799         if (btrfs_mixed_space_info(space_info))
3800                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3801
3802         /*
3803          * if we're doing a data chunk, go ahead and make sure that
3804          * we keep a reasonable number of metadata chunks allocated in the
3805          * FS as well.
3806          */
3807         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3808                 fs_info->data_chunk_allocations++;
3809                 if (!(fs_info->data_chunk_allocations %
3810                       fs_info->metadata_ratio))
3811                         force_metadata_allocation(fs_info);
3812         }
3813
3814         /*
3815          * Check if we have enough space in SYSTEM chunk because we may need
3816          * to update devices.
3817          */
3818         check_system_chunk(trans, extent_root, flags);
3819
3820         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3821         trans->allocating_chunk = false;
3822
3823         spin_lock(&space_info->lock);
3824         if (ret < 0 && ret != -ENOSPC)
3825                 goto out;
3826         if (ret)
3827                 space_info->full = 1;
3828         else
3829                 ret = 1;
3830
3831         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3832 out:
3833         space_info->chunk_alloc = 0;
3834         spin_unlock(&space_info->lock);
3835         mutex_unlock(&fs_info->chunk_mutex);
3836         return ret;
3837 }
3838
3839 static int can_overcommit(struct btrfs_root *root,
3840                           struct btrfs_space_info *space_info, u64 bytes,
3841                           enum btrfs_reserve_flush_enum flush)
3842 {
3843         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3844         u64 profile = btrfs_get_alloc_profile(root, 0);
3845         u64 rsv_size = 0;
3846         u64 avail;
3847         u64 used;
3848         u64 to_add;
3849
3850         used = space_info->bytes_used + space_info->bytes_reserved +
3851                 space_info->bytes_pinned + space_info->bytes_readonly;
3852
3853         spin_lock(&global_rsv->lock);
3854         rsv_size = global_rsv->size;
3855         spin_unlock(&global_rsv->lock);
3856
3857         /*
3858          * We only want to allow over committing if we have lots of actual space
3859          * free, but if we don't have enough space to handle the global reserve
3860          * space then we could end up having a real enospc problem when trying
3861          * to allocate a chunk or some other such important allocation.
3862          */
3863         rsv_size <<= 1;
3864         if (used + rsv_size >= space_info->total_bytes)
3865                 return 0;
3866
3867         used += space_info->bytes_may_use;
3868
3869         spin_lock(&root->fs_info->free_chunk_lock);
3870         avail = root->fs_info->free_chunk_space;
3871         spin_unlock(&root->fs_info->free_chunk_lock);
3872
3873         /*
3874          * If we have dup, raid1 or raid10 then only half of the free
3875          * space is actually useable.  For raid56, the space info used
3876          * doesn't include the parity drive, so we don't have to
3877          * change the math
3878          */
3879         if (profile & (BTRFS_BLOCK_GROUP_DUP |
3880                        BTRFS_BLOCK_GROUP_RAID1 |
3881                        BTRFS_BLOCK_GROUP_RAID10))
3882                 avail >>= 1;
3883
3884         to_add = space_info->total_bytes;
3885
3886         /*
3887          * If we aren't flushing all things, let us overcommit up to
3888          * 1/2th of the space. If we can flush, don't let us overcommit
3889          * too much, let it overcommit up to 1/8 of the space.
3890          */
3891         if (flush == BTRFS_RESERVE_FLUSH_ALL)
3892                 to_add >>= 3;
3893         else
3894                 to_add >>= 1;
3895
3896         /*
3897          * Limit the overcommit to the amount of free space we could possibly
3898          * allocate for chunks.
3899          */
3900         to_add = min(avail, to_add);
3901
3902         if (used + bytes < space_info->total_bytes + to_add)
3903                 return 1;
3904         return 0;
3905 }
3906
3907 void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3908                                   unsigned long nr_pages)
3909 {
3910         struct super_block *sb = root->fs_info->sb;
3911         int started;
3912
3913         /* If we can not start writeback, just sync all the delalloc file. */
3914         started = try_to_writeback_inodes_sb_nr(sb, nr_pages,
3915                                                       WB_REASON_FS_FREE_SPACE);
3916         if (!started) {
3917                 /*
3918                  * We needn't worry the filesystem going from r/w to r/o though
3919                  * we don't acquire ->s_umount mutex, because the filesystem
3920                  * should guarantee the delalloc inodes list be empty after
3921                  * the filesystem is readonly(all dirty pages are written to
3922                  * the disk).
3923                  */
3924                 btrfs_start_delalloc_inodes(root, 0);
3925                 btrfs_wait_ordered_extents(root, 0);
3926         }
3927 }
3928
3929 /*
3930  * shrink metadata reservation for delalloc
3931  */
3932 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3933                             bool wait_ordered)
3934 {
3935         struct btrfs_block_rsv *block_rsv;
3936         struct btrfs_space_info *space_info;
3937         struct btrfs_trans_handle *trans;
3938         u64 delalloc_bytes;
3939         u64 max_reclaim;
3940         long time_left;
3941         unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3942         int loops = 0;
3943         enum btrfs_reserve_flush_enum flush;
3944
3945         trans = (struct btrfs_trans_handle *)current->journal_info;
3946         block_rsv = &root->fs_info->delalloc_block_rsv;
3947         space_info = block_rsv->space_info;
3948
3949         smp_mb();
3950         delalloc_bytes = percpu_counter_sum_positive(
3951                                                 &root->fs_info->delalloc_bytes);
3952         if (delalloc_bytes == 0) {
3953                 if (trans)
3954                         return;
3955                 btrfs_wait_ordered_extents(root, 0);
3956                 return;
3957         }
3958
3959         while (delalloc_bytes && loops < 3) {
3960                 max_reclaim = min(delalloc_bytes, to_reclaim);
3961                 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3962                 btrfs_writeback_inodes_sb_nr(root, nr_pages);
3963                 /*
3964                  * We need to wait for the async pages to actually start before
3965                  * we do anything.
3966                  */
3967                 wait_event(root->fs_info->async_submit_wait,
3968                            !atomic_read(&root->fs_info->async_delalloc_pages));
3969
3970                 if (!trans)
3971                         flush = BTRFS_RESERVE_FLUSH_ALL;
3972                 else
3973                         flush = BTRFS_RESERVE_NO_FLUSH;
3974                 spin_lock(&space_info->lock);
3975                 if (can_overcommit(root, space_info, orig, flush)) {
3976                         spin_unlock(&space_info->lock);
3977                         break;
3978                 }
3979                 spin_unlock(&space_info->lock);
3980
3981                 loops++;
3982                 if (wait_ordered && !trans) {
3983                         btrfs_wait_ordered_extents(root, 0);
3984                 } else {
3985                         time_left = schedule_timeout_killable(1);
3986                         if (time_left)
3987                                 break;
3988                 }
3989                 smp_mb();
3990                 delalloc_bytes = percpu_counter_sum_positive(
3991                                                 &root->fs_info->delalloc_bytes);
3992         }
3993 }
3994
3995 /**
3996  * maybe_commit_transaction - possibly commit the transaction if its ok to
3997  * @root - the root we're allocating for
3998  * @bytes - the number of bytes we want to reserve
3999  * @force - force the commit
4000  *
4001  * This will check to make sure that committing the transaction will actually
4002  * get us somewhere and then commit the transaction if it does.  Otherwise it
4003  * will return -ENOSPC.
4004  */
4005 static int may_commit_transaction(struct btrfs_root *root,
4006                                   struct btrfs_space_info *space_info,
4007                                   u64 bytes, int force)
4008 {
4009         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4010         struct btrfs_trans_handle *trans;
4011
4012         trans = (struct btrfs_trans_handle *)current->journal_info;
4013         if (trans)
4014                 return -EAGAIN;
4015
4016         if (force)
4017                 goto commit;
4018
4019         /* See if there is enough pinned space to make this reservation */
4020         spin_lock(&space_info->lock);
4021         if (space_info->bytes_pinned >= bytes) {
4022                 spin_unlock(&space_info->lock);
4023                 goto commit;
4024         }
4025         spin_unlock(&space_info->lock);
4026
4027         /*
4028          * See if there is some space in the delayed insertion reservation for
4029          * this reservation.
4030          */
4031         if (space_info != delayed_rsv->space_info)
4032                 return -ENOSPC;
4033
4034         spin_lock(&space_info->lock);
4035         spin_lock(&delayed_rsv->lock);
4036         if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
4037                 spin_unlock(&delayed_rsv->lock);
4038                 spin_unlock(&space_info->lock);
4039                 return -ENOSPC;
4040         }
4041         spin_unlock(&delayed_rsv->lock);
4042         spin_unlock(&space_info->lock);
4043
4044 commit:
4045         trans = btrfs_join_transaction(root);
4046         if (IS_ERR(trans))
4047                 return -ENOSPC;
4048
4049         return btrfs_commit_transaction(trans, root);
4050 }
4051
4052 enum flush_state {
4053         FLUSH_DELAYED_ITEMS_NR  =       1,
4054         FLUSH_DELAYED_ITEMS     =       2,
4055         FLUSH_DELALLOC          =       3,
4056         FLUSH_DELALLOC_WAIT     =       4,
4057         ALLOC_CHUNK             =       5,
4058         COMMIT_TRANS            =       6,
4059 };
4060
4061 static int flush_space(struct btrfs_root *root,
4062                        struct btrfs_space_info *space_info, u64 num_bytes,
4063                        u64 orig_bytes, int state)
4064 {
4065         struct btrfs_trans_handle *trans;
4066         int nr;
4067         int ret = 0;
4068
4069         switch (state) {
4070         case FLUSH_DELAYED_ITEMS_NR:
4071         case FLUSH_DELAYED_ITEMS:
4072                 if (state == FLUSH_DELAYED_ITEMS_NR) {
4073                         u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4074
4075                         nr = (int)div64_u64(num_bytes, bytes);
4076                         if (!nr)
4077                                 nr = 1;
4078                         nr *= 2;
4079                 } else {
4080                         nr = -1;
4081                 }
4082                 trans = btrfs_join_transaction(root);
4083                 if (IS_ERR(trans)) {
4084                         ret = PTR_ERR(trans);
4085                         break;
4086                 }
4087                 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4088                 btrfs_end_transaction(trans, root);
4089                 break;
4090         case FLUSH_DELALLOC:
4091         case FLUSH_DELALLOC_WAIT:
4092                 shrink_delalloc(root, num_bytes, orig_bytes,
4093                                 state == FLUSH_DELALLOC_WAIT);
4094                 break;
4095         case ALLOC_CHUNK:
4096                 trans = btrfs_join_transaction(root);
4097                 if (IS_ERR(trans)) {
4098                         ret = PTR_ERR(trans);
4099                         break;
4100                 }
4101                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4102                                      btrfs_get_alloc_profile(root, 0),
4103                                      CHUNK_ALLOC_NO_FORCE);
4104                 btrfs_end_transaction(trans, root);
4105                 if (ret == -ENOSPC)
4106                         ret = 0;
4107                 break;
4108         case COMMIT_TRANS:
4109                 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4110                 break;
4111         default:
4112                 ret = -ENOSPC;
4113                 break;
4114         }
4115
4116         return ret;
4117 }
4118 /**
4119  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4120  * @root - the root we're allocating for
4121  * @block_rsv - the block_rsv we're allocating for
4122  * @orig_bytes - the number of bytes we want
4123  * @flush - whether or not we can flush to make our reservation
4124  *
4125  * This will reserve orgi_bytes number of bytes from the space info associated
4126  * with the block_rsv.  If there is not enough space it will make an attempt to
4127  * flush out space to make room.  It will do this by flushing delalloc if
4128  * possible or committing the transaction.  If flush is 0 then no attempts to
4129  * regain reservations will be made and this will fail if there is not enough
4130  * space already.
4131  */
4132 static int reserve_metadata_bytes(struct btrfs_root *root,
4133                                   struct btrfs_block_rsv *block_rsv,
4134                                   u64 orig_bytes,
4135                                   enum btrfs_reserve_flush_enum flush)
4136 {
4137         struct btrfs_space_info *space_info = block_rsv->space_info;
4138         u64 used;
4139         u64 num_bytes = orig_bytes;
4140         int flush_state = FLUSH_DELAYED_ITEMS_NR;
4141         int ret = 0;
4142         bool flushing = false;
4143
4144 again:
4145         ret = 0;
4146         spin_lock(&space_info->lock);
4147         /*
4148          * We only want to wait if somebody other than us is flushing and we
4149          * are actually allowed to flush all things.
4150          */
4151         while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4152                space_info->flush) {
4153                 spin_unlock(&space_info->lock);
4154                 /*
4155                  * If we have a trans handle we can't wait because the flusher
4156                  * may have to commit the transaction, which would mean we would
4157                  * deadlock since we are waiting for the flusher to finish, but
4158                  * hold the current transaction open.
4159                  */
4160                 if (current->journal_info)
4161                         return -EAGAIN;
4162                 ret = wait_event_killable(space_info->wait, !space_info->flush);
4163                 /* Must have been killed, return */
4164                 if (ret)
4165                         return -EINTR;
4166
4167                 spin_lock(&space_info->lock);
4168         }
4169
4170         ret = -ENOSPC;
4171         used = space_info->bytes_used + space_info->bytes_reserved +
4172                 space_info->bytes_pinned + space_info->bytes_readonly +
4173                 space_info->bytes_may_use;
4174
4175         /*
4176          * The idea here is that we've not already over-reserved the block group
4177          * then we can go ahead and save our reservation first and then start
4178          * flushing if we need to.  Otherwise if we've already overcommitted
4179          * lets start flushing stuff first and then come back and try to make
4180          * our reservation.
4181          */
4182         if (used <= space_info->total_bytes) {
4183                 if (used + orig_bytes <= space_info->total_bytes) {
4184                         space_info->bytes_may_use += orig_bytes;
4185                         trace_btrfs_space_reservation(root->fs_info,
4186                                 "space_info", space_info->flags, orig_bytes, 1);
4187                         ret = 0;
4188                 } else {
4189                         /*
4190                          * Ok set num_bytes to orig_bytes since we aren't
4191                          * overocmmitted, this way we only try and reclaim what
4192                          * we need.
4193                          */
4194                         num_bytes = orig_bytes;
4195                 }
4196         } else {
4197                 /*
4198                  * Ok we're over committed, set num_bytes to the overcommitted
4199                  * amount plus the amount of bytes that we need for this
4200                  * reservation.
4201                  */
4202                 num_bytes = used - space_info->total_bytes +
4203                         (orig_bytes * 2);
4204         }
4205
4206         if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4207                 space_info->bytes_may_use += orig_bytes;
4208                 trace_btrfs_space_reservation(root->fs_info, "space_info",
4209                                               space_info->flags, orig_bytes,
4210                                               1);
4211                 ret = 0;
4212         }
4213
4214         /*
4215          * Couldn't make our reservation, save our place so while we're trying
4216          * to reclaim space we can actually use it instead of somebody else
4217          * stealing it from us.
4218          *
4219          * We make the other tasks wait for the flush only when we can flush
4220          * all things.
4221          */
4222         if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4223                 flushing = true;
4224                 space_info->flush = 1;
4225         }
4226
4227         spin_unlock(&space_info->lock);
4228
4229         if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4230                 goto out;
4231
4232         ret = flush_space(root, space_info, num_bytes, orig_bytes,
4233                           flush_state);
4234         flush_state++;
4235
4236         /*
4237          * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4238          * would happen. So skip delalloc flush.
4239          */
4240         if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4241             (flush_state == FLUSH_DELALLOC ||
4242              flush_state == FLUSH_DELALLOC_WAIT))
4243                 flush_state = ALLOC_CHUNK;
4244
4245         if (!ret)
4246                 goto again;
4247         else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4248                  flush_state < COMMIT_TRANS)
4249                 goto again;
4250         else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4251                  flush_state <= COMMIT_TRANS)
4252                 goto again;
4253
4254 out:
4255         if (ret == -ENOSPC &&
4256             unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4257                 struct btrfs_block_rsv *global_rsv =
4258                         &root->fs_info->global_block_rsv;
4259
4260                 if (block_rsv != global_rsv &&
4261                     !block_rsv_use_bytes(global_rsv, orig_bytes))
4262                         ret = 0;
4263         }
4264         if (flushing) {
4265                 spin_lock(&space_info->lock);
4266                 space_info->flush = 0;
4267                 wake_up_all(&space_info->wait);
4268                 spin_unlock(&space_info->lock);
4269         }
4270         return ret;
4271 }
4272
4273 static struct btrfs_block_rsv *get_block_rsv(
4274                                         const struct btrfs_trans_handle *trans,
4275                                         const struct btrfs_root *root)
4276 {
4277         struct btrfs_block_rsv *block_rsv = NULL;
4278
4279         if (root->ref_cows)
4280                 block_rsv = trans->block_rsv;
4281
4282         if (root == root->fs_info->csum_root && trans->adding_csums)
4283                 block_rsv = trans->block_rsv;
4284
4285         if (!block_rsv)
4286                 block_rsv = root->block_rsv;
4287
4288         if (!block_rsv)
4289                 block_rsv = &root->fs_info->empty_block_rsv;
4290
4291         return block_rsv;
4292 }
4293
4294 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4295                                u64 num_bytes)
4296 {
4297         int ret = -ENOSPC;
4298         spin_lock(&block_rsv->lock);
4299         if (block_rsv->reserved >= num_bytes) {
4300                 block_rsv->reserved -= num_bytes;
4301                 if (block_rsv->reserved < block_rsv->size)
4302                         block_rsv->full = 0;
4303                 ret = 0;
4304         }
4305         spin_unlock(&block_rsv->lock);
4306         return ret;
4307 }
4308
4309 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4310                                 u64 num_bytes, int update_size)
4311 {
4312         spin_lock(&block_rsv->lock);
4313         block_rsv->reserved += num_bytes;
4314         if (update_size)
4315                 block_rsv->size += num_bytes;
4316         else if (block_rsv->reserved >= block_rsv->size)
4317                 block_rsv->full = 1;
4318         spin_unlock(&block_rsv->lock);
4319 }
4320
4321 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4322                                     struct btrfs_block_rsv *block_rsv,
4323                                     struct btrfs_block_rsv *dest, u64 num_bytes)
4324 {
4325         struct btrfs_space_info *space_info = block_rsv->space_info;
4326
4327         spin_lock(&block_rsv->lock);
4328         if (num_bytes == (u64)-1)
4329                 num_bytes = block_rsv->size;
4330         block_rsv->size -= num_bytes;
4331         if (block_rsv->reserved >= block_rsv->size) {
4332                 num_bytes = block_rsv->reserved - block_rsv->size;
4333                 block_rsv->reserved = block_rsv->size;
4334                 block_rsv->full = 1;
4335         } else {
4336                 num_bytes = 0;
4337         }
4338         spin_unlock(&block_rsv->lock);
4339
4340         if (num_bytes > 0) {
4341                 if (dest) {
4342                         spin_lock(&dest->lock);
4343                         if (!dest->full) {
4344                                 u64 bytes_to_add;
4345
4346                                 bytes_to_add = dest->size - dest->reserved;
4347                                 bytes_to_add = min(num_bytes, bytes_to_add);
4348                                 dest->reserved += bytes_to_add;
4349                                 if (dest->reserved >= dest->size)
4350                                         dest->full = 1;
4351                                 num_bytes -= bytes_to_add;
4352                         }
4353                         spin_unlock(&dest->lock);
4354                 }
4355                 if (num_bytes) {
4356                         spin_lock(&space_info->lock);
4357                         space_info->bytes_may_use -= num_bytes;
4358                         trace_btrfs_space_reservation(fs_info, "space_info",
4359                                         space_info->flags, num_bytes, 0);
4360                         space_info->reservation_progress++;
4361                         spin_unlock(&space_info->lock);
4362                 }
4363         }
4364 }
4365
4366 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4367                                    struct btrfs_block_rsv *dst, u64 num_bytes)
4368 {
4369         int ret;
4370
4371         ret = block_rsv_use_bytes(src, num_bytes);
4372         if (ret)
4373                 return ret;
4374
4375         block_rsv_add_bytes(dst, num_bytes, 1);
4376         return 0;
4377 }
4378
4379 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4380 {
4381         memset(rsv, 0, sizeof(*rsv));
4382         spin_lock_init(&rsv->lock);
4383         rsv->type = type;
4384 }
4385
4386 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4387                                               unsigned short type)
4388 {
4389         struct btrfs_block_rsv *block_rsv;
4390         struct btrfs_fs_info *fs_info = root->fs_info;
4391
4392         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4393         if (!block_rsv)
4394                 return NULL;
4395
4396         btrfs_init_block_rsv(block_rsv, type);
4397         block_rsv->space_info = __find_space_info(fs_info,
4398                                                   BTRFS_BLOCK_GROUP_METADATA);
4399         return block_rsv;
4400 }
4401
4402 void btrfs_free_block_rsv(struct btrfs_root *root,
4403                           struct btrfs_block_rsv *rsv)
4404 {
4405         if (!rsv)
4406                 return;
4407         btrfs_block_rsv_release(root, rsv, (u64)-1);
4408         kfree(rsv);
4409 }
4410
4411 int btrfs_block_rsv_add(struct btrfs_root *root,
4412                         struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4413                         enum btrfs_reserve_flush_enum flush)
4414 {
4415         int ret;
4416
4417         if (num_bytes == 0)
4418                 return 0;
4419
4420         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4421         if (!ret) {
4422                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4423                 return 0;
4424         }
4425
4426         return ret;
4427 }
4428
4429 int btrfs_block_rsv_check(struct btrfs_root *root,
4430                           struct btrfs_block_rsv *block_rsv, int min_factor)
4431 {
4432         u64 num_bytes = 0;
4433         int ret = -ENOSPC;
4434
4435         if (!block_rsv)
4436                 return 0;
4437
4438         spin_lock(&block_rsv->lock);
4439         num_bytes = div_factor(block_rsv->size, min_factor);
4440         if (block_rsv->reserved >= num_bytes)
4441                 ret = 0;
4442         spin_unlock(&block_rsv->lock);
4443
4444         return ret;
4445 }
4446
4447 int btrfs_block_rsv_refill(struct btrfs_root *root,
4448                            struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4449                            enum btrfs_reserve_flush_enum flush)
4450 {
4451         u64 num_bytes = 0;
4452         int ret = -ENOSPC;
4453
4454         if (!block_rsv)
4455                 return 0;
4456
4457         spin_lock(&block_rsv->lock);
4458         num_bytes = min_reserved;
4459         if (block_rsv->reserved >= num_bytes)
4460                 ret = 0;
4461         else
4462                 num_bytes -= block_rsv->reserved;
4463         spin_unlock(&block_rsv->lock);
4464
4465         if (!ret)
4466                 return 0;
4467
4468         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4469         if (!ret) {
4470                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4471                 return 0;
4472         }
4473
4474         return ret;
4475 }
4476
4477 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4478                             struct btrfs_block_rsv *dst_rsv,
4479                             u64 num_bytes)
4480 {
4481         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4482 }
4483
4484 void btrfs_block_rsv_release(struct btrfs_root *root,
4485                              struct btrfs_block_rsv *block_rsv,
4486                              u64 num_bytes)
4487 {
4488         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4489         if (global_rsv->full || global_rsv == block_rsv ||
4490             block_rsv->space_info != global_rsv->space_info)
4491                 global_rsv = NULL;
4492         block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4493                                 num_bytes);
4494 }
4495
4496 /*
4497  * helper to calculate size of global block reservation.
4498  * the desired value is sum of space used by extent tree,
4499  * checksum tree and root tree
4500  */
4501 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4502 {
4503         struct btrfs_space_info *sinfo;
4504         u64 num_bytes;
4505         u64 meta_used;
4506         u64 data_used;
4507         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4508
4509         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4510         spin_lock(&sinfo->lock);
4511         data_used = sinfo->bytes_used;
4512         spin_unlock(&sinfo->lock);
4513
4514         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4515         spin_lock(&sinfo->lock);
4516         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4517                 data_used = 0;
4518         meta_used = sinfo->bytes_used;
4519         spin_unlock(&sinfo->lock);
4520
4521         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4522                     csum_size * 2;
4523         num_bytes += div64_u64(data_used + meta_used, 50);
4524
4525         if (num_bytes * 3 > meta_used)
4526                 num_bytes = div64_u64(meta_used, 3);
4527
4528         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4529 }
4530
4531 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4532 {
4533         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4534         struct btrfs_space_info *sinfo = block_rsv->space_info;
4535         u64 num_bytes;
4536
4537         num_bytes = calc_global_metadata_size(fs_info);
4538
4539         spin_lock(&sinfo->lock);
4540         spin_lock(&block_rsv->lock);
4541
4542         block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4543
4544         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4545                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4546                     sinfo->bytes_may_use;
4547
4548         if (sinfo->total_bytes > num_bytes) {
4549                 num_bytes = sinfo->total_bytes - num_bytes;
4550                 block_rsv->reserved += num_bytes;
4551                 sinfo->bytes_may_use += num_bytes;
4552                 trace_btrfs_space_reservation(fs_info, "space_info",
4553                                       sinfo->flags, num_bytes, 1);
4554         }
4555
4556         if (block_rsv->reserved >= block_rsv->size) {
4557                 num_bytes = block_rsv->reserved - block_rsv->size;
4558                 sinfo->bytes_may_use -= num_bytes;
4559                 trace_btrfs_space_reservation(fs_info, "space_info",
4560                                       sinfo->flags, num_bytes, 0);
4561                 sinfo->reservation_progress++;
4562                 block_rsv->reserved = block_rsv->size;
4563                 block_rsv->full = 1;
4564         }
4565
4566         spin_unlock(&block_rsv->lock);
4567         spin_unlock(&sinfo->lock);
4568 }
4569
4570 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4571 {
4572         struct btrfs_space_info *space_info;
4573
4574         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4575         fs_info->chunk_block_rsv.space_info = space_info;
4576
4577         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4578         fs_info->global_block_rsv.space_info = space_info;
4579         fs_info->delalloc_block_rsv.space_info = space_info;
4580         fs_info->trans_block_rsv.space_info = space_info;
4581         fs_info->empty_block_rsv.space_info = space_info;
4582         fs_info->delayed_block_rsv.space_info = space_info;
4583
4584         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4585         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4586         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4587         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4588         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4589
4590         update_global_block_rsv(fs_info);
4591 }
4592
4593 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4594 {
4595         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4596                                 (u64)-1);
4597         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4598         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4599         WARN_ON(fs_info->trans_block_rsv.size > 0);
4600         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4601         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4602         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4603         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4604         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4605 }
4606
4607 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4608                                   struct btrfs_root *root)
4609 {
4610         if (!trans->block_rsv)
4611                 return;
4612
4613         if (!trans->bytes_reserved)
4614                 return;
4615
4616         trace_btrfs_space_reservation(root->fs_info, "transaction",
4617                                       trans->transid, trans->bytes_reserved, 0);
4618         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4619         trans->bytes_reserved = 0;
4620 }
4621
4622 /* Can only return 0 or -ENOSPC */
4623 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4624                                   struct inode *inode)
4625 {
4626         struct btrfs_root *root = BTRFS_I(inode)->root;
4627         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4628         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4629
4630         /*
4631          * We need to hold space in order to delete our orphan item once we've
4632          * added it, so this takes the reservation so we can release it later
4633          * when we are truly done with the orphan item.
4634          */
4635         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4636         trace_btrfs_space_reservation(root->fs_info, "orphan",
4637                                       btrfs_ino(inode), num_bytes, 1);
4638         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4639 }
4640
4641 void btrfs_orphan_release_metadata(struct inode *inode)
4642 {
4643         struct btrfs_root *root = BTRFS_I(inode)->root;
4644         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4645         trace_btrfs_space_reservation(root->fs_info, "orphan",
4646                                       btrfs_ino(inode), num_bytes, 0);
4647         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4648 }
4649
4650 /*
4651  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4652  * root: the root of the parent directory
4653  * rsv: block reservation
4654  * items: the number of items that we need do reservation
4655  * qgroup_reserved: used to return the reserved size in qgroup
4656  *
4657  * This function is used to reserve the space for snapshot/subvolume
4658  * creation and deletion. Those operations are different with the
4659  * common file/directory operations, they change two fs/file trees
4660  * and root tree, the number of items that the qgroup reserves is
4661  * different with the free space reservation. So we can not use
4662  * the space reseravtion mechanism in start_transaction().
4663  */
4664 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4665                                      struct btrfs_block_rsv *rsv,
4666                                      int items,
4667                                      u64 *qgroup_reserved)
4668 {
4669         u64 num_bytes;
4670         int ret;
4671
4672         if (root->fs_info->quota_enabled) {
4673                 /* One for parent inode, two for dir entries */
4674                 num_bytes = 3 * root->leafsize;
4675                 ret = btrfs_qgroup_reserve(root, num_bytes);
4676                 if (ret)
4677                         return ret;
4678         } else {
4679                 num_bytes = 0;
4680         }
4681
4682         *qgroup_reserved = num_bytes;
4683
4684         num_bytes = btrfs_calc_trans_metadata_size(root, items);
4685         rsv->space_info = __find_space_info(root->fs_info,
4686                                             BTRFS_BLOCK_GROUP_METADATA);
4687         ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4688                                   BTRFS_RESERVE_FLUSH_ALL);
4689         if (ret) {
4690                 if (*qgroup_reserved)
4691                         btrfs_qgroup_free(root, *qgroup_reserved);
4692         }
4693
4694         return ret;
4695 }
4696
4697 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4698                                       struct btrfs_block_rsv *rsv,
4699                                       u64 qgroup_reserved)
4700 {
4701         btrfs_block_rsv_release(root, rsv, (u64)-1);
4702         if (qgroup_reserved)
4703                 btrfs_qgroup_free(root, qgroup_reserved);
4704 }
4705
4706 /**
4707  * drop_outstanding_extent - drop an outstanding extent
4708  * @inode: the inode we're dropping the extent for
4709  *
4710  * This is called when we are freeing up an outstanding extent, either called
4711  * after an error or after an extent is written.  This will return the number of
4712  * reserved extents that need to be freed.  This must be called with
4713  * BTRFS_I(inode)->lock held.
4714  */
4715 static unsigned drop_outstanding_extent(struct inode *inode)
4716 {
4717         unsigned drop_inode_space = 0;
4718         unsigned dropped_extents = 0;
4719
4720         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4721         BTRFS_I(inode)->outstanding_extents--;
4722
4723         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4724             test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4725                                &BTRFS_I(inode)->runtime_flags))
4726                 drop_inode_space = 1;
4727
4728         /*
4729          * If we have more or the same amount of outsanding extents than we have
4730          * reserved then we need to leave the reserved extents count alone.
4731          */
4732         if (BTRFS_I(inode)->outstanding_extents >=
4733             BTRFS_I(inode)->reserved_extents)
4734                 return drop_inode_space;
4735
4736         dropped_extents = BTRFS_I(inode)->reserved_extents -
4737                 BTRFS_I(inode)->outstanding_extents;
4738         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4739         return dropped_extents + drop_inode_space;
4740 }
4741
4742 /**
4743  * calc_csum_metadata_size - return the amount of metada space that must be
4744  *      reserved/free'd for the given bytes.
4745  * @inode: the inode we're manipulating
4746  * @num_bytes: the number of bytes in question
4747  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4748  *
4749  * This adjusts the number of csum_bytes in the inode and then returns the
4750  * correct amount of metadata that must either be reserved or freed.  We
4751  * calculate how many checksums we can fit into one leaf and then divide the
4752  * number of bytes that will need to be checksumed by this value to figure out
4753  * how many checksums will be required.  If we are adding bytes then the number
4754  * may go up and we will return the number of additional bytes that must be
4755  * reserved.  If it is going down we will return the number of bytes that must
4756  * be freed.
4757  *
4758  * This must be called with BTRFS_I(inode)->lock held.
4759  */
4760 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4761                                    int reserve)
4762 {
4763         struct btrfs_root *root = BTRFS_I(inode)->root;
4764         u64 csum_size;
4765         int num_csums_per_leaf;
4766         int num_csums;
4767         int old_csums;
4768
4769         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4770             BTRFS_I(inode)->csum_bytes == 0)
4771                 return 0;
4772
4773         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4774         if (reserve)
4775                 BTRFS_I(inode)->csum_bytes += num_bytes;
4776         else
4777                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4778         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4779         num_csums_per_leaf = (int)div64_u64(csum_size,
4780                                             sizeof(struct btrfs_csum_item) +
4781                                             sizeof(struct btrfs_disk_key));
4782         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4783         num_csums = num_csums + num_csums_per_leaf - 1;
4784         num_csums = num_csums / num_csums_per_leaf;
4785
4786         old_csums = old_csums + num_csums_per_leaf - 1;
4787         old_csums = old_csums / num_csums_per_leaf;
4788
4789         /* No change, no need to reserve more */
4790         if (old_csums == num_csums)
4791                 return 0;
4792
4793         if (reserve)
4794                 return btrfs_calc_trans_metadata_size(root,
4795                                                       num_csums - old_csums);
4796
4797         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4798 }
4799
4800 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4801 {
4802         struct btrfs_root *root = BTRFS_I(inode)->root;
4803         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4804         u64 to_reserve = 0;
4805         u64 csum_bytes;
4806         unsigned nr_extents = 0;
4807         int extra_reserve = 0;
4808         enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4809         int ret = 0;
4810         bool delalloc_lock = true;
4811         u64 to_free = 0;
4812         unsigned dropped;
4813
4814         /* If we are a free space inode we need to not flush since we will be in
4815          * the middle of a transaction commit.  We also don't need the delalloc
4816          * mutex since we won't race with anybody.  We need this mostly to make
4817          * lockdep shut its filthy mouth.
4818          */
4819         if (btrfs_is_free_space_inode(inode)) {
4820                 flush = BTRFS_RESERVE_NO_FLUSH;
4821                 delalloc_lock = false;
4822         }
4823
4824         if (flush != BTRFS_RESERVE_NO_FLUSH &&
4825             btrfs_transaction_in_commit(root->fs_info))
4826                 schedule_timeout(1);
4827
4828         if (delalloc_lock)
4829                 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4830
4831         num_bytes = ALIGN(num_bytes, root->sectorsize);
4832
4833         spin_lock(&BTRFS_I(inode)->lock);
4834         BTRFS_I(inode)->outstanding_extents++;
4835
4836         if (BTRFS_I(inode)->outstanding_extents >
4837             BTRFS_I(inode)->reserved_extents)
4838                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4839                         BTRFS_I(inode)->reserved_extents;
4840
4841         /*
4842          * Add an item to reserve for updating the inode when we complete the
4843          * delalloc io.
4844          */
4845         if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4846                       &BTRFS_I(inode)->runtime_flags)) {
4847                 nr_extents++;
4848                 extra_reserve = 1;
4849         }
4850
4851         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4852         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4853         csum_bytes = BTRFS_I(inode)->csum_bytes;
4854         spin_unlock(&BTRFS_I(inode)->lock);
4855
4856         if (root->fs_info->quota_enabled) {
4857                 ret = btrfs_qgroup_reserve(root, num_bytes +
4858                                            nr_extents * root->leafsize);
4859                 if (ret)
4860                         goto out_fail;
4861         }
4862
4863         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4864         if (unlikely(ret)) {
4865                 if (root->fs_info->quota_enabled)
4866                         btrfs_qgroup_free(root, num_bytes +
4867                                                 nr_extents * root->leafsize);
4868                 goto out_fail;
4869         }
4870
4871         spin_lock(&BTRFS_I(inode)->lock);
4872         if (extra_reserve) {
4873                 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4874                         &BTRFS_I(inode)->runtime_flags);
4875                 nr_extents--;
4876         }
4877         BTRFS_I(inode)->reserved_extents += nr_extents;
4878         spin_unlock(&BTRFS_I(inode)->lock);
4879
4880         if (delalloc_lock)
4881                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4882
4883         if (to_reserve)
4884                 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4885                                               btrfs_ino(inode), to_reserve, 1);
4886         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4887
4888         return 0;
4889
4890 out_fail:
4891         spin_lock(&BTRFS_I(inode)->lock);
4892         dropped = drop_outstanding_extent(inode);
4893         /*
4894          * If the inodes csum_bytes is the same as the original
4895          * csum_bytes then we know we haven't raced with any free()ers
4896          * so we can just reduce our inodes csum bytes and carry on.
4897          */
4898         if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
4899                 calc_csum_metadata_size(inode, num_bytes, 0);
4900         } else {
4901                 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
4902                 u64 bytes;
4903
4904                 /*
4905                  * This is tricky, but first we need to figure out how much we
4906                  * free'd from any free-ers that occured during this
4907                  * reservation, so we reset ->csum_bytes to the csum_bytes
4908                  * before we dropped our lock, and then call the free for the
4909                  * number of bytes that were freed while we were trying our
4910                  * reservation.
4911                  */
4912                 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
4913                 BTRFS_I(inode)->csum_bytes = csum_bytes;
4914                 to_free = calc_csum_metadata_size(inode, bytes, 0);
4915
4916
4917                 /*
4918                  * Now we need to see how much we would have freed had we not
4919                  * been making this reservation and our ->csum_bytes were not
4920                  * artificially inflated.
4921                  */
4922                 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
4923                 bytes = csum_bytes - orig_csum_bytes;
4924                 bytes = calc_csum_metadata_size(inode, bytes, 0);
4925
4926                 /*
4927                  * Now reset ->csum_bytes to what it should be.  If bytes is
4928                  * more than to_free then we would have free'd more space had we
4929                  * not had an artificially high ->csum_bytes, so we need to free
4930                  * the remainder.  If bytes is the same or less then we don't
4931                  * need to do anything, the other free-ers did the correct
4932                  * thing.
4933                  */
4934                 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
4935                 if (bytes > to_free)
4936                         to_free = bytes - to_free;
4937                 else
4938                         to_free = 0;
4939         }
4940         spin_unlock(&BTRFS_I(inode)->lock);
4941         if (dropped)
4942                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4943
4944         if (to_free) {
4945                 btrfs_block_rsv_release(root, block_rsv, to_free);
4946                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4947                                               btrfs_ino(inode), to_free, 0);
4948         }
4949         if (delalloc_lock)
4950                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4951         return ret;
4952 }
4953
4954 /**
4955  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4956  * @inode: the inode to release the reservation for
4957  * @num_bytes: the number of bytes we're releasing
4958  *
4959  * This will release the metadata reservation for an inode.  This can be called
4960  * once we complete IO for a given set of bytes to release their metadata
4961  * reservations.
4962  */
4963 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4964 {
4965         struct btrfs_root *root = BTRFS_I(inode)->root;
4966         u64 to_free = 0;
4967         unsigned dropped;
4968
4969         num_bytes = ALIGN(num_bytes, root->sectorsize);
4970         spin_lock(&BTRFS_I(inode)->lock);
4971         dropped = drop_outstanding_extent(inode);
4972
4973         if (num_bytes)
4974                 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4975         spin_unlock(&BTRFS_I(inode)->lock);
4976         if (dropped > 0)
4977                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4978
4979         trace_btrfs_space_reservation(root->fs_info, "delalloc",
4980                                       btrfs_ino(inode), to_free, 0);
4981         if (root->fs_info->quota_enabled) {
4982                 btrfs_qgroup_free(root, num_bytes +
4983                                         dropped * root->leafsize);
4984         }
4985
4986         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4987                                 to_free);
4988 }
4989
4990 /**
4991  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4992  * @inode: inode we're writing to
4993  * @num_bytes: the number of bytes we want to allocate
4994  *
4995  * This will do the following things
4996  *
4997  * o reserve space in the data space info for num_bytes
4998  * o reserve space in the metadata space info based on number of outstanding
4999  *   extents and how much csums will be needed
5000  * o add to the inodes ->delalloc_bytes
5001  * o add it to the fs_info's delalloc inodes list.
5002  *
5003  * This will return 0 for success and -ENOSPC if there is no space left.
5004  */
5005 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5006 {
5007         int ret;
5008
5009         ret = btrfs_check_data_free_space(inode, num_bytes);
5010         if (ret)
5011                 return ret;
5012
5013         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5014         if (ret) {
5015                 btrfs_free_reserved_data_space(inode, num_bytes);
5016                 return ret;
5017         }
5018
5019         return 0;
5020 }
5021
5022 /**
5023  * btrfs_delalloc_release_space - release data and metadata space for delalloc
5024  * @inode: inode we're releasing space for
5025  * @num_bytes: the number of bytes we want to free up
5026  *
5027  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
5028  * called in the case that we don't need the metadata AND data reservations
5029  * anymore.  So if there is an error or we insert an inline extent.
5030  *
5031  * This function will release the metadata space that was not used and will
5032  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5033  * list if there are no delalloc bytes left.
5034  */
5035 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5036 {
5037         btrfs_delalloc_release_metadata(inode, num_bytes);
5038         btrfs_free_reserved_data_space(inode, num_bytes);
5039 }
5040
5041 static int update_block_group(struct btrfs_root *root,
5042                               u64 bytenr, u64 num_bytes, int alloc)
5043 {
5044         struct btrfs_block_group_cache *cache = NULL;
5045         struct btrfs_fs_info *info = root->fs_info;
5046         u64 total = num_bytes;
5047         u64 old_val;
5048         u64 byte_in_group;
5049         int factor;
5050
5051         /* block accounting for super block */
5052         spin_lock(&info->delalloc_lock);
5053         old_val = btrfs_super_bytes_used(info->super_copy);
5054         if (alloc)
5055                 old_val += num_bytes;
5056         else
5057                 old_val -= num_bytes;
5058         btrfs_set_super_bytes_used(info->super_copy, old_val);
5059         spin_unlock(&info->delalloc_lock);
5060
5061         while (total) {
5062                 cache = btrfs_lookup_block_group(info, bytenr);
5063                 if (!cache)
5064                         return -ENOENT;
5065                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5066                                     BTRFS_BLOCK_GROUP_RAID1 |
5067                                     BTRFS_BLOCK_GROUP_RAID10))
5068                         factor = 2;
5069                 else
5070                         factor = 1;
5071                 /*
5072                  * If this block group has free space cache written out, we
5073                  * need to make sure to load it if we are removing space.  This
5074                  * is because we need the unpinning stage to actually add the
5075                  * space back to the block group, otherwise we will leak space.
5076                  */
5077                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5078                         cache_block_group(cache, 1);
5079
5080                 byte_in_group = bytenr - cache->key.objectid;
5081                 WARN_ON(byte_in_group > cache->key.offset);
5082
5083                 spin_lock(&cache->space_info->lock);
5084                 spin_lock(&cache->lock);
5085
5086                 if (btrfs_test_opt(root, SPACE_CACHE) &&
5087                     cache->disk_cache_state < BTRFS_DC_CLEAR)
5088                         cache->disk_cache_state = BTRFS_DC_CLEAR;
5089
5090                 cache->dirty = 1;
5091                 old_val = btrfs_block_group_used(&cache->item);
5092                 num_bytes = min(total, cache->key.offset - byte_in_group);
5093                 if (alloc) {
5094                         old_val += num_bytes;
5095                         btrfs_set_block_group_used(&cache->item, old_val);
5096                         cache->reserved -= num_bytes;
5097                         cache->space_info->bytes_reserved -= num_bytes;
5098                         cache->space_info->bytes_used += num_bytes;
5099                         cache->space_info->disk_used += num_bytes * factor;
5100                         spin_unlock(&cache->lock);
5101                         spin_unlock(&cache->space_info->lock);
5102                 } else {
5103                         old_val -= num_bytes;
5104                         btrfs_set_block_group_used(&cache->item, old_val);
5105                         cache->pinned += num_bytes;
5106                         cache->space_info->bytes_pinned += num_bytes;
5107                         cache->space_info->bytes_used -= num_bytes;
5108                         cache->space_info->disk_used -= num_bytes * factor;
5109                         spin_unlock(&cache->lock);
5110                         spin_unlock(&cache->space_info->lock);
5111
5112                         set_extent_dirty(info->pinned_extents,
5113                                          bytenr, bytenr + num_bytes - 1,
5114                                          GFP_NOFS | __GFP_NOFAIL);
5115                 }
5116                 btrfs_put_block_group(cache);
5117                 total -= num_bytes;
5118                 bytenr += num_bytes;
5119         }
5120         return 0;
5121 }
5122
5123 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5124 {
5125         struct btrfs_block_group_cache *cache;
5126         u64 bytenr;
5127
5128         spin_lock(&root->fs_info->block_group_cache_lock);
5129         bytenr = root->fs_info->first_logical_byte;
5130         spin_unlock(&root->fs_info->block_group_cache_lock);
5131
5132         if (bytenr < (u64)-1)
5133                 return bytenr;
5134
5135         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5136         if (!cache)
5137                 return 0;
5138
5139         bytenr = cache->key.objectid;
5140         btrfs_put_block_group(cache);
5141
5142         return bytenr;
5143 }
5144
5145 static int pin_down_extent(struct btrfs_root *root,
5146                            struct btrfs_block_group_cache *cache,
5147                            u64 bytenr, u64 num_bytes, int reserved)
5148 {
5149         spin_lock(&cache->space_info->lock);
5150         spin_lock(&cache->lock);
5151         cache->pinned += num_bytes;
5152         cache->space_info->bytes_pinned += num_bytes;
5153         if (reserved) {
5154                 cache->reserved -= num_bytes;
5155                 cache->space_info->bytes_reserved -= num_bytes;
5156         }
5157         spin_unlock(&cache->lock);
5158         spin_unlock(&cache->space_info->lock);
5159
5160         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5161                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5162         return 0;
5163 }
5164
5165 /*
5166  * this function must be called within transaction
5167  */
5168 int btrfs_pin_extent(struct btrfs_root *root,
5169                      u64 bytenr, u64 num_bytes, int reserved)
5170 {
5171         struct btrfs_block_group_cache *cache;
5172
5173         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5174         BUG_ON(!cache); /* Logic error */
5175
5176         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5177
5178         btrfs_put_block_group(cache);
5179         return 0;
5180 }
5181
5182 /*
5183  * this function must be called within transaction
5184  */
5185 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5186                                     u64 bytenr, u64 num_bytes)
5187 {
5188         struct btrfs_block_group_cache *cache;
5189
5190         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5191         BUG_ON(!cache); /* Logic error */
5192
5193         /*
5194          * pull in the free space cache (if any) so that our pin
5195          * removes the free space from the cache.  We have load_only set
5196          * to one because the slow code to read in the free extents does check
5197          * the pinned extents.
5198          */
5199         cache_block_group(cache, 1);
5200
5201         pin_down_extent(root, cache, bytenr, num_bytes, 0);
5202
5203         /* remove us from the free space cache (if we're there at all) */
5204         btrfs_remove_free_space(cache, bytenr, num_bytes);
5205         btrfs_put_block_group(cache);
5206         return 0;
5207 }
5208
5209 /**
5210  * btrfs_update_reserved_bytes - update the block_group and space info counters
5211  * @cache:      The cache we are manipulating
5212  * @num_bytes:  The number of bytes in question
5213  * @reserve:    One of the reservation enums
5214  *
5215  * This is called by the allocator when it reserves space, or by somebody who is
5216  * freeing space that was never actually used on disk.  For example if you
5217  * reserve some space for a new leaf in transaction A and before transaction A
5218  * commits you free that leaf, you call this with reserve set to 0 in order to
5219  * clear the reservation.
5220  *
5221  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5222  * ENOSPC accounting.  For data we handle the reservation through clearing the
5223  * delalloc bits in the io_tree.  We have to do this since we could end up
5224  * allocating less disk space for the amount of data we have reserved in the
5225  * case of compression.
5226  *
5227  * If this is a reservation and the block group has become read only we cannot
5228  * make the reservation and return -EAGAIN, otherwise this function always
5229  * succeeds.
5230  */
5231 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5232                                        u64 num_bytes, int reserve)
5233 {
5234         struct btrfs_space_info *space_info = cache->space_info;
5235         int ret = 0;
5236
5237         spin_lock(&space_info->lock);
5238         spin_lock(&cache->lock);
5239         if (reserve != RESERVE_FREE) {
5240                 if (cache->ro) {
5241                         ret = -EAGAIN;
5242                 } else {
5243                         cache->reserved += num_bytes;
5244                         space_info->bytes_reserved += num_bytes;
5245                         if (reserve == RESERVE_ALLOC) {
5246                                 trace_btrfs_space_reservation(cache->fs_info,
5247                                                 "space_info", space_info->flags,
5248                                                 num_bytes, 0);
5249                                 space_info->bytes_may_use -= num_bytes;
5250                         }
5251                 }
5252         } else {
5253                 if (cache->ro)
5254                         space_info->bytes_readonly += num_bytes;
5255                 cache->reserved -= num_bytes;
5256                 space_info->bytes_reserved -= num_bytes;
5257                 space_info->reservation_progress++;
5258         }
5259         spin_unlock(&cache->lock);
5260         spin_unlock(&space_info->lock);
5261         return ret;
5262 }
5263
5264 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5265                                 struct btrfs_root *root)
5266 {
5267         struct btrfs_fs_info *fs_info = root->fs_info;
5268         struct btrfs_caching_control *next;
5269         struct btrfs_caching_control *caching_ctl;
5270         struct btrfs_block_group_cache *cache;
5271
5272         down_write(&fs_info->extent_commit_sem);
5273
5274         list_for_each_entry_safe(caching_ctl, next,
5275                                  &fs_info->caching_block_groups, list) {
5276                 cache = caching_ctl->block_group;
5277                 if (block_group_cache_done(cache)) {
5278                         cache->last_byte_to_unpin = (u64)-1;
5279                         list_del_init(&caching_ctl->list);
5280                         put_caching_control(caching_ctl);
5281                 } else {
5282                         cache->last_byte_to_unpin = caching_ctl->progress;
5283                 }
5284         }
5285
5286         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5287                 fs_info->pinned_extents = &fs_info->freed_extents[1];
5288         else
5289                 fs_info->pinned_extents = &fs_info->freed_extents[0];
5290
5291         up_write(&fs_info->extent_commit_sem);
5292
5293         update_global_block_rsv(fs_info);
5294 }
5295
5296 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5297 {
5298         struct btrfs_fs_info *fs_info = root->fs_info;
5299         struct btrfs_block_group_cache *cache = NULL;
5300         struct btrfs_space_info *space_info;
5301         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5302         u64 len;
5303         bool readonly;
5304
5305         while (start <= end) {
5306                 readonly = false;
5307                 if (!cache ||
5308                     start >= cache->key.objectid + cache->key.offset) {
5309                         if (cache)
5310                                 btrfs_put_block_group(cache);
5311                         cache = btrfs_lookup_block_group(fs_info, start);
5312                         BUG_ON(!cache); /* Logic error */
5313                 }
5314
5315                 len = cache->key.objectid + cache->key.offset - start;
5316                 len = min(len, end + 1 - start);
5317
5318                 if (start < cache->last_byte_to_unpin) {
5319                         len = min(len, cache->last_byte_to_unpin - start);
5320                         btrfs_add_free_space(cache, start, len);
5321                 }
5322
5323                 start += len;
5324                 space_info = cache->space_info;
5325
5326                 spin_lock(&space_info->lock);
5327                 spin_lock(&cache->lock);
5328                 cache->pinned -= len;
5329                 space_info->bytes_pinned -= len;
5330                 if (cache->ro) {
5331                         space_info->bytes_readonly += len;
5332                         readonly = true;
5333                 }
5334                 spin_unlock(&cache->lock);
5335                 if (!readonly && global_rsv->space_info == space_info) {
5336                         spin_lock(&global_rsv->lock);
5337                         if (!global_rsv->full) {
5338                                 len = min(len, global_rsv->size -
5339                                           global_rsv->reserved);
5340                                 global_rsv->reserved += len;
5341                                 space_info->bytes_may_use += len;
5342                                 if (global_rsv->reserved >= global_rsv->size)
5343                                         global_rsv->full = 1;
5344                         }
5345                         spin_unlock(&global_rsv->lock);
5346                 }
5347                 spin_unlock(&space_info->lock);
5348         }
5349
5350         if (cache)
5351                 btrfs_put_block_group(cache);
5352         return 0;
5353 }
5354
5355 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5356                                struct btrfs_root *root)
5357 {
5358         struct btrfs_fs_info *fs_info = root->fs_info;
5359         struct extent_io_tree *unpin;
5360         u64 start;
5361         u64 end;
5362         int ret;
5363
5364         if (trans->aborted)
5365                 return 0;
5366
5367         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5368                 unpin = &fs_info->freed_extents[1];
5369         else
5370                 unpin = &fs_info->freed_extents[0];
5371
5372         while (1) {
5373                 ret = find_first_extent_bit(unpin, 0, &start, &end,
5374                                             EXTENT_DIRTY, NULL);
5375                 if (ret)
5376                         break;
5377
5378                 if (btrfs_test_opt(root, DISCARD))
5379                         ret = btrfs_discard_extent(root, start,
5380                                                    end + 1 - start, NULL);
5381
5382                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5383                 unpin_extent_range(root, start, end);
5384                 cond_resched();
5385         }
5386
5387         return 0;
5388 }
5389
5390 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5391                                 struct btrfs_root *root,
5392                                 u64 bytenr, u64 num_bytes, u64 parent,
5393                                 u64 root_objectid, u64 owner_objectid,
5394                                 u64 owner_offset, int refs_to_drop,
5395                                 struct btrfs_delayed_extent_op *extent_op)
5396 {
5397         struct btrfs_key key;
5398         struct btrfs_path *path;
5399         struct btrfs_fs_info *info = root->fs_info;
5400         struct btrfs_root *extent_root = info->extent_root;
5401         struct extent_buffer *leaf;
5402         struct btrfs_extent_item *ei;
5403         struct btrfs_extent_inline_ref *iref;
5404         int ret;
5405         int is_data;
5406         int extent_slot = 0;
5407         int found_extent = 0;
5408         int num_to_del = 1;
5409         u32 item_size;
5410         u64 refs;
5411         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5412                                                  SKINNY_METADATA);
5413
5414         path = btrfs_alloc_path();
5415         if (!path)
5416                 return -ENOMEM;
5417
5418         path->reada = 1;
5419         path->leave_spinning = 1;
5420
5421         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5422         BUG_ON(!is_data && refs_to_drop != 1);
5423
5424         if (is_data)
5425                 skinny_metadata = 0;
5426
5427         ret = lookup_extent_backref(trans, extent_root, path, &iref,
5428                                     bytenr, num_bytes, parent,
5429                                     root_objectid, owner_objectid,
5430                                     owner_offset);
5431         if (ret == 0) {
5432                 extent_slot = path->slots[0];
5433                 while (extent_slot >= 0) {
5434                         btrfs_item_key_to_cpu(path->nodes[0], &key,
5435                                               extent_slot);
5436                         if (key.objectid != bytenr)
5437                                 break;
5438                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5439                             key.offset == num_bytes) {
5440                                 found_extent = 1;
5441                                 break;
5442                         }
5443                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
5444                             key.offset == owner_objectid) {
5445                                 found_extent = 1;
5446                                 break;
5447                         }
5448                         if (path->slots[0] - extent_slot > 5)
5449                                 break;
5450                         extent_slot--;
5451                 }
5452 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5453                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5454                 if (found_extent && item_size < sizeof(*ei))
5455                         found_extent = 0;
5456 #endif
5457                 if (!found_extent) {
5458                         BUG_ON(iref);
5459                         ret = remove_extent_backref(trans, extent_root, path,
5460                                                     NULL, refs_to_drop,
5461                                                     is_data);
5462                         if (ret) {
5463                                 btrfs_abort_transaction(trans, extent_root, ret);
5464                                 goto out;
5465                         }
5466                         btrfs_release_path(path);
5467                         path->leave_spinning = 1;
5468
5469                         key.objectid = bytenr;
5470                         key.type = BTRFS_EXTENT_ITEM_KEY;
5471                         key.offset = num_bytes;
5472
5473                         if (!is_data && skinny_metadata) {
5474                                 key.type = BTRFS_METADATA_ITEM_KEY;
5475                                 key.offset = owner_objectid;
5476                         }
5477
5478                         ret = btrfs_search_slot(trans, extent_root,
5479                                                 &key, path, -1, 1);
5480                         if (ret > 0 && skinny_metadata && path->slots[0]) {
5481                                 /*
5482                                  * Couldn't find our skinny metadata item,
5483                                  * see if we have ye olde extent item.
5484                                  */
5485                                 path->slots[0]--;
5486                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
5487                                                       path->slots[0]);
5488                                 if (key.objectid == bytenr &&
5489                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
5490                                     key.offset == num_bytes)
5491                                         ret = 0;
5492                         }
5493
5494                         if (ret > 0 && skinny_metadata) {
5495                                 skinny_metadata = false;
5496                                 key.type = BTRFS_EXTENT_ITEM_KEY;
5497                                 key.offset = num_bytes;
5498                                 btrfs_release_path(path);
5499                                 ret = btrfs_search_slot(trans, extent_root,
5500                                                         &key, path, -1, 1);
5501                         }
5502
5503                         if (ret) {
5504                                 printk(KERN_ERR "umm, got %d back from search"
5505                                        ", was looking for %llu\n", ret,
5506                                        (unsigned long long)bytenr);
5507                                 if (ret > 0)
5508                                         btrfs_print_leaf(extent_root,
5509                                                          path->nodes[0]);
5510                         }
5511                         if (ret < 0) {
5512                                 btrfs_abort_transaction(trans, extent_root, ret);
5513                                 goto out;
5514                         }
5515                         extent_slot = path->slots[0];
5516                 }
5517         } else if (ret == -ENOENT) {
5518                 btrfs_print_leaf(extent_root, path->nodes[0]);
5519                 WARN_ON(1);
5520                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5521                        "parent %llu root %llu  owner %llu offset %llu\n",
5522                        (unsigned long long)bytenr,
5523                        (unsigned long long)parent,
5524                        (unsigned long long)root_objectid,
5525                        (unsigned long long)owner_objectid,
5526                        (unsigned long long)owner_offset);
5527         } else {
5528                 btrfs_abort_transaction(trans, extent_root, ret);
5529                 goto out;
5530         }
5531
5532         leaf = path->nodes[0];
5533         item_size = btrfs_item_size_nr(leaf, extent_slot);
5534 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5535         if (item_size < sizeof(*ei)) {
5536                 BUG_ON(found_extent || extent_slot != path->slots[0]);
5537                 ret = convert_extent_item_v0(trans, extent_root, path,
5538                                              owner_objectid, 0);
5539                 if (ret < 0) {
5540                         btrfs_abort_transaction(trans, extent_root, ret);
5541                         goto out;
5542                 }
5543
5544                 btrfs_release_path(path);
5545                 path->leave_spinning = 1;
5546
5547                 key.objectid = bytenr;
5548                 key.type = BTRFS_EXTENT_ITEM_KEY;
5549                 key.offset = num_bytes;
5550
5551                 ret = btrfs_search_slot(trans, extent_root, &key, path,
5552                                         -1, 1);
5553                 if (ret) {
5554                         printk(KERN_ERR "umm, got %d back from search"
5555                                ", was looking for %llu\n", ret,
5556                                (unsigned long long)bytenr);
5557                         btrfs_print_leaf(extent_root, path->nodes[0]);
5558                 }
5559                 if (ret < 0) {
5560                         btrfs_abort_transaction(trans, extent_root, ret);
5561                         goto out;
5562                 }
5563
5564                 extent_slot = path->slots[0];
5565                 leaf = path->nodes[0];
5566                 item_size = btrfs_item_size_nr(leaf, extent_slot);
5567         }
5568 #endif
5569         BUG_ON(item_size < sizeof(*ei));
5570         ei = btrfs_item_ptr(leaf, extent_slot,
5571                             struct btrfs_extent_item);
5572         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5573             key.type == BTRFS_EXTENT_ITEM_KEY) {
5574                 struct btrfs_tree_block_info *bi;
5575                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5576                 bi = (struct btrfs_tree_block_info *)(ei + 1);
5577                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5578         }
5579
5580         refs = btrfs_extent_refs(leaf, ei);
5581         BUG_ON(refs < refs_to_drop);
5582         refs -= refs_to_drop;
5583
5584         if (refs > 0) {
5585                 if (extent_op)
5586                         __run_delayed_extent_op(extent_op, leaf, ei);
5587                 /*
5588                  * In the case of inline back ref, reference count will
5589                  * be updated by remove_extent_backref
5590                  */
5591                 if (iref) {
5592                         BUG_ON(!found_extent);
5593                 } else {
5594                         btrfs_set_extent_refs(leaf, ei, refs);
5595                         btrfs_mark_buffer_dirty(leaf);
5596                 }
5597                 if (found_extent) {
5598                         ret = remove_extent_backref(trans, extent_root, path,
5599                                                     iref, refs_to_drop,
5600                                                     is_data);
5601                         if (ret) {
5602                                 btrfs_abort_transaction(trans, extent_root, ret);
5603                                 goto out;
5604                         }
5605                 }
5606         } else {
5607                 if (found_extent) {
5608                         BUG_ON(is_data && refs_to_drop !=
5609                                extent_data_ref_count(root, path, iref));
5610                         if (iref) {
5611                                 BUG_ON(path->slots[0] != extent_slot);
5612                         } else {
5613                                 BUG_ON(path->slots[0] != extent_slot + 1);
5614                                 path->slots[0] = extent_slot;
5615                                 num_to_del = 2;
5616                         }
5617                 }
5618
5619                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5620                                       num_to_del);
5621                 if (ret) {
5622                         btrfs_abort_transaction(trans, extent_root, ret);
5623                         goto out;
5624                 }
5625                 btrfs_release_path(path);
5626
5627                 if (is_data) {
5628                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5629                         if (ret) {
5630                                 btrfs_abort_transaction(trans, extent_root, ret);
5631                                 goto out;
5632                         }
5633                 }
5634
5635                 ret = update_block_group(root, bytenr, num_bytes, 0);
5636                 if (ret) {
5637                         btrfs_abort_transaction(trans, extent_root, ret);
5638                         goto out;
5639                 }
5640         }
5641 out:
5642         btrfs_free_path(path);
5643         return ret;
5644 }
5645
5646 /*
5647  * when we free an block, it is possible (and likely) that we free the last
5648  * delayed ref for that extent as well.  This searches the delayed ref tree for
5649  * a given extent, and if there are no other delayed refs to be processed, it
5650  * removes it from the tree.
5651  */
5652 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5653                                       struct btrfs_root *root, u64 bytenr)
5654 {
5655         struct btrfs_delayed_ref_head *head;
5656         struct btrfs_delayed_ref_root *delayed_refs;
5657         struct btrfs_delayed_ref_node *ref;
5658         struct rb_node *node;
5659         int ret = 0;
5660
5661         delayed_refs = &trans->transaction->delayed_refs;
5662         spin_lock(&delayed_refs->lock);
5663         head = btrfs_find_delayed_ref_head(trans, bytenr);
5664         if (!head)
5665                 goto out;
5666
5667         node = rb_prev(&head->node.rb_node);
5668         if (!node)
5669                 goto out;
5670
5671         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5672
5673         /* there are still entries for this ref, we can't drop it */
5674         if (ref->bytenr == bytenr)
5675                 goto out;
5676
5677         if (head->extent_op) {
5678                 if (!head->must_insert_reserved)
5679                         goto out;
5680                 btrfs_free_delayed_extent_op(head->extent_op);
5681                 head->extent_op = NULL;
5682         }
5683
5684         /*
5685          * waiting for the lock here would deadlock.  If someone else has it
5686          * locked they are already in the process of dropping it anyway
5687          */
5688         if (!mutex_trylock(&head->mutex))
5689                 goto out;
5690
5691         /*
5692          * at this point we have a head with no other entries.  Go
5693          * ahead and process it.
5694          */
5695         head->node.in_tree = 0;
5696         rb_erase(&head->node.rb_node, &delayed_refs->root);
5697
5698         delayed_refs->num_entries--;
5699
5700         /*
5701          * we don't take a ref on the node because we're removing it from the
5702          * tree, so we just steal the ref the tree was holding.
5703          */
5704         delayed_refs->num_heads--;
5705         if (list_empty(&head->cluster))
5706                 delayed_refs->num_heads_ready--;
5707
5708         list_del_init(&head->cluster);
5709         spin_unlock(&delayed_refs->lock);
5710
5711         BUG_ON(head->extent_op);
5712         if (head->must_insert_reserved)
5713                 ret = 1;
5714
5715         mutex_unlock(&head->mutex);
5716         btrfs_put_delayed_ref(&head->node);
5717         return ret;
5718 out:
5719         spin_unlock(&delayed_refs->lock);
5720         return 0;
5721 }
5722
5723 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5724                            struct btrfs_root *root,
5725                            struct extent_buffer *buf,
5726                            u64 parent, int last_ref)
5727 {
5728         struct btrfs_block_group_cache *cache = NULL;
5729         int ret;
5730
5731         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5732                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5733                                         buf->start, buf->len,
5734                                         parent, root->root_key.objectid,
5735                                         btrfs_header_level(buf),
5736                                         BTRFS_DROP_DELAYED_REF, NULL, 0);
5737                 BUG_ON(ret); /* -ENOMEM */
5738         }
5739
5740         if (!last_ref)
5741                 return;
5742
5743         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5744
5745         if (btrfs_header_generation(buf) == trans->transid) {
5746                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5747                         ret = check_ref_cleanup(trans, root, buf->start);
5748                         if (!ret)
5749                                 goto out;
5750                 }
5751
5752                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5753                         pin_down_extent(root, cache, buf->start, buf->len, 1);
5754                         goto out;
5755                 }
5756
5757                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5758
5759                 btrfs_add_free_space(cache, buf->start, buf->len);
5760                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5761         }
5762 out:
5763         /*
5764          * Deleting the buffer, clear the corrupt flag since it doesn't matter
5765          * anymore.
5766          */
5767         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5768         btrfs_put_block_group(cache);
5769 }
5770
5771 /* Can return -ENOMEM */
5772 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5773                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5774                       u64 owner, u64 offset, int for_cow)
5775 {
5776         int ret;
5777         struct btrfs_fs_info *fs_info = root->fs_info;
5778
5779         /*
5780          * tree log blocks never actually go into the extent allocation
5781          * tree, just update pinning info and exit early.
5782          */
5783         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5784                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5785                 /* unlocks the pinned mutex */
5786                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5787                 ret = 0;
5788         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5789                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5790                                         num_bytes,
5791                                         parent, root_objectid, (int)owner,
5792                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5793         } else {
5794                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5795                                                 num_bytes,
5796                                                 parent, root_objectid, owner,
5797                                                 offset, BTRFS_DROP_DELAYED_REF,
5798                                                 NULL, for_cow);
5799         }
5800         return ret;
5801 }
5802
5803 static u64 stripe_align(struct btrfs_root *root,
5804                         struct btrfs_block_group_cache *cache,
5805                         u64 val, u64 num_bytes)
5806 {
5807         u64 ret = ALIGN(val, root->stripesize);
5808         return ret;
5809 }
5810
5811 /*
5812  * when we wait for progress in the block group caching, its because
5813  * our allocation attempt failed at least once.  So, we must sleep
5814  * and let some progress happen before we try again.
5815  *
5816  * This function will sleep at least once waiting for new free space to
5817  * show up, and then it will check the block group free space numbers
5818  * for our min num_bytes.  Another option is to have it go ahead
5819  * and look in the rbtree for a free extent of a given size, but this
5820  * is a good start.
5821  */
5822 static noinline int
5823 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5824                                 u64 num_bytes)
5825 {
5826         struct btrfs_caching_control *caching_ctl;
5827
5828         caching_ctl = get_caching_control(cache);
5829         if (!caching_ctl)
5830                 return 0;
5831
5832         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5833                    (cache->free_space_ctl->free_space >= num_bytes));
5834
5835         put_caching_control(caching_ctl);
5836         return 0;
5837 }
5838
5839 static noinline int
5840 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5841 {
5842         struct btrfs_caching_control *caching_ctl;
5843
5844         caching_ctl = get_caching_control(cache);
5845         if (!caching_ctl)
5846                 return 0;
5847
5848         wait_event(caching_ctl->wait, block_group_cache_done(cache));
5849
5850         put_caching_control(caching_ctl);
5851         return 0;
5852 }
5853
5854 int __get_raid_index(u64 flags)
5855 {
5856         if (flags & BTRFS_BLOCK_GROUP_RAID10)
5857                 return BTRFS_RAID_RAID10;
5858         else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5859                 return BTRFS_RAID_RAID1;
5860         else if (flags & BTRFS_BLOCK_GROUP_DUP)
5861                 return BTRFS_RAID_DUP;
5862         else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5863                 return BTRFS_RAID_RAID0;
5864         else if (flags & BTRFS_BLOCK_GROUP_RAID5)
5865                 return BTRFS_RAID_RAID5;
5866         else if (flags & BTRFS_BLOCK_GROUP_RAID6)
5867                 return BTRFS_RAID_RAID6;
5868
5869         return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
5870 }
5871
5872 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5873 {
5874         return __get_raid_index(cache->flags);
5875 }
5876
5877 enum btrfs_loop_type {
5878         LOOP_CACHING_NOWAIT = 0,
5879         LOOP_CACHING_WAIT = 1,
5880         LOOP_ALLOC_CHUNK = 2,
5881         LOOP_NO_EMPTY_SIZE = 3,
5882 };
5883
5884 /*
5885  * walks the btree of allocated extents and find a hole of a given size.
5886  * The key ins is changed to record the hole:
5887  * ins->objectid == block start
5888  * ins->flags = BTRFS_EXTENT_ITEM_KEY
5889  * ins->offset == number of blocks
5890  * Any available blocks before search_start are skipped.
5891  */
5892 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5893                                      struct btrfs_root *orig_root,
5894                                      u64 num_bytes, u64 empty_size,
5895                                      u64 hint_byte, struct btrfs_key *ins,
5896                                      u64 data)
5897 {
5898         int ret = 0;
5899         struct btrfs_root *root = orig_root->fs_info->extent_root;
5900         struct btrfs_free_cluster *last_ptr = NULL;
5901         struct btrfs_block_group_cache *block_group = NULL;
5902         struct btrfs_block_group_cache *used_block_group;
5903         u64 search_start = 0;
5904         int empty_cluster = 2 * 1024 * 1024;
5905         struct btrfs_space_info *space_info;
5906         int loop = 0;
5907         int index = __get_raid_index(data);
5908         int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5909                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5910         bool found_uncached_bg = false;
5911         bool failed_cluster_refill = false;
5912         bool failed_alloc = false;
5913         bool use_cluster = true;
5914         bool have_caching_bg = false;
5915
5916         WARN_ON(num_bytes < root->sectorsize);
5917         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5918         ins->objectid = 0;
5919         ins->offset = 0;
5920
5921         trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5922
5923         space_info = __find_space_info(root->fs_info, data);
5924         if (!space_info) {
5925                 printk(KERN_ERR "No space info for %llu\n", data);
5926                 return -ENOSPC;
5927         }
5928
5929         /*
5930          * If the space info is for both data and metadata it means we have a
5931          * small filesystem and we can't use the clustering stuff.
5932          */
5933         if (btrfs_mixed_space_info(space_info))
5934                 use_cluster = false;
5935
5936         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5937                 last_ptr = &root->fs_info->meta_alloc_cluster;
5938                 if (!btrfs_test_opt(root, SSD))
5939                         empty_cluster = 64 * 1024;
5940         }
5941
5942         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5943             btrfs_test_opt(root, SSD)) {
5944                 last_ptr = &root->fs_info->data_alloc_cluster;
5945         }
5946
5947         if (last_ptr) {
5948                 spin_lock(&last_ptr->lock);
5949                 if (last_ptr->block_group)
5950                         hint_byte = last_ptr->window_start;
5951                 spin_unlock(&last_ptr->lock);
5952         }
5953
5954         search_start = max(search_start, first_logical_byte(root, 0));
5955         search_start = max(search_start, hint_byte);
5956
5957         if (!last_ptr)
5958                 empty_cluster = 0;
5959
5960         if (search_start == hint_byte) {
5961                 block_group = btrfs_lookup_block_group(root->fs_info,
5962                                                        search_start);
5963                 used_block_group = block_group;
5964                 /*
5965                  * we don't want to use the block group if it doesn't match our
5966                  * allocation bits, or if its not cached.
5967                  *
5968                  * However if we are re-searching with an ideal block group
5969                  * picked out then we don't care that the block group is cached.
5970                  */
5971                 if (block_group && block_group_bits(block_group, data) &&
5972                     block_group->cached != BTRFS_CACHE_NO) {
5973                         down_read(&space_info->groups_sem);
5974                         if (list_empty(&block_group->list) ||
5975                             block_group->ro) {
5976                                 /*
5977                                  * someone is removing this block group,
5978                                  * we can't jump into the have_block_group
5979                                  * target because our list pointers are not
5980                                  * valid
5981                                  */
5982                                 btrfs_put_block_group(block_group);
5983                                 up_read(&space_info->groups_sem);
5984                         } else {
5985                                 index = get_block_group_index(block_group);
5986                                 goto have_block_group;
5987                         }
5988                 } else if (block_group) {
5989                         btrfs_put_block_group(block_group);
5990                 }
5991         }
5992 search:
5993         have_caching_bg = false;
5994         down_read(&space_info->groups_sem);
5995         list_for_each_entry(block_group, &space_info->block_groups[index],
5996                             list) {
5997                 u64 offset;
5998                 int cached;
5999
6000                 used_block_group = block_group;
6001                 btrfs_get_block_group(block_group);
6002                 search_start = block_group->key.objectid;
6003
6004                 /*
6005                  * this can happen if we end up cycling through all the
6006                  * raid types, but we want to make sure we only allocate
6007                  * for the proper type.
6008                  */
6009                 if (!block_group_bits(block_group, data)) {
6010                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
6011                                 BTRFS_BLOCK_GROUP_RAID1 |
6012                                 BTRFS_BLOCK_GROUP_RAID5 |
6013                                 BTRFS_BLOCK_GROUP_RAID6 |
6014                                 BTRFS_BLOCK_GROUP_RAID10;
6015
6016                         /*
6017                          * if they asked for extra copies and this block group
6018                          * doesn't provide them, bail.  This does allow us to
6019                          * fill raid0 from raid1.
6020                          */
6021                         if ((data & extra) && !(block_group->flags & extra))
6022                                 goto loop;
6023                 }
6024
6025 have_block_group:
6026                 cached = block_group_cache_done(block_group);
6027                 if (unlikely(!cached)) {
6028                         found_uncached_bg = true;
6029                         ret = cache_block_group(block_group, 0);
6030                         BUG_ON(ret < 0);
6031                         ret = 0;
6032                 }
6033
6034                 if (unlikely(block_group->ro))
6035                         goto loop;
6036
6037                 /*
6038                  * Ok we want to try and use the cluster allocator, so
6039                  * lets look there
6040                  */
6041                 if (last_ptr) {
6042                         unsigned long aligned_cluster;
6043                         /*
6044                          * the refill lock keeps out other
6045                          * people trying to start a new cluster
6046                          */
6047                         spin_lock(&last_ptr->refill_lock);
6048                         used_block_group = last_ptr->block_group;
6049                         if (used_block_group != block_group &&
6050                             (!used_block_group ||
6051                              used_block_group->ro ||
6052                              !block_group_bits(used_block_group, data))) {
6053                                 used_block_group = block_group;
6054                                 goto refill_cluster;
6055                         }
6056
6057                         if (used_block_group != block_group)
6058                                 btrfs_get_block_group(used_block_group);
6059
6060                         offset = btrfs_alloc_from_cluster(used_block_group,
6061                           last_ptr, num_bytes, used_block_group->key.objectid);
6062                         if (offset) {
6063                                 /* we have a block, we're done */
6064                                 spin_unlock(&last_ptr->refill_lock);
6065                                 trace_btrfs_reserve_extent_cluster(root,
6066                                         block_group, search_start, num_bytes);
6067                                 goto checks;
6068                         }
6069
6070                         WARN_ON(last_ptr->block_group != used_block_group);
6071                         if (used_block_group != block_group) {
6072                                 btrfs_put_block_group(used_block_group);
6073                                 used_block_group = block_group;
6074                         }
6075 refill_cluster:
6076                         BUG_ON(used_block_group != block_group);
6077                         /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6078                          * set up a new clusters, so lets just skip it
6079                          * and let the allocator find whatever block
6080                          * it can find.  If we reach this point, we
6081                          * will have tried the cluster allocator
6082                          * plenty of times and not have found
6083                          * anything, so we are likely way too
6084                          * fragmented for the clustering stuff to find
6085                          * anything.
6086                          *
6087                          * However, if the cluster is taken from the
6088                          * current block group, release the cluster
6089                          * first, so that we stand a better chance of
6090                          * succeeding in the unclustered
6091                          * allocation.  */
6092                         if (loop >= LOOP_NO_EMPTY_SIZE &&
6093                             last_ptr->block_group != block_group) {
6094                                 spin_unlock(&last_ptr->refill_lock);
6095                                 goto unclustered_alloc;
6096                         }
6097
6098                         /*
6099                          * this cluster didn't work out, free it and
6100                          * start over
6101                          */
6102                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
6103
6104                         if (loop >= LOOP_NO_EMPTY_SIZE) {
6105                                 spin_unlock(&last_ptr->refill_lock);
6106                                 goto unclustered_alloc;
6107                         }
6108
6109                         aligned_cluster = max_t(unsigned long,
6110                                                 empty_cluster + empty_size,
6111                                               block_group->full_stripe_len);
6112
6113                         /* allocate a cluster in this block group */
6114                         ret = btrfs_find_space_cluster(trans, root,
6115                                                block_group, last_ptr,
6116                                                search_start, num_bytes,
6117                                                aligned_cluster);
6118                         if (ret == 0) {
6119                                 /*
6120                                  * now pull our allocation out of this
6121                                  * cluster
6122                                  */
6123                                 offset = btrfs_alloc_from_cluster(block_group,
6124                                                   last_ptr, num_bytes,
6125                                                   search_start);
6126                                 if (offset) {
6127                                         /* we found one, proceed */
6128                                         spin_unlock(&last_ptr->refill_lock);
6129                                         trace_btrfs_reserve_extent_cluster(root,
6130                                                 block_group, search_start,
6131                                                 num_bytes);
6132                                         goto checks;
6133                                 }
6134                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
6135                                    && !failed_cluster_refill) {
6136                                 spin_unlock(&last_ptr->refill_lock);
6137
6138                                 failed_cluster_refill = true;
6139                                 wait_block_group_cache_progress(block_group,
6140                                        num_bytes + empty_cluster + empty_size);
6141                                 goto have_block_group;
6142                         }
6143
6144                         /*
6145                          * at this point we either didn't find a cluster
6146                          * or we weren't able to allocate a block from our
6147                          * cluster.  Free the cluster we've been trying
6148                          * to use, and go to the next block group
6149                          */
6150                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
6151                         spin_unlock(&last_ptr->refill_lock);
6152                         goto loop;
6153                 }
6154
6155 unclustered_alloc:
6156                 spin_lock(&block_group->free_space_ctl->tree_lock);
6157                 if (cached &&
6158                     block_group->free_space_ctl->free_space <
6159                     num_bytes + empty_cluster + empty_size) {
6160                         spin_unlock(&block_group->free_space_ctl->tree_lock);
6161                         goto loop;
6162                 }
6163                 spin_unlock(&block_group->free_space_ctl->tree_lock);
6164
6165                 offset = btrfs_find_space_for_alloc(block_group, search_start,
6166                                                     num_bytes, empty_size);
6167                 /*
6168                  * If we didn't find a chunk, and we haven't failed on this
6169                  * block group before, and this block group is in the middle of
6170                  * caching and we are ok with waiting, then go ahead and wait
6171                  * for progress to be made, and set failed_alloc to true.
6172                  *
6173                  * If failed_alloc is true then we've already waited on this
6174                  * block group once and should move on to the next block group.
6175                  */
6176                 if (!offset && !failed_alloc && !cached &&
6177                     loop > LOOP_CACHING_NOWAIT) {
6178                         wait_block_group_cache_progress(block_group,
6179                                                 num_bytes + empty_size);
6180                         failed_alloc = true;
6181                         goto have_block_group;
6182                 } else if (!offset) {
6183                         if (!cached)
6184                                 have_caching_bg = true;
6185                         goto loop;
6186                 }
6187 checks:
6188                 search_start = stripe_align(root, used_block_group,
6189                                             offset, num_bytes);
6190
6191                 /* move on to the next group */
6192                 if (search_start + num_bytes >
6193                     used_block_group->key.objectid + used_block_group->key.offset) {
6194                         btrfs_add_free_space(used_block_group, offset, num_bytes);
6195                         goto loop;
6196                 }
6197
6198                 if (offset < search_start)
6199                         btrfs_add_free_space(used_block_group, offset,
6200                                              search_start - offset);
6201                 BUG_ON(offset > search_start);
6202
6203                 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6204                                                   alloc_type);
6205                 if (ret == -EAGAIN) {
6206                         btrfs_add_free_space(used_block_group, offset, num_bytes);
6207                         goto loop;
6208                 }
6209
6210                 /* we are all good, lets return */
6211                 ins->objectid = search_start;
6212                 ins->offset = num_bytes;
6213
6214                 trace_btrfs_reserve_extent(orig_root, block_group,
6215                                            search_start, num_bytes);
6216                 if (used_block_group != block_group)
6217                         btrfs_put_block_group(used_block_group);
6218                 btrfs_put_block_group(block_group);
6219                 break;
6220 loop:
6221                 failed_cluster_refill = false;
6222                 failed_alloc = false;
6223                 BUG_ON(index != get_block_group_index(block_group));
6224                 if (used_block_group != block_group)
6225                         btrfs_put_block_group(used_block_group);
6226                 btrfs_put_block_group(block_group);
6227         }
6228         up_read(&space_info->groups_sem);
6229
6230         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6231                 goto search;
6232
6233         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6234                 goto search;
6235
6236         /*
6237          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6238          *                      caching kthreads as we move along
6239          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6240          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6241          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6242          *                      again
6243          */
6244         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6245                 index = 0;
6246                 loop++;
6247                 if (loop == LOOP_ALLOC_CHUNK) {
6248                         ret = do_chunk_alloc(trans, root, data,
6249                                              CHUNK_ALLOC_FORCE);
6250                         /*
6251                          * Do not bail out on ENOSPC since we
6252                          * can do more things.
6253                          */
6254                         if (ret < 0 && ret != -ENOSPC) {
6255                                 btrfs_abort_transaction(trans,
6256                                                         root, ret);
6257                                 goto out;
6258                         }
6259                 }
6260
6261                 if (loop == LOOP_NO_EMPTY_SIZE) {
6262                         empty_size = 0;
6263                         empty_cluster = 0;
6264                 }
6265
6266                 goto search;
6267         } else if (!ins->objectid) {
6268                 ret = -ENOSPC;
6269         } else if (ins->objectid) {
6270                 ret = 0;
6271         }
6272 out:
6273
6274         return ret;
6275 }
6276
6277 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6278                             int dump_block_groups)
6279 {
6280         struct btrfs_block_group_cache *cache;
6281         int index = 0;
6282
6283         spin_lock(&info->lock);
6284         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6285                (unsigned long long)info->flags,
6286                (unsigned long long)(info->total_bytes - info->bytes_used -
6287                                     info->bytes_pinned - info->bytes_reserved -
6288                                     info->bytes_readonly),
6289                (info->full) ? "" : "not ");
6290         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6291                "reserved=%llu, may_use=%llu, readonly=%llu\n",
6292                (unsigned long long)info->total_bytes,
6293                (unsigned long long)info->bytes_used,
6294                (unsigned long long)info->bytes_pinned,
6295                (unsigned long long)info->bytes_reserved,
6296                (unsigned long long)info->bytes_may_use,
6297                (unsigned long long)info->bytes_readonly);
6298         spin_unlock(&info->lock);
6299
6300         if (!dump_block_groups)
6301                 return;
6302
6303         down_read(&info->groups_sem);
6304 again:
6305         list_for_each_entry(cache, &info->block_groups[index], list) {
6306                 spin_lock(&cache->lock);
6307                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6308                        (unsigned long long)cache->key.objectid,
6309                        (unsigned long long)cache->key.offset,
6310                        (unsigned long long)btrfs_block_group_used(&cache->item),
6311                        (unsigned long long)cache->pinned,
6312                        (unsigned long long)cache->reserved,
6313                        cache->ro ? "[readonly]" : "");
6314                 btrfs_dump_free_space(cache, bytes);
6315                 spin_unlock(&cache->lock);
6316         }
6317         if (++index < BTRFS_NR_RAID_TYPES)
6318                 goto again;
6319         up_read(&info->groups_sem);
6320 }
6321
6322 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6323                          struct btrfs_root *root,
6324                          u64 num_bytes, u64 min_alloc_size,
6325                          u64 empty_size, u64 hint_byte,
6326                          struct btrfs_key *ins, u64 data)
6327 {
6328         bool final_tried = false;
6329         int ret;
6330
6331         data = btrfs_get_alloc_profile(root, data);
6332 again:
6333         WARN_ON(num_bytes < root->sectorsize);
6334         ret = find_free_extent(trans, root, num_bytes, empty_size,
6335                                hint_byte, ins, data);
6336
6337         if (ret == -ENOSPC) {
6338                 if (!final_tried) {
6339                         num_bytes = num_bytes >> 1;
6340                         num_bytes = round_down(num_bytes, root->sectorsize);
6341                         num_bytes = max(num_bytes, min_alloc_size);
6342                         if (num_bytes == min_alloc_size)
6343                                 final_tried = true;
6344                         goto again;
6345                 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6346                         struct btrfs_space_info *sinfo;
6347
6348                         sinfo = __find_space_info(root->fs_info, data);
6349                         printk(KERN_ERR "btrfs allocation failed flags %llu, "
6350                                "wanted %llu\n", (unsigned long long)data,
6351                                (unsigned long long)num_bytes);
6352                         if (sinfo)
6353                                 dump_space_info(sinfo, num_bytes, 1);
6354                 }
6355         }
6356
6357         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6358
6359         return ret;
6360 }
6361
6362 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6363                                         u64 start, u64 len, int pin)
6364 {
6365         struct btrfs_block_group_cache *cache;
6366         int ret = 0;
6367
6368         cache = btrfs_lookup_block_group(root->fs_info, start);
6369         if (!cache) {
6370                 printk(KERN_ERR "Unable to find block group for %llu\n",
6371                        (unsigned long long)start);
6372                 return -ENOSPC;
6373         }
6374
6375         if (btrfs_test_opt(root, DISCARD))
6376                 ret = btrfs_discard_extent(root, start, len, NULL);
6377
6378         if (pin)
6379                 pin_down_extent(root, cache, start, len, 1);
6380         else {
6381                 btrfs_add_free_space(cache, start, len);
6382                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6383         }
6384         btrfs_put_block_group(cache);
6385
6386         trace_btrfs_reserved_extent_free(root, start, len);
6387
6388         return ret;
6389 }
6390
6391 int btrfs_free_reserved_extent(struct btrfs_root *root,
6392                                         u64 start, u64 len)
6393 {
6394         return __btrfs_free_reserved_extent(root, start, len, 0);
6395 }
6396
6397 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6398                                        u64 start, u64 len)
6399 {
6400         return __btrfs_free_reserved_extent(root, start, len, 1);
6401 }
6402
6403 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6404                                       struct btrfs_root *root,
6405                                       u64 parent, u64 root_objectid,
6406                                       u64 flags, u64 owner, u64 offset,
6407                                       struct btrfs_key *ins, int ref_mod)
6408 {
6409         int ret;
6410         struct btrfs_fs_info *fs_info = root->fs_info;
6411         struct btrfs_extent_item *extent_item;
6412         struct btrfs_extent_inline_ref *iref;
6413         struct btrfs_path *path;
6414         struct extent_buffer *leaf;
6415         int type;
6416         u32 size;
6417
6418         if (parent > 0)
6419                 type = BTRFS_SHARED_DATA_REF_KEY;
6420         else
6421                 type = BTRFS_EXTENT_DATA_REF_KEY;
6422
6423         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6424
6425         path = btrfs_alloc_path();
6426         if (!path)
6427                 return -ENOMEM;
6428
6429         path->leave_spinning = 1;
6430         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6431                                       ins, size);
6432         if (ret) {
6433                 btrfs_free_path(path);
6434                 return ret;
6435         }
6436
6437         leaf = path->nodes[0];
6438         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6439                                      struct btrfs_extent_item);
6440         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6441         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6442         btrfs_set_extent_flags(leaf, extent_item,
6443                                flags | BTRFS_EXTENT_FLAG_DATA);
6444
6445         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6446         btrfs_set_extent_inline_ref_type(leaf, iref, type);
6447         if (parent > 0) {
6448                 struct btrfs_shared_data_ref *ref;
6449                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6450                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6451                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6452         } else {
6453                 struct btrfs_extent_data_ref *ref;
6454                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6455                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6456                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6457                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6458                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6459         }
6460
6461         btrfs_mark_buffer_dirty(path->nodes[0]);
6462         btrfs_free_path(path);
6463
6464         ret = update_block_group(root, ins->objectid, ins->offset, 1);
6465         if (ret) { /* -ENOENT, logic error */
6466                 printk(KERN_ERR "btrfs update block group failed for %llu "
6467                        "%llu\n", (unsigned long long)ins->objectid,
6468                        (unsigned long long)ins->offset);
6469                 BUG();
6470         }
6471         return ret;
6472 }
6473
6474 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6475                                      struct btrfs_root *root,
6476                                      u64 parent, u64 root_objectid,
6477                                      u64 flags, struct btrfs_disk_key *key,
6478                                      int level, struct btrfs_key *ins)
6479 {
6480         int ret;
6481         struct btrfs_fs_info *fs_info = root->fs_info;
6482         struct btrfs_extent_item *extent_item;
6483         struct btrfs_tree_block_info *block_info;
6484         struct btrfs_extent_inline_ref *iref;
6485         struct btrfs_path *path;
6486         struct extent_buffer *leaf;
6487         u32 size = sizeof(*extent_item) + sizeof(*iref);
6488         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6489                                                  SKINNY_METADATA);
6490
6491         if (!skinny_metadata)
6492                 size += sizeof(*block_info);
6493
6494         path = btrfs_alloc_path();
6495         if (!path)
6496                 return -ENOMEM;
6497
6498         path->leave_spinning = 1;
6499         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6500                                       ins, size);
6501         if (ret) {
6502                 btrfs_free_path(path);
6503                 return ret;
6504         }
6505
6506         leaf = path->nodes[0];
6507         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6508                                      struct btrfs_extent_item);
6509         btrfs_set_extent_refs(leaf, extent_item, 1);
6510         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6511         btrfs_set_extent_flags(leaf, extent_item,
6512                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6513
6514         if (skinny_metadata) {
6515                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6516         } else {
6517                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6518                 btrfs_set_tree_block_key(leaf, block_info, key);
6519                 btrfs_set_tree_block_level(leaf, block_info, level);
6520                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6521         }
6522
6523         if (parent > 0) {
6524                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6525                 btrfs_set_extent_inline_ref_type(leaf, iref,
6526                                                  BTRFS_SHARED_BLOCK_REF_KEY);
6527                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6528         } else {
6529                 btrfs_set_extent_inline_ref_type(leaf, iref,
6530                                                  BTRFS_TREE_BLOCK_REF_KEY);
6531                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6532         }
6533
6534         btrfs_mark_buffer_dirty(leaf);
6535         btrfs_free_path(path);
6536
6537         ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6538         if (ret) { /* -ENOENT, logic error */
6539                 printk(KERN_ERR "btrfs update block group failed for %llu "
6540                        "%llu\n", (unsigned long long)ins->objectid,
6541                        (unsigned long long)ins->offset);
6542                 BUG();
6543         }
6544         return ret;
6545 }
6546
6547 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6548                                      struct btrfs_root *root,
6549                                      u64 root_objectid, u64 owner,
6550                                      u64 offset, struct btrfs_key *ins)
6551 {
6552         int ret;
6553
6554         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6555
6556         ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6557                                          ins->offset, 0,
6558                                          root_objectid, owner, offset,
6559                                          BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6560         return ret;
6561 }
6562
6563 /*
6564  * this is used by the tree logging recovery code.  It records that
6565  * an extent has been allocated and makes sure to clear the free
6566  * space cache bits as well
6567  */
6568 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6569                                    struct btrfs_root *root,
6570                                    u64 root_objectid, u64 owner, u64 offset,
6571                                    struct btrfs_key *ins)
6572 {
6573         int ret;
6574         struct btrfs_block_group_cache *block_group;
6575         struct btrfs_caching_control *caching_ctl;
6576         u64 start = ins->objectid;
6577         u64 num_bytes = ins->offset;
6578
6579         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6580         cache_block_group(block_group, 0);
6581         caching_ctl = get_caching_control(block_group);
6582
6583         if (!caching_ctl) {
6584                 BUG_ON(!block_group_cache_done(block_group));
6585                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6586                 BUG_ON(ret); /* -ENOMEM */
6587         } else {
6588                 mutex_lock(&caching_ctl->mutex);
6589
6590                 if (start >= caching_ctl->progress) {
6591                         ret = add_excluded_extent(root, start, num_bytes);
6592                         BUG_ON(ret); /* -ENOMEM */
6593                 } else if (start + num_bytes <= caching_ctl->progress) {
6594                         ret = btrfs_remove_free_space(block_group,
6595                                                       start, num_bytes);
6596                         BUG_ON(ret); /* -ENOMEM */
6597                 } else {
6598                         num_bytes = caching_ctl->progress - start;
6599                         ret = btrfs_remove_free_space(block_group,
6600                                                       start, num_bytes);
6601                         BUG_ON(ret); /* -ENOMEM */
6602
6603                         start = caching_ctl->progress;
6604                         num_bytes = ins->objectid + ins->offset -
6605                                     caching_ctl->progress;
6606                         ret = add_excluded_extent(root, start, num_bytes);
6607                         BUG_ON(ret); /* -ENOMEM */
6608                 }
6609
6610                 mutex_unlock(&caching_ctl->mutex);
6611                 put_caching_control(caching_ctl);
6612         }
6613
6614         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6615                                           RESERVE_ALLOC_NO_ACCOUNT);
6616         BUG_ON(ret); /* logic error */
6617         btrfs_put_block_group(block_group);
6618         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6619                                          0, owner, offset, ins, 1);
6620         return ret;
6621 }
6622
6623 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6624                                             struct btrfs_root *root,
6625                                             u64 bytenr, u32 blocksize,
6626                                             int level)
6627 {
6628         struct extent_buffer *buf;
6629
6630         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6631         if (!buf)
6632                 return ERR_PTR(-ENOMEM);
6633         btrfs_set_header_generation(buf, trans->transid);
6634         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6635         btrfs_tree_lock(buf);
6636         clean_tree_block(trans, root, buf);
6637         clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6638
6639         btrfs_set_lock_blocking(buf);
6640         btrfs_set_buffer_uptodate(buf);
6641
6642         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6643                 /*
6644                  * we allow two log transactions at a time, use different
6645                  * EXENT bit to differentiate dirty pages.
6646                  */
6647                 if (root->log_transid % 2 == 0)
6648                         set_extent_dirty(&root->dirty_log_pages, buf->start,
6649                                         buf->start + buf->len - 1, GFP_NOFS);
6650                 else
6651                         set_extent_new(&root->dirty_log_pages, buf->start,
6652                                         buf->start + buf->len - 1, GFP_NOFS);
6653         } else {
6654                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6655                          buf->start + buf->len - 1, GFP_NOFS);
6656         }
6657         trans->blocks_used++;
6658         /* this returns a buffer locked for blocking */
6659         return buf;
6660 }
6661
6662 static struct btrfs_block_rsv *
6663 use_block_rsv(struct btrfs_trans_handle *trans,
6664               struct btrfs_root *root, u32 blocksize)
6665 {
6666         struct btrfs_block_rsv *block_rsv;
6667         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6668         int ret;
6669
6670         block_rsv = get_block_rsv(trans, root);
6671
6672         if (block_rsv->size == 0) {
6673                 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6674                                              BTRFS_RESERVE_NO_FLUSH);
6675                 /*
6676                  * If we couldn't reserve metadata bytes try and use some from
6677                  * the global reserve.
6678                  */
6679                 if (ret && block_rsv != global_rsv) {
6680                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6681                         if (!ret)
6682                                 return global_rsv;
6683                         return ERR_PTR(ret);
6684                 } else if (ret) {
6685                         return ERR_PTR(ret);
6686                 }
6687                 return block_rsv;
6688         }
6689
6690         ret = block_rsv_use_bytes(block_rsv, blocksize);
6691         if (!ret)
6692                 return block_rsv;
6693         if (ret && !block_rsv->failfast) {
6694                 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6695                         static DEFINE_RATELIMIT_STATE(_rs,
6696                                         DEFAULT_RATELIMIT_INTERVAL * 10,
6697                                         /*DEFAULT_RATELIMIT_BURST*/ 1);
6698                         if (__ratelimit(&_rs))
6699                                 WARN(1, KERN_DEBUG
6700                                         "btrfs: block rsv returned %d\n", ret);
6701                 }
6702                 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6703                                              BTRFS_RESERVE_NO_FLUSH);
6704                 if (!ret) {
6705                         return block_rsv;
6706                 } else if (ret && block_rsv != global_rsv) {
6707                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6708                         if (!ret)
6709                                 return global_rsv;
6710                 }
6711         }
6712
6713         return ERR_PTR(-ENOSPC);
6714 }
6715
6716 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6717                             struct btrfs_block_rsv *block_rsv, u32 blocksize)
6718 {
6719         block_rsv_add_bytes(block_rsv, blocksize, 0);
6720         block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6721 }
6722
6723 /*
6724  * finds a free extent and does all the dirty work required for allocation
6725  * returns the key for the extent through ins, and a tree buffer for
6726  * the first block of the extent through buf.
6727  *
6728  * returns the tree buffer or NULL.
6729  */
6730 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6731                                         struct btrfs_root *root, u32 blocksize,
6732                                         u64 parent, u64 root_objectid,
6733                                         struct btrfs_disk_key *key, int level,
6734                                         u64 hint, u64 empty_size)
6735 {
6736         struct btrfs_key ins;
6737         struct btrfs_block_rsv *block_rsv;
6738         struct extent_buffer *buf;
6739         u64 flags = 0;
6740         int ret;
6741         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6742                                                  SKINNY_METADATA);
6743
6744         block_rsv = use_block_rsv(trans, root, blocksize);
6745         if (IS_ERR(block_rsv))
6746                 return ERR_CAST(block_rsv);
6747
6748         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6749                                    empty_size, hint, &ins, 0);
6750         if (ret) {
6751                 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6752                 return ERR_PTR(ret);
6753         }
6754
6755         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6756                                     blocksize, level);
6757         BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6758
6759         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6760                 if (parent == 0)
6761                         parent = ins.objectid;
6762                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6763         } else
6764                 BUG_ON(parent > 0);
6765
6766         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6767                 struct btrfs_delayed_extent_op *extent_op;
6768                 extent_op = btrfs_alloc_delayed_extent_op();
6769                 BUG_ON(!extent_op); /* -ENOMEM */
6770                 if (key)
6771                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
6772                 else
6773                         memset(&extent_op->key, 0, sizeof(extent_op->key));
6774                 extent_op->flags_to_set = flags;
6775                 if (skinny_metadata)
6776                         extent_op->update_key = 0;
6777                 else
6778                         extent_op->update_key = 1;
6779                 extent_op->update_flags = 1;
6780                 extent_op->is_data = 0;
6781
6782                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6783                                         ins.objectid,
6784                                         ins.offset, parent, root_objectid,
6785                                         level, BTRFS_ADD_DELAYED_EXTENT,
6786                                         extent_op, 0);
6787                 BUG_ON(ret); /* -ENOMEM */
6788         }
6789         return buf;
6790 }
6791
6792 struct walk_control {
6793         u64 refs[BTRFS_MAX_LEVEL];
6794         u64 flags[BTRFS_MAX_LEVEL];
6795         struct btrfs_key update_progress;
6796         int stage;
6797         int level;
6798         int shared_level;
6799         int update_ref;
6800         int keep_locks;
6801         int reada_slot;
6802         int reada_count;
6803         int for_reloc;
6804 };
6805
6806 #define DROP_REFERENCE  1
6807 #define UPDATE_BACKREF  2
6808
6809 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6810                                      struct btrfs_root *root,
6811                                      struct walk_control *wc,
6812                                      struct btrfs_path *path)
6813 {
6814         u64 bytenr;
6815         u64 generation;
6816         u64 refs;
6817         u64 flags;
6818         u32 nritems;
6819         u32 blocksize;
6820         struct btrfs_key key;
6821         struct extent_buffer *eb;
6822         int ret;
6823         int slot;
6824         int nread = 0;
6825
6826         if (path->slots[wc->level] < wc->reada_slot) {
6827                 wc->reada_count = wc->reada_count * 2 / 3;
6828                 wc->reada_count = max(wc->reada_count, 2);
6829         } else {
6830                 wc->reada_count = wc->reada_count * 3 / 2;
6831                 wc->reada_count = min_t(int, wc->reada_count,
6832                                         BTRFS_NODEPTRS_PER_BLOCK(root));
6833         }
6834
6835         eb = path->nodes[wc->level];
6836         nritems = btrfs_header_nritems(eb);
6837         blocksize = btrfs_level_size(root, wc->level - 1);
6838
6839         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6840                 if (nread >= wc->reada_count)
6841                         break;
6842
6843                 cond_resched();
6844                 bytenr = btrfs_node_blockptr(eb, slot);
6845                 generation = btrfs_node_ptr_generation(eb, slot);
6846
6847                 if (slot == path->slots[wc->level])
6848                         goto reada;
6849
6850                 if (wc->stage == UPDATE_BACKREF &&
6851                     generation <= root->root_key.offset)
6852                         continue;
6853
6854                 /* We don't lock the tree block, it's OK to be racy here */
6855                 ret = btrfs_lookup_extent_info(trans, root, bytenr,
6856                                                wc->level - 1, 1, &refs,
6857                                                &flags);
6858                 /* We don't care about errors in readahead. */
6859                 if (ret < 0)
6860                         continue;
6861                 BUG_ON(refs == 0);
6862
6863                 if (wc->stage == DROP_REFERENCE) {
6864                         if (refs == 1)
6865                                 goto reada;
6866
6867                         if (wc->level == 1 &&
6868                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6869                                 continue;
6870                         if (!wc->update_ref ||
6871                             generation <= root->root_key.offset)
6872                                 continue;
6873                         btrfs_node_key_to_cpu(eb, &key, slot);
6874                         ret = btrfs_comp_cpu_keys(&key,
6875                                                   &wc->update_progress);
6876                         if (ret < 0)
6877                                 continue;
6878                 } else {
6879                         if (wc->level == 1 &&
6880                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6881                                 continue;
6882                 }
6883 reada:
6884                 ret = readahead_tree_block(root, bytenr, blocksize,
6885                                            generation);
6886                 if (ret)
6887                         break;
6888                 nread++;
6889         }
6890         wc->reada_slot = slot;
6891 }
6892
6893 /*
6894  * helper to process tree block while walking down the tree.
6895  *
6896  * when wc->stage == UPDATE_BACKREF, this function updates
6897  * back refs for pointers in the block.
6898  *
6899  * NOTE: return value 1 means we should stop walking down.
6900  */
6901 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6902                                    struct btrfs_root *root,
6903                                    struct btrfs_path *path,
6904                                    struct walk_control *wc, int lookup_info)
6905 {
6906         int level = wc->level;
6907         struct extent_buffer *eb = path->nodes[level];
6908         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6909         int ret;
6910
6911         if (wc->stage == UPDATE_BACKREF &&
6912             btrfs_header_owner(eb) != root->root_key.objectid)
6913                 return 1;
6914
6915         /*
6916          * when reference count of tree block is 1, it won't increase
6917          * again. once full backref flag is set, we never clear it.
6918          */
6919         if (lookup_info &&
6920             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6921              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6922                 BUG_ON(!path->locks[level]);
6923                 ret = btrfs_lookup_extent_info(trans, root,
6924                                                eb->start, level, 1,
6925                                                &wc->refs[level],
6926                                                &wc->flags[level]);
6927                 BUG_ON(ret == -ENOMEM);
6928                 if (ret)
6929                         return ret;
6930                 BUG_ON(wc->refs[level] == 0);
6931         }
6932
6933         if (wc->stage == DROP_REFERENCE) {
6934                 if (wc->refs[level] > 1)
6935                         return 1;
6936
6937                 if (path->locks[level] && !wc->keep_locks) {
6938                         btrfs_tree_unlock_rw(eb, path->locks[level]);
6939                         path->locks[level] = 0;
6940                 }
6941                 return 0;
6942         }
6943
6944         /* wc->stage == UPDATE_BACKREF */
6945         if (!(wc->flags[level] & flag)) {
6946                 BUG_ON(!path->locks[level]);
6947                 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6948                 BUG_ON(ret); /* -ENOMEM */
6949                 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6950                 BUG_ON(ret); /* -ENOMEM */
6951                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6952                                                   eb->len, flag, 0);
6953                 BUG_ON(ret); /* -ENOMEM */
6954                 wc->flags[level] |= flag;
6955         }
6956
6957         /*
6958          * the block is shared by multiple trees, so it's not good to
6959          * keep the tree lock
6960          */
6961         if (path->locks[level] && level > 0) {
6962                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6963                 path->locks[level] = 0;
6964         }
6965         return 0;
6966 }
6967
6968 /*
6969  * helper to process tree block pointer.
6970  *
6971  * when wc->stage == DROP_REFERENCE, this function checks
6972  * reference count of the block pointed to. if the block
6973  * is shared and we need update back refs for the subtree
6974  * rooted at the block, this function changes wc->stage to
6975  * UPDATE_BACKREF. if the block is shared and there is no
6976  * need to update back, this function drops the reference
6977  * to the block.
6978  *
6979  * NOTE: return value 1 means we should stop walking down.
6980  */
6981 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6982                                  struct btrfs_root *root,
6983                                  struct btrfs_path *path,
6984                                  struct walk_control *wc, int *lookup_info)
6985 {
6986         u64 bytenr;
6987         u64 generation;
6988         u64 parent;
6989         u32 blocksize;
6990         struct btrfs_key key;
6991         struct extent_buffer *next;
6992         int level = wc->level;
6993         int reada = 0;
6994         int ret = 0;
6995
6996         generation = btrfs_node_ptr_generation(path->nodes[level],
6997                                                path->slots[level]);
6998         /*
6999          * if the lower level block was created before the snapshot
7000          * was created, we know there is no need to update back refs
7001          * for the subtree
7002          */
7003         if (wc->stage == UPDATE_BACKREF &&
7004             generation <= root->root_key.offset) {
7005                 *lookup_info = 1;
7006                 return 1;
7007         }
7008
7009         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7010         blocksize = btrfs_level_size(root, level - 1);
7011
7012         next = btrfs_find_tree_block(root, bytenr, blocksize);
7013         if (!next) {
7014                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7015                 if (!next)
7016                         return -ENOMEM;
7017                 reada = 1;
7018         }
7019         btrfs_tree_lock(next);
7020         btrfs_set_lock_blocking(next);
7021
7022         ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7023                                        &wc->refs[level - 1],
7024                                        &wc->flags[level - 1]);
7025         if (ret < 0) {
7026                 btrfs_tree_unlock(next);
7027                 return ret;
7028         }
7029
7030         BUG_ON(wc->refs[level - 1] == 0);
7031         *lookup_info = 0;
7032
7033         if (wc->stage == DROP_REFERENCE) {
7034                 if (wc->refs[level - 1] > 1) {
7035                         if (level == 1 &&
7036                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7037                                 goto skip;
7038
7039                         if (!wc->update_ref ||
7040                             generation <= root->root_key.offset)
7041                                 goto skip;
7042
7043                         btrfs_node_key_to_cpu(path->nodes[level], &key,
7044                                               path->slots[level]);
7045                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7046                         if (ret < 0)
7047                                 goto skip;
7048
7049                         wc->stage = UPDATE_BACKREF;
7050                         wc->shared_level = level - 1;
7051                 }
7052         } else {
7053                 if (level == 1 &&
7054                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7055                         goto skip;
7056         }
7057
7058         if (!btrfs_buffer_uptodate(next, generation, 0)) {
7059                 btrfs_tree_unlock(next);
7060                 free_extent_buffer(next);
7061                 next = NULL;
7062                 *lookup_info = 1;
7063         }
7064
7065         if (!next) {
7066                 if (reada && level == 1)
7067                         reada_walk_down(trans, root, wc, path);
7068                 next = read_tree_block(root, bytenr, blocksize, generation);
7069                 if (!next)
7070                         return -EIO;
7071                 btrfs_tree_lock(next);
7072                 btrfs_set_lock_blocking(next);
7073         }
7074
7075         level--;
7076         BUG_ON(level != btrfs_header_level(next));
7077         path->nodes[level] = next;
7078         path->slots[level] = 0;
7079         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7080         wc->level = level;
7081         if (wc->level == 1)
7082                 wc->reada_slot = 0;
7083         return 0;
7084 skip:
7085         wc->refs[level - 1] = 0;
7086         wc->flags[level - 1] = 0;
7087         if (wc->stage == DROP_REFERENCE) {
7088                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7089                         parent = path->nodes[level]->start;
7090                 } else {
7091                         BUG_ON(root->root_key.objectid !=
7092                                btrfs_header_owner(path->nodes[level]));
7093                         parent = 0;
7094                 }
7095
7096                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7097                                 root->root_key.objectid, level - 1, 0, 0);
7098                 BUG_ON(ret); /* -ENOMEM */
7099         }
7100         btrfs_tree_unlock(next);
7101         free_extent_buffer(next);
7102         *lookup_info = 1;
7103         return 1;
7104 }
7105
7106 /*
7107  * helper to process tree block while walking up the tree.
7108  *
7109  * when wc->stage == DROP_REFERENCE, this function drops
7110  * reference count on the block.
7111  *
7112  * when wc->stage == UPDATE_BACKREF, this function changes
7113  * wc->stage back to DROP_REFERENCE if we changed wc->stage
7114  * to UPDATE_BACKREF previously while processing the block.
7115  *
7116  * NOTE: return value 1 means we should stop walking up.
7117  */
7118 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7119                                  struct btrfs_root *root,
7120                                  struct btrfs_path *path,
7121                                  struct walk_control *wc)
7122 {
7123         int ret;
7124         int level = wc->level;
7125         struct extent_buffer *eb = path->nodes[level];
7126         u64 parent = 0;
7127
7128         if (wc->stage == UPDATE_BACKREF) {
7129                 BUG_ON(wc->shared_level < level);
7130                 if (level < wc->shared_level)
7131                         goto out;
7132
7133                 ret = find_next_key(path, level + 1, &wc->update_progress);
7134                 if (ret > 0)
7135                         wc->update_ref = 0;
7136
7137                 wc->stage = DROP_REFERENCE;
7138                 wc->shared_level = -1;
7139                 path->slots[level] = 0;
7140
7141                 /*
7142                  * check reference count again if the block isn't locked.
7143                  * we should start walking down the tree again if reference
7144                  * count is one.
7145                  */
7146                 if (!path->locks[level]) {
7147                         BUG_ON(level == 0);
7148                         btrfs_tree_lock(eb);
7149                         btrfs_set_lock_blocking(eb);
7150                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7151
7152                         ret = btrfs_lookup_extent_info(trans, root,
7153                                                        eb->start, level, 1,
7154                                                        &wc->refs[level],
7155                                                        &wc->flags[level]);
7156                         if (ret < 0) {
7157                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7158                                 path->locks[level] = 0;
7159                                 return ret;
7160                         }
7161                         BUG_ON(wc->refs[level] == 0);
7162                         if (wc->refs[level] == 1) {
7163                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
7164                                 path->locks[level] = 0;
7165                                 return 1;
7166                         }
7167                 }
7168         }
7169
7170         /* wc->stage == DROP_REFERENCE */
7171         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7172
7173         if (wc->refs[level] == 1) {
7174                 if (level == 0) {
7175                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7176                                 ret = btrfs_dec_ref(trans, root, eb, 1,
7177                                                     wc->for_reloc);
7178                         else
7179                                 ret = btrfs_dec_ref(trans, root, eb, 0,
7180                                                     wc->for_reloc);
7181                         BUG_ON(ret); /* -ENOMEM */
7182                 }
7183                 /* make block locked assertion in clean_tree_block happy */
7184                 if (!path->locks[level] &&
7185                     btrfs_header_generation(eb) == trans->transid) {
7186                         btrfs_tree_lock(eb);
7187                         btrfs_set_lock_blocking(eb);
7188                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7189                 }
7190                 clean_tree_block(trans, root, eb);
7191         }
7192
7193         if (eb == root->node) {
7194                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7195                         parent = eb->start;
7196                 else
7197                         BUG_ON(root->root_key.objectid !=
7198                                btrfs_header_owner(eb));
7199         } else {
7200                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7201                         parent = path->nodes[level + 1]->start;
7202                 else
7203                         BUG_ON(root->root_key.objectid !=
7204                                btrfs_header_owner(path->nodes[level + 1]));
7205         }
7206
7207         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7208 out:
7209         wc->refs[level] = 0;
7210         wc->flags[level] = 0;
7211         return 0;
7212 }
7213
7214 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7215                                    struct btrfs_root *root,
7216                                    struct btrfs_path *path,
7217                                    struct walk_control *wc)
7218 {
7219         int level = wc->level;
7220         int lookup_info = 1;
7221         int ret;
7222
7223         while (level >= 0) {
7224                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7225                 if (ret > 0)
7226                         break;
7227
7228                 if (level == 0)
7229                         break;
7230
7231                 if (path->slots[level] >=
7232                     btrfs_header_nritems(path->nodes[level]))
7233                         break;
7234
7235                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7236                 if (ret > 0) {
7237                         path->slots[level]++;
7238                         continue;
7239                 } else if (ret < 0)
7240                         return ret;
7241                 level = wc->level;
7242         }
7243         return 0;
7244 }
7245
7246 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7247                                  struct btrfs_root *root,
7248                                  struct btrfs_path *path,
7249                                  struct walk_control *wc, int max_level)
7250 {
7251         int level = wc->level;
7252         int ret;
7253
7254         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7255         while (level < max_level && path->nodes[level]) {
7256                 wc->level = level;
7257                 if (path->slots[level] + 1 <
7258                     btrfs_header_nritems(path->nodes[level])) {
7259                         path->slots[level]++;
7260                         return 0;
7261                 } else {
7262                         ret = walk_up_proc(trans, root, path, wc);
7263                         if (ret > 0)
7264                                 return 0;
7265
7266                         if (path->locks[level]) {
7267                                 btrfs_tree_unlock_rw(path->nodes[level],
7268                                                      path->locks[level]);
7269                                 path->locks[level] = 0;
7270                         }
7271                         free_extent_buffer(path->nodes[level]);
7272                         path->nodes[level] = NULL;
7273                         level++;
7274                 }
7275         }
7276         return 1;
7277 }
7278
7279 /*
7280  * drop a subvolume tree.
7281  *
7282  * this function traverses the tree freeing any blocks that only
7283  * referenced by the tree.
7284  *
7285  * when a shared tree block is found. this function decreases its
7286  * reference count by one. if update_ref is true, this function
7287  * also make sure backrefs for the shared block and all lower level
7288  * blocks are properly updated.
7289  */
7290 int btrfs_drop_snapshot(struct btrfs_root *root,
7291                          struct btrfs_block_rsv *block_rsv, int update_ref,
7292                          int for_reloc)
7293 {
7294         struct btrfs_path *path;
7295         struct btrfs_trans_handle *trans;
7296         struct btrfs_root *tree_root = root->fs_info->tree_root;
7297         struct btrfs_root_item *root_item = &root->root_item;
7298         struct walk_control *wc;
7299         struct btrfs_key key;
7300         int err = 0;
7301         int ret;
7302         int level;
7303
7304         path = btrfs_alloc_path();
7305         if (!path) {
7306                 err = -ENOMEM;
7307                 goto out;
7308         }
7309
7310         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7311         if (!wc) {
7312                 btrfs_free_path(path);
7313                 err = -ENOMEM;
7314                 goto out;
7315         }
7316
7317         trans = btrfs_start_transaction(tree_root, 0);
7318         if (IS_ERR(trans)) {
7319                 err = PTR_ERR(trans);
7320                 goto out_free;
7321         }
7322
7323         if (block_rsv)
7324                 trans->block_rsv = block_rsv;
7325
7326         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7327                 level = btrfs_header_level(root->node);
7328                 path->nodes[level] = btrfs_lock_root_node(root);
7329                 btrfs_set_lock_blocking(path->nodes[level]);
7330                 path->slots[level] = 0;
7331                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7332                 memset(&wc->update_progress, 0,
7333                        sizeof(wc->update_progress));
7334         } else {
7335                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7336                 memcpy(&wc->update_progress, &key,
7337                        sizeof(wc->update_progress));
7338
7339                 level = root_item->drop_level;
7340                 BUG_ON(level == 0);
7341                 path->lowest_level = level;
7342                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7343                 path->lowest_level = 0;
7344                 if (ret < 0) {
7345                         err = ret;
7346                         goto out_end_trans;
7347                 }
7348                 WARN_ON(ret > 0);
7349
7350                 /*
7351                  * unlock our path, this is safe because only this
7352                  * function is allowed to delete this snapshot
7353                  */
7354                 btrfs_unlock_up_safe(path, 0);
7355
7356                 level = btrfs_header_level(root->node);
7357                 while (1) {
7358                         btrfs_tree_lock(path->nodes[level]);
7359                         btrfs_set_lock_blocking(path->nodes[level]);
7360
7361                         ret = btrfs_lookup_extent_info(trans, root,
7362                                                 path->nodes[level]->start,
7363                                                 level, 1, &wc->refs[level],
7364                                                 &wc->flags[level]);
7365                         if (ret < 0) {
7366                                 err = ret;
7367                                 goto out_end_trans;
7368                         }
7369                         BUG_ON(wc->refs[level] == 0);
7370
7371                         if (level == root_item->drop_level)
7372                                 break;
7373
7374                         btrfs_tree_unlock(path->nodes[level]);
7375                         WARN_ON(wc->refs[level] != 1);
7376                         level--;
7377                 }
7378         }
7379
7380         wc->level = level;
7381         wc->shared_level = -1;
7382         wc->stage = DROP_REFERENCE;
7383         wc->update_ref = update_ref;
7384         wc->keep_locks = 0;
7385         wc->for_reloc = for_reloc;
7386         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7387
7388         while (1) {
7389                 ret = walk_down_tree(trans, root, path, wc);
7390                 if (ret < 0) {
7391                         err = ret;
7392                         break;
7393                 }
7394
7395                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7396                 if (ret < 0) {
7397                         err = ret;
7398                         break;
7399                 }
7400
7401                 if (ret > 0) {
7402                         BUG_ON(wc->stage != DROP_REFERENCE);
7403                         break;
7404                 }
7405
7406                 if (wc->stage == DROP_REFERENCE) {
7407                         level = wc->level;
7408                         btrfs_node_key(path->nodes[level],
7409                                        &root_item->drop_progress,
7410                                        path->slots[level]);
7411                         root_item->drop_level = level;
7412                 }
7413
7414                 BUG_ON(wc->level == 0);
7415                 if (btrfs_should_end_transaction(trans, tree_root)) {
7416                         ret = btrfs_update_root(trans, tree_root,
7417                                                 &root->root_key,
7418                                                 root_item);
7419                         if (ret) {
7420                                 btrfs_abort_transaction(trans, tree_root, ret);
7421                                 err = ret;
7422                                 goto out_end_trans;
7423                         }
7424
7425                         btrfs_end_transaction_throttle(trans, tree_root);
7426                         trans = btrfs_start_transaction(tree_root, 0);
7427                         if (IS_ERR(trans)) {
7428                                 err = PTR_ERR(trans);
7429                                 goto out_free;
7430                         }
7431                         if (block_rsv)
7432                                 trans->block_rsv = block_rsv;
7433                 }
7434         }
7435         btrfs_release_path(path);
7436         if (err)
7437                 goto out_end_trans;
7438
7439         ret = btrfs_del_root(trans, tree_root, &root->root_key);
7440         if (ret) {
7441                 btrfs_abort_transaction(trans, tree_root, ret);
7442                 goto out_end_trans;
7443         }
7444
7445         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7446                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7447                                            NULL, NULL);
7448                 if (ret < 0) {
7449                         btrfs_abort_transaction(trans, tree_root, ret);
7450                         err = ret;
7451                         goto out_end_trans;
7452                 } else if (ret > 0) {
7453                         /* if we fail to delete the orphan item this time
7454                          * around, it'll get picked up the next time.
7455                          *
7456                          * The most common failure here is just -ENOENT.
7457                          */
7458                         btrfs_del_orphan_item(trans, tree_root,
7459                                               root->root_key.objectid);
7460                 }
7461         }
7462
7463         if (root->in_radix) {
7464                 btrfs_free_fs_root(tree_root->fs_info, root);
7465         } else {
7466                 free_extent_buffer(root->node);
7467                 free_extent_buffer(root->commit_root);
7468                 kfree(root);
7469         }
7470 out_end_trans:
7471         btrfs_end_transaction_throttle(trans, tree_root);
7472 out_free:
7473         kfree(wc);
7474         btrfs_free_path(path);
7475 out:
7476         if (err)
7477                 btrfs_std_error(root->fs_info, err);
7478         return err;
7479 }
7480
7481 /*
7482  * drop subtree rooted at tree block 'node'.
7483  *
7484  * NOTE: this function will unlock and release tree block 'node'
7485  * only used by relocation code
7486  */
7487 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7488                         struct btrfs_root *root,
7489                         struct extent_buffer *node,
7490                         struct extent_buffer *parent)
7491 {
7492         struct btrfs_path *path;
7493         struct walk_control *wc;
7494         int level;
7495         int parent_level;
7496         int ret = 0;
7497         int wret;
7498
7499         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7500
7501         path = btrfs_alloc_path();
7502         if (!path)
7503                 return -ENOMEM;
7504
7505         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7506         if (!wc) {
7507                 btrfs_free_path(path);
7508                 return -ENOMEM;
7509         }
7510
7511         btrfs_assert_tree_locked(parent);
7512         parent_level = btrfs_header_level(parent);
7513         extent_buffer_get(parent);
7514         path->nodes[parent_level] = parent;
7515         path->slots[parent_level] = btrfs_header_nritems(parent);
7516
7517         btrfs_assert_tree_locked(node);
7518         level = btrfs_header_level(node);
7519         path->nodes[level] = node;
7520         path->slots[level] = 0;
7521         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7522
7523         wc->refs[parent_level] = 1;
7524         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7525         wc->level = level;
7526         wc->shared_level = -1;
7527         wc->stage = DROP_REFERENCE;
7528         wc->update_ref = 0;
7529         wc->keep_locks = 1;
7530         wc->for_reloc = 1;
7531         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7532
7533         while (1) {
7534                 wret = walk_down_tree(trans, root, path, wc);
7535                 if (wret < 0) {
7536                         ret = wret;
7537                         break;
7538                 }
7539
7540                 wret = walk_up_tree(trans, root, path, wc, parent_level);
7541                 if (wret < 0)
7542                         ret = wret;
7543                 if (wret != 0)
7544                         break;
7545         }
7546
7547         kfree(wc);
7548         btrfs_free_path(path);
7549         return ret;
7550 }
7551
7552 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7553 {
7554         u64 num_devices;
7555         u64 stripped;
7556
7557         /*
7558          * if restripe for this chunk_type is on pick target profile and
7559          * return, otherwise do the usual balance
7560          */
7561         stripped = get_restripe_target(root->fs_info, flags);
7562         if (stripped)
7563                 return extended_to_chunk(stripped);
7564
7565         /*
7566          * we add in the count of missing devices because we want
7567          * to make sure that any RAID levels on a degraded FS
7568          * continue to be honored.
7569          */
7570         num_devices = root->fs_info->fs_devices->rw_devices +
7571                 root->fs_info->fs_devices->missing_devices;
7572
7573         stripped = BTRFS_BLOCK_GROUP_RAID0 |
7574                 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7575                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7576
7577         if (num_devices == 1) {
7578                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7579                 stripped = flags & ~stripped;
7580
7581                 /* turn raid0 into single device chunks */
7582                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7583                         return stripped;
7584
7585                 /* turn mirroring into duplication */
7586                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7587                              BTRFS_BLOCK_GROUP_RAID10))
7588                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7589         } else {
7590                 /* they already had raid on here, just return */
7591                 if (flags & stripped)
7592                         return flags;
7593
7594                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7595                 stripped = flags & ~stripped;
7596
7597                 /* switch duplicated blocks with raid1 */
7598                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7599                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7600
7601                 /* this is drive concat, leave it alone */
7602         }
7603
7604         return flags;
7605 }
7606
7607 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7608 {
7609         struct btrfs_space_info *sinfo = cache->space_info;
7610         u64 num_bytes;
7611         u64 min_allocable_bytes;
7612         int ret = -ENOSPC;
7613
7614
7615         /*
7616          * We need some metadata space and system metadata space for
7617          * allocating chunks in some corner cases until we force to set
7618          * it to be readonly.
7619          */
7620         if ((sinfo->flags &
7621              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7622             !force)
7623                 min_allocable_bytes = 1 * 1024 * 1024;
7624         else
7625                 min_allocable_bytes = 0;
7626
7627         spin_lock(&sinfo->lock);
7628         spin_lock(&cache->lock);
7629
7630         if (cache->ro) {
7631                 ret = 0;
7632                 goto out;
7633         }
7634
7635         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7636                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7637
7638         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7639             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7640             min_allocable_bytes <= sinfo->total_bytes) {
7641                 sinfo->bytes_readonly += num_bytes;
7642                 cache->ro = 1;
7643                 ret = 0;
7644         }
7645 out:
7646         spin_unlock(&cache->lock);
7647         spin_unlock(&sinfo->lock);
7648         return ret;
7649 }
7650
7651 int btrfs_set_block_group_ro(struct btrfs_root *root,
7652                              struct btrfs_block_group_cache *cache)
7653
7654 {
7655         struct btrfs_trans_handle *trans;
7656         u64 alloc_flags;
7657         int ret;
7658
7659         BUG_ON(cache->ro);
7660
7661         trans = btrfs_join_transaction(root);
7662         if (IS_ERR(trans))
7663                 return PTR_ERR(trans);
7664
7665         alloc_flags = update_block_group_flags(root, cache->flags);
7666         if (alloc_flags != cache->flags) {
7667                 ret = do_chunk_alloc(trans, root, alloc_flags,
7668                                      CHUNK_ALLOC_FORCE);
7669                 if (ret < 0)
7670                         goto out;
7671         }
7672
7673         ret = set_block_group_ro(cache, 0);
7674         if (!ret)
7675                 goto out;
7676         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7677         ret = do_chunk_alloc(trans, root, alloc_flags,
7678                              CHUNK_ALLOC_FORCE);
7679         if (ret < 0)
7680                 goto out;
7681         ret = set_block_group_ro(cache, 0);
7682 out:
7683         btrfs_end_transaction(trans, root);
7684         return ret;
7685 }
7686
7687 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7688                             struct btrfs_root *root, u64 type)
7689 {
7690         u64 alloc_flags = get_alloc_profile(root, type);
7691         return do_chunk_alloc(trans, root, alloc_flags,
7692                               CHUNK_ALLOC_FORCE);
7693 }
7694
7695 /*
7696  * helper to account the unused space of all the readonly block group in the
7697  * list. takes mirrors into account.
7698  */
7699 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7700 {
7701         struct btrfs_block_group_cache *block_group;
7702         u64 free_bytes = 0;
7703         int factor;
7704
7705         list_for_each_entry(block_group, groups_list, list) {
7706                 spin_lock(&block_group->lock);
7707
7708                 if (!block_group->ro) {
7709                         spin_unlock(&block_group->lock);
7710                         continue;
7711                 }
7712
7713                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7714                                           BTRFS_BLOCK_GROUP_RAID10 |
7715                                           BTRFS_BLOCK_GROUP_DUP))
7716                         factor = 2;
7717                 else
7718                         factor = 1;
7719
7720                 free_bytes += (block_group->key.offset -
7721                                btrfs_block_group_used(&block_group->item)) *
7722                                factor;
7723
7724                 spin_unlock(&block_group->lock);
7725         }
7726
7727         return free_bytes;
7728 }
7729
7730 /*
7731  * helper to account the unused space of all the readonly block group in the
7732  * space_info. takes mirrors into account.
7733  */
7734 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7735 {
7736         int i;
7737         u64 free_bytes = 0;
7738
7739         spin_lock(&sinfo->lock);
7740
7741         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7742                 if (!list_empty(&sinfo->block_groups[i]))
7743                         free_bytes += __btrfs_get_ro_block_group_free_space(
7744                                                 &sinfo->block_groups[i]);
7745
7746         spin_unlock(&sinfo->lock);
7747
7748         return free_bytes;
7749 }
7750
7751 void btrfs_set_block_group_rw(struct btrfs_root *root,
7752                               struct btrfs_block_group_cache *cache)
7753 {
7754         struct btrfs_space_info *sinfo = cache->space_info;
7755         u64 num_bytes;
7756
7757         BUG_ON(!cache->ro);
7758
7759         spin_lock(&sinfo->lock);
7760         spin_lock(&cache->lock);
7761         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7762                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7763         sinfo->bytes_readonly -= num_bytes;
7764         cache->ro = 0;
7765         spin_unlock(&cache->lock);
7766         spin_unlock(&sinfo->lock);
7767 }
7768
7769 /*
7770  * checks to see if its even possible to relocate this block group.
7771  *
7772  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7773  * ok to go ahead and try.
7774  */
7775 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7776 {
7777         struct btrfs_block_group_cache *block_group;
7778         struct btrfs_space_info *space_info;
7779         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7780         struct btrfs_device *device;
7781         u64 min_free;
7782         u64 dev_min = 1;
7783         u64 dev_nr = 0;
7784         u64 target;
7785         int index;
7786         int full = 0;
7787         int ret = 0;
7788
7789         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7790
7791         /* odd, couldn't find the block group, leave it alone */
7792         if (!block_group)
7793                 return -1;
7794
7795         min_free = btrfs_block_group_used(&block_group->item);
7796
7797         /* no bytes used, we're good */
7798         if (!min_free)
7799                 goto out;
7800
7801         space_info = block_group->space_info;
7802         spin_lock(&space_info->lock);
7803
7804         full = space_info->full;
7805
7806         /*
7807          * if this is the last block group we have in this space, we can't
7808          * relocate it unless we're able to allocate a new chunk below.
7809          *
7810          * Otherwise, we need to make sure we have room in the space to handle
7811          * all of the extents from this block group.  If we can, we're good
7812          */
7813         if ((space_info->total_bytes != block_group->key.offset) &&
7814             (space_info->bytes_used + space_info->bytes_reserved +
7815              space_info->bytes_pinned + space_info->bytes_readonly +
7816              min_free < space_info->total_bytes)) {
7817                 spin_unlock(&space_info->lock);
7818                 goto out;
7819         }
7820         spin_unlock(&space_info->lock);
7821
7822         /*
7823          * ok we don't have enough space, but maybe we have free space on our
7824          * devices to allocate new chunks for relocation, so loop through our
7825          * alloc devices and guess if we have enough space.  if this block
7826          * group is going to be restriped, run checks against the target
7827          * profile instead of the current one.
7828          */
7829         ret = -1;
7830
7831         /*
7832          * index:
7833          *      0: raid10
7834          *      1: raid1
7835          *      2: dup
7836          *      3: raid0
7837          *      4: single
7838          */
7839         target = get_restripe_target(root->fs_info, block_group->flags);
7840         if (target) {
7841                 index = __get_raid_index(extended_to_chunk(target));
7842         } else {
7843                 /*
7844                  * this is just a balance, so if we were marked as full
7845                  * we know there is no space for a new chunk
7846                  */
7847                 if (full)
7848                         goto out;
7849
7850                 index = get_block_group_index(block_group);
7851         }
7852
7853         if (index == BTRFS_RAID_RAID10) {
7854                 dev_min = 4;
7855                 /* Divide by 2 */
7856                 min_free >>= 1;
7857         } else if (index == BTRFS_RAID_RAID1) {
7858                 dev_min = 2;
7859         } else if (index == BTRFS_RAID_DUP) {
7860                 /* Multiply by 2 */
7861                 min_free <<= 1;
7862         } else if (index == BTRFS_RAID_RAID0) {
7863                 dev_min = fs_devices->rw_devices;
7864                 do_div(min_free, dev_min);
7865         }
7866
7867         mutex_lock(&root->fs_info->chunk_mutex);
7868         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7869                 u64 dev_offset;
7870
7871                 /*
7872                  * check to make sure we can actually find a chunk with enough
7873                  * space to fit our block group in.
7874                  */
7875                 if (device->total_bytes > device->bytes_used + min_free &&
7876                     !device->is_tgtdev_for_dev_replace) {
7877                         ret = find_free_dev_extent(device, min_free,
7878                                                    &dev_offset, NULL);
7879                         if (!ret)
7880                                 dev_nr++;
7881
7882                         if (dev_nr >= dev_min)
7883                                 break;
7884
7885                         ret = -1;
7886                 }
7887         }
7888         mutex_unlock(&root->fs_info->chunk_mutex);
7889 out:
7890         btrfs_put_block_group(block_group);
7891         return ret;
7892 }
7893
7894 static int find_first_block_group(struct btrfs_root *root,
7895                 struct btrfs_path *path, struct btrfs_key *key)
7896 {
7897         int ret = 0;
7898         struct btrfs_key found_key;
7899         struct extent_buffer *leaf;
7900         int slot;
7901
7902         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7903         if (ret < 0)
7904                 goto out;
7905
7906         while (1) {
7907                 slot = path->slots[0];
7908                 leaf = path->nodes[0];
7909                 if (slot >= btrfs_header_nritems(leaf)) {
7910                         ret = btrfs_next_leaf(root, path);
7911                         if (ret == 0)
7912                                 continue;
7913                         if (ret < 0)
7914                                 goto out;
7915                         break;
7916                 }
7917                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7918
7919                 if (found_key.objectid >= key->objectid &&
7920                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7921                         ret = 0;
7922                         goto out;
7923                 }
7924                 path->slots[0]++;
7925         }
7926 out:
7927         return ret;
7928 }
7929
7930 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7931 {
7932         struct btrfs_block_group_cache *block_group;
7933         u64 last = 0;
7934
7935         while (1) {
7936                 struct inode *inode;
7937
7938                 block_group = btrfs_lookup_first_block_group(info, last);
7939                 while (block_group) {
7940                         spin_lock(&block_group->lock);
7941                         if (block_group->iref)
7942                                 break;
7943                         spin_unlock(&block_group->lock);
7944                         block_group = next_block_group(info->tree_root,
7945                                                        block_group);
7946                 }
7947                 if (!block_group) {
7948                         if (last == 0)
7949                                 break;
7950                         last = 0;
7951                         continue;
7952                 }
7953
7954                 inode = block_group->inode;
7955                 block_group->iref = 0;
7956                 block_group->inode = NULL;
7957                 spin_unlock(&block_group->lock);
7958                 iput(inode);
7959                 last = block_group->key.objectid + block_group->key.offset;
7960                 btrfs_put_block_group(block_group);
7961         }
7962 }
7963
7964 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7965 {
7966         struct btrfs_block_group_cache *block_group;
7967         struct btrfs_space_info *space_info;
7968         struct btrfs_caching_control *caching_ctl;
7969         struct rb_node *n;
7970
7971         down_write(&info->extent_commit_sem);
7972         while (!list_empty(&info->caching_block_groups)) {
7973                 caching_ctl = list_entry(info->caching_block_groups.next,
7974                                          struct btrfs_caching_control, list);
7975                 list_del(&caching_ctl->list);
7976                 put_caching_control(caching_ctl);
7977         }
7978         up_write(&info->extent_commit_sem);
7979
7980         spin_lock(&info->block_group_cache_lock);
7981         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7982                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7983                                        cache_node);
7984                 rb_erase(&block_group->cache_node,
7985                          &info->block_group_cache_tree);
7986                 spin_unlock(&info->block_group_cache_lock);
7987
7988                 down_write(&block_group->space_info->groups_sem);
7989                 list_del(&block_group->list);
7990                 up_write(&block_group->space_info->groups_sem);
7991
7992                 if (block_group->cached == BTRFS_CACHE_STARTED)
7993                         wait_block_group_cache_done(block_group);
7994
7995                 /*
7996                  * We haven't cached this block group, which means we could
7997                  * possibly have excluded extents on this block group.
7998                  */
7999                 if (block_group->cached == BTRFS_CACHE_NO)
8000                         free_excluded_extents(info->extent_root, block_group);
8001
8002                 btrfs_remove_free_space_cache(block_group);
8003                 btrfs_put_block_group(block_group);
8004
8005                 spin_lock(&info->block_group_cache_lock);
8006         }
8007         spin_unlock(&info->block_group_cache_lock);
8008
8009         /* now that all the block groups are freed, go through and
8010          * free all the space_info structs.  This is only called during
8011          * the final stages of unmount, and so we know nobody is
8012          * using them.  We call synchronize_rcu() once before we start,
8013          * just to be on the safe side.
8014          */
8015         synchronize_rcu();
8016
8017         release_global_block_rsv(info);
8018
8019         while(!list_empty(&info->space_info)) {
8020                 space_info = list_entry(info->space_info.next,
8021                                         struct btrfs_space_info,
8022                                         list);
8023                 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8024                         if (space_info->bytes_pinned > 0 ||
8025                             space_info->bytes_reserved > 0 ||
8026                             space_info->bytes_may_use > 0) {
8027                                 WARN_ON(1);
8028                                 dump_space_info(space_info, 0, 0);
8029                         }
8030                 }
8031                 list_del(&space_info->list);
8032                 kfree(space_info);
8033         }
8034         return 0;
8035 }
8036
8037 static void __link_block_group(struct btrfs_space_info *space_info,
8038                                struct btrfs_block_group_cache *cache)
8039 {
8040         int index = get_block_group_index(cache);
8041
8042         down_write(&space_info->groups_sem);
8043         list_add_tail(&cache->list, &space_info->block_groups[index]);
8044         up_write(&space_info->groups_sem);
8045 }
8046
8047 int btrfs_read_block_groups(struct btrfs_root *root)
8048 {
8049         struct btrfs_path *path;
8050         int ret;
8051         struct btrfs_block_group_cache *cache;
8052         struct btrfs_fs_info *info = root->fs_info;
8053         struct btrfs_space_info *space_info;
8054         struct btrfs_key key;
8055         struct btrfs_key found_key;
8056         struct extent_buffer *leaf;
8057         int need_clear = 0;
8058         u64 cache_gen;
8059
8060         root = info->extent_root;
8061         key.objectid = 0;
8062         key.offset = 0;
8063         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8064         path = btrfs_alloc_path();
8065         if (!path)
8066                 return -ENOMEM;
8067         path->reada = 1;
8068
8069         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8070         if (btrfs_test_opt(root, SPACE_CACHE) &&
8071             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8072                 need_clear = 1;
8073         if (btrfs_test_opt(root, CLEAR_CACHE))
8074                 need_clear = 1;
8075
8076         while (1) {
8077                 ret = find_first_block_group(root, path, &key);
8078                 if (ret > 0)
8079                         break;
8080                 if (ret != 0)
8081                         goto error;
8082                 leaf = path->nodes[0];
8083                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8084                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8085                 if (!cache) {
8086                         ret = -ENOMEM;
8087                         goto error;
8088                 }
8089                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8090                                                 GFP_NOFS);
8091                 if (!cache->free_space_ctl) {
8092                         kfree(cache);
8093                         ret = -ENOMEM;
8094                         goto error;
8095                 }
8096
8097                 atomic_set(&cache->count, 1);
8098                 spin_lock_init(&cache->lock);
8099                 cache->fs_info = info;
8100                 INIT_LIST_HEAD(&cache->list);
8101                 INIT_LIST_HEAD(&cache->cluster_list);
8102
8103                 if (need_clear) {
8104                         /*
8105                          * When we mount with old space cache, we need to
8106                          * set BTRFS_DC_CLEAR and set dirty flag.
8107                          *
8108                          * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8109                          *    truncate the old free space cache inode and
8110                          *    setup a new one.
8111                          * b) Setting 'dirty flag' makes sure that we flush
8112                          *    the new space cache info onto disk.
8113                          */
8114                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8115                         if (btrfs_test_opt(root, SPACE_CACHE))
8116                                 cache->dirty = 1;
8117                 }
8118
8119                 read_extent_buffer(leaf, &cache->item,
8120                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8121                                    sizeof(cache->item));
8122                 memcpy(&cache->key, &found_key, sizeof(found_key));
8123
8124                 key.objectid = found_key.objectid + found_key.offset;
8125                 btrfs_release_path(path);
8126                 cache->flags = btrfs_block_group_flags(&cache->item);
8127                 cache->sectorsize = root->sectorsize;
8128                 cache->full_stripe_len = btrfs_full_stripe_len(root,
8129                                                &root->fs_info->mapping_tree,
8130                                                found_key.objectid);
8131                 btrfs_init_free_space_ctl(cache);
8132
8133                 /*
8134                  * We need to exclude the super stripes now so that the space
8135                  * info has super bytes accounted for, otherwise we'll think
8136                  * we have more space than we actually do.
8137                  */
8138                 ret = exclude_super_stripes(root, cache);
8139                 if (ret) {
8140                         /*
8141                          * We may have excluded something, so call this just in
8142                          * case.
8143                          */
8144                         free_excluded_extents(root, cache);
8145                         kfree(cache->free_space_ctl);
8146                         kfree(cache);
8147                         goto error;
8148                 }
8149
8150                 /*
8151                  * check for two cases, either we are full, and therefore
8152                  * don't need to bother with the caching work since we won't
8153                  * find any space, or we are empty, and we can just add all
8154                  * the space in and be done with it.  This saves us _alot_ of
8155                  * time, particularly in the full case.
8156                  */
8157                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8158                         cache->last_byte_to_unpin = (u64)-1;
8159                         cache->cached = BTRFS_CACHE_FINISHED;
8160                         free_excluded_extents(root, cache);
8161                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8162                         cache->last_byte_to_unpin = (u64)-1;
8163                         cache->cached = BTRFS_CACHE_FINISHED;
8164                         add_new_free_space(cache, root->fs_info,
8165                                            found_key.objectid,
8166                                            found_key.objectid +
8167                                            found_key.offset);
8168                         free_excluded_extents(root, cache);
8169                 }
8170
8171                 ret = update_space_info(info, cache->flags, found_key.offset,
8172                                         btrfs_block_group_used(&cache->item),
8173                                         &space_info);
8174                 BUG_ON(ret); /* -ENOMEM */
8175                 cache->space_info = space_info;
8176                 spin_lock(&cache->space_info->lock);
8177                 cache->space_info->bytes_readonly += cache->bytes_super;
8178                 spin_unlock(&cache->space_info->lock);
8179
8180                 __link_block_group(space_info, cache);
8181
8182                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8183                 BUG_ON(ret); /* Logic error */
8184
8185                 set_avail_alloc_bits(root->fs_info, cache->flags);
8186                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8187                         set_block_group_ro(cache, 1);
8188         }
8189
8190         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8191                 if (!(get_alloc_profile(root, space_info->flags) &
8192                       (BTRFS_BLOCK_GROUP_RAID10 |
8193                        BTRFS_BLOCK_GROUP_RAID1 |
8194                        BTRFS_BLOCK_GROUP_RAID5 |
8195                        BTRFS_BLOCK_GROUP_RAID6 |
8196                        BTRFS_BLOCK_GROUP_DUP)))
8197                         continue;
8198                 /*
8199                  * avoid allocating from un-mirrored block group if there are
8200                  * mirrored block groups.
8201                  */
8202                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8203                         set_block_group_ro(cache, 1);
8204                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8205                         set_block_group_ro(cache, 1);
8206         }
8207
8208         init_global_block_rsv(info);
8209         ret = 0;
8210 error:
8211         btrfs_free_path(path);
8212         return ret;
8213 }
8214
8215 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8216                                        struct btrfs_root *root)
8217 {
8218         struct btrfs_block_group_cache *block_group, *tmp;
8219         struct btrfs_root *extent_root = root->fs_info->extent_root;
8220         struct btrfs_block_group_item item;
8221         struct btrfs_key key;
8222         int ret = 0;
8223
8224         list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8225                                  new_bg_list) {
8226                 list_del_init(&block_group->new_bg_list);
8227
8228                 if (ret)
8229                         continue;
8230
8231                 spin_lock(&block_group->lock);
8232                 memcpy(&item, &block_group->item, sizeof(item));
8233                 memcpy(&key, &block_group->key, sizeof(key));
8234                 spin_unlock(&block_group->lock);
8235
8236                 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8237                                         sizeof(item));
8238                 if (ret)
8239                         btrfs_abort_transaction(trans, extent_root, ret);
8240         }
8241 }
8242
8243 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8244                            struct btrfs_root *root, u64 bytes_used,
8245                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8246                            u64 size)
8247 {
8248         int ret;
8249         struct btrfs_root *extent_root;
8250         struct btrfs_block_group_cache *cache;
8251
8252         extent_root = root->fs_info->extent_root;
8253
8254         root->fs_info->last_trans_log_full_commit = trans->transid;
8255
8256         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8257         if (!cache)
8258                 return -ENOMEM;
8259         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8260                                         GFP_NOFS);
8261         if (!cache->free_space_ctl) {
8262                 kfree(cache);
8263                 return -ENOMEM;
8264         }
8265
8266         cache->key.objectid = chunk_offset;
8267         cache->key.offset = size;
8268         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8269         cache->sectorsize = root->sectorsize;
8270         cache->fs_info = root->fs_info;
8271         cache->full_stripe_len = btrfs_full_stripe_len(root,
8272                                                &root->fs_info->mapping_tree,
8273                                                chunk_offset);
8274
8275         atomic_set(&cache->count, 1);
8276         spin_lock_init(&cache->lock);
8277         INIT_LIST_HEAD(&cache->list);
8278         INIT_LIST_HEAD(&cache->cluster_list);
8279         INIT_LIST_HEAD(&cache->new_bg_list);
8280
8281         btrfs_init_free_space_ctl(cache);
8282
8283         btrfs_set_block_group_used(&cache->item, bytes_used);
8284         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8285         cache->flags = type;
8286         btrfs_set_block_group_flags(&cache->item, type);
8287
8288         cache->last_byte_to_unpin = (u64)-1;
8289         cache->cached = BTRFS_CACHE_FINISHED;
8290         ret = exclude_super_stripes(root, cache);
8291         if (ret) {
8292                 /*
8293                  * We may have excluded something, so call this just in
8294                  * case.
8295                  */
8296                 free_excluded_extents(root, cache);
8297                 kfree(cache->free_space_ctl);
8298                 kfree(cache);
8299                 return ret;
8300         }
8301
8302         add_new_free_space(cache, root->fs_info, chunk_offset,
8303                            chunk_offset + size);
8304
8305         free_excluded_extents(root, cache);
8306
8307         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8308                                 &cache->space_info);
8309         BUG_ON(ret); /* -ENOMEM */
8310         update_global_block_rsv(root->fs_info);
8311
8312         spin_lock(&cache->space_info->lock);
8313         cache->space_info->bytes_readonly += cache->bytes_super;
8314         spin_unlock(&cache->space_info->lock);
8315
8316         __link_block_group(cache->space_info, cache);
8317
8318         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8319         BUG_ON(ret); /* Logic error */
8320
8321         list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8322
8323         set_avail_alloc_bits(extent_root->fs_info, type);
8324
8325         return 0;
8326 }
8327
8328 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8329 {
8330         u64 extra_flags = chunk_to_extended(flags) &
8331                                 BTRFS_EXTENDED_PROFILE_MASK;
8332
8333         write_seqlock(&fs_info->profiles_lock);
8334         if (flags & BTRFS_BLOCK_GROUP_DATA)
8335                 fs_info->avail_data_alloc_bits &= ~extra_flags;
8336         if (flags & BTRFS_BLOCK_GROUP_METADATA)
8337                 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8338         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8339                 fs_info->avail_system_alloc_bits &= ~extra_flags;
8340         write_sequnlock(&fs_info->profiles_lock);
8341 }
8342
8343 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8344                              struct btrfs_root *root, u64 group_start)
8345 {
8346         struct btrfs_path *path;
8347         struct btrfs_block_group_cache *block_group;
8348         struct btrfs_free_cluster *cluster;
8349         struct btrfs_root *tree_root = root->fs_info->tree_root;
8350         struct btrfs_key key;
8351         struct inode *inode;
8352         int ret;
8353         int index;
8354         int factor;
8355
8356         root = root->fs_info->extent_root;
8357
8358         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8359         BUG_ON(!block_group);
8360         BUG_ON(!block_group->ro);
8361
8362         /*
8363          * Free the reserved super bytes from this block group before
8364          * remove it.
8365          */
8366         free_excluded_extents(root, block_group);
8367
8368         memcpy(&key, &block_group->key, sizeof(key));
8369         index = get_block_group_index(block_group);
8370         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8371                                   BTRFS_BLOCK_GROUP_RAID1 |
8372                                   BTRFS_BLOCK_GROUP_RAID10))
8373                 factor = 2;
8374         else
8375                 factor = 1;
8376
8377         /* make sure this block group isn't part of an allocation cluster */
8378         cluster = &root->fs_info->data_alloc_cluster;
8379         spin_lock(&cluster->refill_lock);
8380         btrfs_return_cluster_to_free_space(block_group, cluster);
8381         spin_unlock(&cluster->refill_lock);
8382
8383         /*
8384          * make sure this block group isn't part of a metadata
8385          * allocation cluster
8386          */
8387         cluster = &root->fs_info->meta_alloc_cluster;
8388         spin_lock(&cluster->refill_lock);
8389         btrfs_return_cluster_to_free_space(block_group, cluster);
8390         spin_unlock(&cluster->refill_lock);
8391
8392         path = btrfs_alloc_path();
8393         if (!path) {
8394                 ret = -ENOMEM;
8395                 goto out;
8396         }
8397
8398         inode = lookup_free_space_inode(tree_root, block_group, path);
8399         if (!IS_ERR(inode)) {
8400                 ret = btrfs_orphan_add(trans, inode);
8401                 if (ret) {
8402                         btrfs_add_delayed_iput(inode);
8403                         goto out;
8404                 }
8405                 clear_nlink(inode);
8406                 /* One for the block groups ref */
8407                 spin_lock(&block_group->lock);
8408                 if (block_group->iref) {
8409                         block_group->iref = 0;
8410                         block_group->inode = NULL;
8411                         spin_unlock(&block_group->lock);
8412                         iput(inode);
8413                 } else {
8414                         spin_unlock(&block_group->lock);
8415                 }
8416                 /* One for our lookup ref */
8417                 btrfs_add_delayed_iput(inode);
8418         }
8419
8420         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8421         key.offset = block_group->key.objectid;
8422         key.type = 0;
8423
8424         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8425         if (ret < 0)
8426                 goto out;
8427         if (ret > 0)
8428                 btrfs_release_path(path);
8429         if (ret == 0) {
8430                 ret = btrfs_del_item(trans, tree_root, path);
8431                 if (ret)
8432                         goto out;
8433                 btrfs_release_path(path);
8434         }
8435
8436         spin_lock(&root->fs_info->block_group_cache_lock);
8437         rb_erase(&block_group->cache_node,
8438                  &root->fs_info->block_group_cache_tree);
8439
8440         if (root->fs_info->first_logical_byte == block_group->key.objectid)
8441                 root->fs_info->first_logical_byte = (u64)-1;
8442         spin_unlock(&root->fs_info->block_group_cache_lock);
8443
8444         down_write(&block_group->space_info->groups_sem);
8445         /*
8446          * we must use list_del_init so people can check to see if they
8447          * are still on the list after taking the semaphore
8448          */
8449         list_del_init(&block_group->list);
8450         if (list_empty(&block_group->space_info->block_groups[index]))
8451                 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8452         up_write(&block_group->space_info->groups_sem);
8453
8454         if (block_group->cached == BTRFS_CACHE_STARTED)
8455                 wait_block_group_cache_done(block_group);
8456
8457         btrfs_remove_free_space_cache(block_group);
8458
8459         spin_lock(&block_group->space_info->lock);
8460         block_group->space_info->total_bytes -= block_group->key.offset;
8461         block_group->space_info->bytes_readonly -= block_group->key.offset;
8462         block_group->space_info->disk_total -= block_group->key.offset * factor;
8463         spin_unlock(&block_group->space_info->lock);
8464
8465         memcpy(&key, &block_group->key, sizeof(key));
8466
8467         btrfs_clear_space_info_full(root->fs_info);
8468
8469         btrfs_put_block_group(block_group);
8470         btrfs_put_block_group(block_group);
8471
8472         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8473         if (ret > 0)
8474                 ret = -EIO;
8475         if (ret < 0)
8476                 goto out;
8477
8478         ret = btrfs_del_item(trans, root, path);
8479 out:
8480         btrfs_free_path(path);
8481         return ret;
8482 }
8483
8484 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8485 {
8486         struct btrfs_space_info *space_info;
8487         struct btrfs_super_block *disk_super;
8488         u64 features;
8489         u64 flags;
8490         int mixed = 0;
8491         int ret;
8492
8493         disk_super = fs_info->super_copy;
8494         if (!btrfs_super_root(disk_super))
8495                 return 1;
8496
8497         features = btrfs_super_incompat_flags(disk_super);
8498         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8499                 mixed = 1;
8500
8501         flags = BTRFS_BLOCK_GROUP_SYSTEM;
8502         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8503         if (ret)
8504                 goto out;
8505
8506         if (mixed) {
8507                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8508                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8509         } else {
8510                 flags = BTRFS_BLOCK_GROUP_METADATA;
8511                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8512                 if (ret)
8513                         goto out;
8514
8515                 flags = BTRFS_BLOCK_GROUP_DATA;
8516                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8517         }
8518 out:
8519         return ret;
8520 }
8521
8522 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8523 {
8524         return unpin_extent_range(root, start, end);
8525 }
8526
8527 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8528                                u64 num_bytes, u64 *actual_bytes)
8529 {
8530         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8531 }
8532
8533 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8534 {
8535         struct btrfs_fs_info *fs_info = root->fs_info;
8536         struct btrfs_block_group_cache *cache = NULL;
8537         u64 group_trimmed;
8538         u64 start;
8539         u64 end;
8540         u64 trimmed = 0;
8541         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8542         int ret = 0;
8543
8544         /*
8545          * try to trim all FS space, our block group may start from non-zero.
8546          */
8547         if (range->len == total_bytes)
8548                 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8549         else
8550                 cache = btrfs_lookup_block_group(fs_info, range->start);
8551
8552         while (cache) {
8553                 if (cache->key.objectid >= (range->start + range->len)) {
8554                         btrfs_put_block_group(cache);
8555                         break;
8556                 }
8557
8558                 start = max(range->start, cache->key.objectid);
8559                 end = min(range->start + range->len,
8560                                 cache->key.objectid + cache->key.offset);
8561
8562                 if (end - start >= range->minlen) {
8563                         if (!block_group_cache_done(cache)) {
8564                                 ret = cache_block_group(cache, 0);
8565                                 if (!ret)
8566                                         wait_block_group_cache_done(cache);
8567                         }
8568                         ret = btrfs_trim_block_group(cache,
8569                                                      &group_trimmed,
8570                                                      start,
8571                                                      end,
8572                                                      range->minlen);
8573
8574                         trimmed += group_trimmed;
8575                         if (ret) {
8576                                 btrfs_put_block_group(cache);
8577                                 break;
8578                         }
8579                 }
8580
8581                 cache = next_block_group(fs_info->tree_root, cache);
8582         }
8583
8584         range->len = trimmed;
8585         return ret;
8586 }