fs/btrfs/extent-tree.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/sched.h>
   7 #include <linux/sched/signal.h>
   8 #include <linux/pagemap.h>
   9 #include <linux/writeback.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/sort.h>
  12 #include <linux/rcupdate.h>
  13 #include <linux/kthread.h>
  14 #include <linux/slab.h>
  15 #include <linux/ratelimit.h>
  16 #include <linux/percpu_counter.h>
  17 #include <linux/lockdep.h>
  18 #include <linux/crc32c.h>
  19 #include "misc.h"
  20 #include "tree-log.h"
  21 #include "disk-io.h"
  22 #include "print-tree.h"
  23 #include "volumes.h"
  24 #include "raid56.h"
  25 #include "locking.h"
  26 #include "free-space-cache.h"
  27 #include "free-space-tree.h"
  28 #include "sysfs.h"
  29 #include "qgroup.h"
  30 #include "ref-verify.h"
  31 #include "space-info.h"
  32 #include "block-rsv.h"
  33 #include "delalloc-space.h"
  34 #include "block-group.h"
  35
  36 #undef SCRAMBLE_DELAYED_REFS
  37
  38
  39 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  40                                struct btrfs_delayed_ref_node *node, u64 parent,
  41                                u64 root_objectid, u64 owner_objectid,
  42                                u64 owner_offset, int refs_to_drop,
  43                                struct btrfs_delayed_extent_op *extra_op);
  44 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  45                                     struct extent_buffer *leaf,
  46                                     struct btrfs_extent_item *ei);
  47 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  48                                       u64 parent, u64 root_objectid,
  49                                       u64 flags, u64 owner, u64 offset,
  50                                       struct btrfs_key *ins, int ref_mod);
  51 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  52                                      struct btrfs_delayed_ref_node *node,
  53                                      struct btrfs_delayed_extent_op *extent_op);
  54 static int find_next_key(struct btrfs_path *path, int level,
  55                          struct btrfs_key *key);
  56
  57 static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
  58 {
  59         return (cache->flags & bits) == bits;
  60 }
  61
  62 int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info,
  63                               u64 start, u64 num_bytes)
  64 {
  65         u64 end = start + num_bytes - 1;
  66         set_extent_bits(&fs_info->freed_extents[0],
  67                         start, end, EXTENT_UPTODATE);
  68         set_extent_bits(&fs_info->freed_extents[1],
  69                         start, end, EXTENT_UPTODATE);
  70         return 0;
  71 }
  72
  73 void btrfs_free_excluded_extents(struct btrfs_block_group *cache)
  74 {
  75         struct btrfs_fs_info *fs_info = cache->fs_info;
  76         u64 start, end;
  77
  78         start = cache->start;
  79         end = start + cache->length - 1;
  80
  81         clear_extent_bits(&fs_info->freed_extents[0],
  82                           start, end, EXTENT_UPTODATE);
  83         clear_extent_bits(&fs_info->freed_extents[1],
  84                           start, end, EXTENT_UPTODATE);
  85 }
  86
  87 static u64 generic_ref_to_space_flags(struct btrfs_ref *ref)
  88 {
  89         if (ref->type == BTRFS_REF_METADATA) {
  90                 if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID)
  91                         return BTRFS_BLOCK_GROUP_SYSTEM;
  92                 else
  93                         return BTRFS_BLOCK_GROUP_METADATA;
  94         }
  95         return BTRFS_BLOCK_GROUP_DATA;
  96 }
  97
  98 static void add_pinned_bytes(struct btrfs_fs_info *fs_info,
  99                              struct btrfs_ref *ref)
 100 {
 101         struct btrfs_space_info *space_info;
 102         u64 flags = generic_ref_to_space_flags(ref);
 103
 104         space_info = btrfs_find_space_info(fs_info, flags);
 105         ASSERT(space_info);
 106         percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len,
 107                     BTRFS_TOTAL_BYTES_PINNED_BATCH);
 108 }
 109
 110 static void sub_pinned_bytes(struct btrfs_fs_info *fs_info,
 111                              struct btrfs_ref *ref)
 112 {
 113         struct btrfs_space_info *space_info;
 114         u64 flags = generic_ref_to_space_flags(ref);
 115
 116         space_info = btrfs_find_space_info(fs_info, flags);
 117         ASSERT(space_info);
 118         percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len,
 119                     BTRFS_TOTAL_BYTES_PINNED_BATCH);
 120 }
 121
 122 /* simple helper to search for an existing data extent at a given offset */
 123 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
 124 {
 125         int ret;
 126         struct btrfs_key key;
 127         struct btrfs_path *path;
 128
 129         path = btrfs_alloc_path();
 130         if (!path)
 131                 return -ENOMEM;
 132
 133         key.objectid = start;
 134         key.offset = len;
 135         key.type = BTRFS_EXTENT_ITEM_KEY;
 136         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
 137         btrfs_free_path(path);
 138         return ret;
 139 }
 140
 141 /*
 142  * helper function to lookup reference count and flags of a tree block.
 143  *
 144  * the head node for delayed ref is used to store the sum of all the
 145  * reference count modifications queued up in the rbtree. the head
 146  * node may also store the extent flags to set. This way you can check
 147  * to see what the reference count and extent flags would be if all of
 148  * the delayed refs are not processed.
 149  */
 150 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 151                              struct btrfs_fs_info *fs_info, u64 bytenr,
 152                              u64 offset, int metadata, u64 *refs, u64 *flags)
 153 {
 154         struct btrfs_delayed_ref_head *head;
 155         struct btrfs_delayed_ref_root *delayed_refs;
 156         struct btrfs_path *path;
 157         struct btrfs_extent_item *ei;
 158         struct extent_buffer *leaf;
 159         struct btrfs_key key;
 160         u32 item_size;
 161         u64 num_refs;
 162         u64 extent_flags;
 163         int ret;
 164
 165         /*
 166          * If we don't have skinny metadata, don't bother doing anything
 167          * different
 168          */
 169         if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
 170                 offset = fs_info->nodesize;
 171                 metadata = 0;
 172         }
 173
 174         path = btrfs_alloc_path();
 175         if (!path)
 176                 return -ENOMEM;
 177
 178         if (!trans) {
 179                 path->skip_locking = 1;
 180                 path->search_commit_root = 1;
 181         }
 182
 183 search_again:
 184         key.objectid = bytenr;
 185         key.offset = offset;
 186         if (metadata)
 187                 key.type = BTRFS_METADATA_ITEM_KEY;
 188         else
 189                 key.type = BTRFS_EXTENT_ITEM_KEY;
 190
 191         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
 192         if (ret < 0)
 193                 goto out_free;
 194
 195         if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
 196                 if (path->slots[0]) {
 197                         path->slots[0]--;
 198                         btrfs_item_key_to_cpu(path->nodes[0], &key,
 199                                               path->slots[0]);
 200                         if (key.objectid == bytenr &&
 201                             key.type == BTRFS_EXTENT_ITEM_KEY &&
 202                             key.offset == fs_info->nodesize)
 203                                 ret = 0;
 204                 }
 205         }
 206
 207         if (ret == 0) {
 208                 leaf = path->nodes[0];
 209                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 210                 if (item_size >= sizeof(*ei)) {
 211                         ei = btrfs_item_ptr(leaf, path->slots[0],
 212                                             struct btrfs_extent_item);
 213                         num_refs = btrfs_extent_refs(leaf, ei);
 214                         extent_flags = btrfs_extent_flags(leaf, ei);
 215                 } else {
 216                         ret = -EINVAL;
 217                         btrfs_print_v0_err(fs_info);
 218                         if (trans)
 219                                 btrfs_abort_transaction(trans, ret);
 220                         else
 221                                 btrfs_handle_fs_error(fs_info, ret, NULL);
 222
 223                         goto out_free;
 224                 }
 225
 226                 BUG_ON(num_refs == 0);
 227         } else {
 228                 num_refs = 0;
 229                 extent_flags = 0;
 230                 ret = 0;
 231         }
 232
 233         if (!trans)
 234                 goto out;
 235
 236         delayed_refs = &trans->transaction->delayed_refs;
 237         spin_lock(&delayed_refs->lock);
 238         head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
 239         if (head) {
 240                 if (!mutex_trylock(&head->mutex)) {
 241                         refcount_inc(&head->refs);
 242                         spin_unlock(&delayed_refs->lock);
 243
 244                         btrfs_release_path(path);
 245
 246                         /*
 247                          * Mutex was contended, block until it's released and try
 248                          * again
 249                          */
 250                         mutex_lock(&head->mutex);
 251                         mutex_unlock(&head->mutex);
 252                         btrfs_put_delayed_ref_head(head);
 253                         goto search_again;
 254                 }
 255                 spin_lock(&head->lock);
 256                 if (head->extent_op && head->extent_op->update_flags)
 257                         extent_flags |= head->extent_op->flags_to_set;
 258                 else
 259                         BUG_ON(num_refs == 0);
 260
 261                 num_refs += head->ref_mod;
 262                 spin_unlock(&head->lock);
 263                 mutex_unlock(&head->mutex);
 264         }
 265         spin_unlock(&delayed_refs->lock);
 266 out:
 267         WARN_ON(num_refs == 0);
 268         if (refs)
 269                 *refs = num_refs;
 270         if (flags)
 271                 *flags = extent_flags;
 272 out_free:
 273         btrfs_free_path(path);
 274         return ret;
 275 }
 276
 277 /*
 278  * Back reference rules.  Back refs have three main goals:
 279  *
 280  * 1) differentiate between all holders of references to an extent so that
 281  *    when a reference is dropped we can make sure it was a valid reference
 282  *    before freeing the extent.
 283  *
 284  * 2) Provide enough information to quickly find the holders of an extent
 285  *    if we notice a given block is corrupted or bad.
 286  *
 287  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 288  *    maintenance.  This is actually the same as #2, but with a slightly
 289  *    different use case.
 290  *
 291  * There are two kinds of back refs. The implicit back refs is optimized
 292  * for pointers in non-shared tree blocks. For a given pointer in a block,
 293  * back refs of this kind provide information about the block's owner tree
 294  * and the pointer's key. These information allow us to find the block by
 295  * b-tree searching. The full back refs is for pointers in tree blocks not
 296  * referenced by their owner trees. The location of tree block is recorded
 297  * in the back refs. Actually the full back refs is generic, and can be
 298  * used in all cases the implicit back refs is used. The major shortcoming
 299  * of the full back refs is its overhead. Every time a tree block gets
 300  * COWed, we have to update back refs entry for all pointers in it.
 301  *
 302  * For a newly allocated tree block, we use implicit back refs for
 303  * pointers in it. This means most tree related operations only involve
 304  * implicit back refs. For a tree block created in old transaction, the
 305  * only way to drop a reference to it is COW it. So we can detect the
 306  * event that tree block loses its owner tree's reference and do the
 307  * back refs conversion.
 308  *
 309  * When a tree block is COWed through a tree, there are four cases:
 310  *
 311  * The reference count of the block is one and the tree is the block's
 312  * owner tree. Nothing to do in this case.
 313  *
 314  * The reference count of the block is one and the tree is not the
 315  * block's owner tree. In this case, full back refs is used for pointers
 316  * in the block. Remove these full back refs, add implicit back refs for
 317  * every pointers in the new block.
 318  *
 319  * The reference count of the block is greater than one and the tree is
 320  * the block's owner tree. In this case, implicit back refs is used for
 321  * pointers in the block. Add full back refs for every pointers in the
 322  * block, increase lower level extents' reference counts. The original
 323  * implicit back refs are entailed to the new block.
 324  *
 325  * The reference count of the block is greater than one and the tree is
 326  * not the block's owner tree. Add implicit back refs for every pointer in
 327  * the new block, increase lower level extents' reference count.
 328  *
 329  * Back Reference Key composing:
 330  *
 331  * The key objectid corresponds to the first byte in the extent,
 332  * The key type is used to differentiate between types of back refs.
 333  * There are different meanings of the key offset for different types
 334  * of back refs.
 335  *
 336  * File extents can be referenced by:
 337  *
 338  * - multiple snapshots, subvolumes, or different generations in one subvol
 339  * - different files inside a single subvolume
 340  * - different offsets inside a file (bookend extents in file.c)
 341  *
 342  * The extent ref structure for the implicit back refs has fields for:
 343  *
 344  * - Objectid of the subvolume root
 345  * - objectid of the file holding the reference
 346  * - original offset in the file
 347  * - how many bookend extents
 348  *
 349  * The key offset for the implicit back refs is hash of the first
 350  * three fields.
 351  *
 352  * The extent ref structure for the full back refs has field for:
 353  *
 354  * - number of pointers in the tree leaf
 355  *
 356  * The key offset for the implicit back refs is the first byte of
 357  * the tree leaf
 358  *
 359  * When a file extent is allocated, The implicit back refs is used.
 360  * the fields are filled in:
 361  *
 362  *     (root_key.objectid, inode objectid, offset in file, 1)
 363  *
 364  * When a file extent is removed file truncation, we find the
 365  * corresponding implicit back refs and check the following fields:
 366  *
 367  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 368  *
 369  * Btree extents can be referenced by:
 370  *
 371  * - Different subvolumes
 372  *
 373  * Both the implicit back refs and the full back refs for tree blocks
 374  * only consist of key. The key offset for the implicit back refs is
 375  * objectid of block's owner tree. The key offset for the full back refs
 376  * is the first byte of parent block.
 377  *
 378  * When implicit back refs is used, information about the lowest key and
 379  * level of the tree block are required. These information are stored in
 380  * tree block info structure.
 381  */
 382
 383 /*
 384  * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
 385  * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
 386  * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
 387  */
 388 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
 389                                      struct btrfs_extent_inline_ref *iref,
 390                                      enum btrfs_inline_ref_type is_data)
 391 {
 392         int type = btrfs_extent_inline_ref_type(eb, iref);
 393         u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
 394
 395         if (type == BTRFS_TREE_BLOCK_REF_KEY ||
 396             type == BTRFS_SHARED_BLOCK_REF_KEY ||
 397             type == BTRFS_SHARED_DATA_REF_KEY ||
 398             type == BTRFS_EXTENT_DATA_REF_KEY) {
 399                 if (is_data == BTRFS_REF_TYPE_BLOCK) {
 400                         if (type == BTRFS_TREE_BLOCK_REF_KEY)
 401                                 return type;
 402                         if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
 403                                 ASSERT(eb->fs_info);
 404                                 /*
 405                                  * Every shared one has parent tree
 406                                  * block, which must be aligned to
 407                                  * nodesize.
 408                                  */
 409                                 if (offset &&
 410                                     IS_ALIGNED(offset, eb->fs_info->nodesize))
 411                                         return type;
 412                         }
 413                 } else if (is_data == BTRFS_REF_TYPE_DATA) {
 414                         if (type == BTRFS_EXTENT_DATA_REF_KEY)
 415                                 return type;
 416                         if (type == BTRFS_SHARED_DATA_REF_KEY) {
 417                                 ASSERT(eb->fs_info);
 418                                 /*
 419                                  * Every shared one has parent tree
 420                                  * block, which must be aligned to
 421                                  * nodesize.
 422                                  */
 423                                 if (offset &&
 424                                     IS_ALIGNED(offset, eb->fs_info->nodesize))
 425                                         return type;
 426                         }
 427                 } else {
 428                         ASSERT(is_data == BTRFS_REF_TYPE_ANY);
 429                         return type;
 430                 }
 431         }
 432
 433         btrfs_print_leaf((struct extent_buffer *)eb);
 434         btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
 435                   eb->start, type);
 436         WARN_ON(1);
 437
 438         return BTRFS_REF_TYPE_INVALID;
 439 }
 440
 441 u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 442 {
 443         u32 high_crc = ~(u32)0;
 444         u32 low_crc = ~(u32)0;
 445         __le64 lenum;
 446
 447         lenum = cpu_to_le64(root_objectid);
 448         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
 449         lenum = cpu_to_le64(owner);
 450         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 451         lenum = cpu_to_le64(offset);
 452         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
 453
 454         return ((u64)high_crc << 31) ^ (u64)low_crc;
 455 }
 456
 457 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 458                                      struct btrfs_extent_data_ref *ref)
 459 {
 460         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 461                                     btrfs_extent_data_ref_objectid(leaf, ref),
 462                                     btrfs_extent_data_ref_offset(leaf, ref));
 463 }
 464
 465 static int match_extent_data_ref(struct extent_buffer *leaf,
 466                                  struct btrfs_extent_data_ref *ref,
 467                                  u64 root_objectid, u64 owner, u64 offset)
 468 {
 469         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 470             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 471             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 472                 return 0;
 473         return 1;
 474 }
 475
 476 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 477                                            struct btrfs_path *path,
 478                                            u64 bytenr, u64 parent,
 479                                            u64 root_objectid,
 480                                            u64 owner, u64 offset)
 481 {
 482         struct btrfs_root *root = trans->fs_info->extent_root;
 483         struct btrfs_key key;
 484         struct btrfs_extent_data_ref *ref;
 485         struct extent_buffer *leaf;
 486         u32 nritems;
 487         int ret;
 488         int recow;
 489         int err = -ENOENT;
 490
 491         key.objectid = bytenr;
 492         if (parent) {
 493                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 494                 key.offset = parent;
 495         } else {
 496                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 497                 key.offset = hash_extent_data_ref(root_objectid,
 498                                                   owner, offset);
 499         }
 500 again:
 501         recow = 0;
 502         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 503         if (ret < 0) {
 504                 err = ret;
 505                 goto fail;
 506         }
 507
 508         if (parent) {
 509                 if (!ret)
 510                         return 0;
 511                 goto fail;
 512         }
 513
 514         leaf = path->nodes[0];
 515         nritems = btrfs_header_nritems(leaf);
 516         while (1) {
 517                 if (path->slots[0] >= nritems) {
 518                         ret = btrfs_next_leaf(root, path);
 519                         if (ret < 0)
 520                                 err = ret;
 521                         if (ret)
 522                                 goto fail;
 523
 524                         leaf = path->nodes[0];
 525                         nritems = btrfs_header_nritems(leaf);
 526                         recow = 1;
 527                 }
 528
 529                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 530                 if (key.objectid != bytenr ||
 531                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 532                         goto fail;
 533
 534                 ref = btrfs_item_ptr(leaf, path->slots[0],
 535                                      struct btrfs_extent_data_ref);
 536
 537                 if (match_extent_data_ref(leaf, ref, root_objectid,
 538                                           owner, offset)) {
 539                         if (recow) {
 540                                 btrfs_release_path(path);
 541                                 goto again;
 542                         }
 543                         err = 0;
 544                         break;
 545                 }
 546                 path->slots[0]++;
 547         }
 548 fail:
 549         return err;
 550 }
 551
 552 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 553                                            struct btrfs_path *path,
 554                                            u64 bytenr, u64 parent,
 555                                            u64 root_objectid, u64 owner,
 556                                            u64 offset, int refs_to_add)
 557 {
 558         struct btrfs_root *root = trans->fs_info->extent_root;
 559         struct btrfs_key key;
 560         struct extent_buffer *leaf;
 561         u32 size;
 562         u32 num_refs;
 563         int ret;
 564
 565         key.objectid = bytenr;
 566         if (parent) {
 567                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 568                 key.offset = parent;
 569                 size = sizeof(struct btrfs_shared_data_ref);
 570         } else {
 571                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 572                 key.offset = hash_extent_data_ref(root_objectid,
 573                                                   owner, offset);
 574                 size = sizeof(struct btrfs_extent_data_ref);
 575         }
 576
 577         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 578         if (ret && ret != -EEXIST)
 579                 goto fail;
 580
 581         leaf = path->nodes[0];
 582         if (parent) {
 583                 struct btrfs_shared_data_ref *ref;
 584                 ref = btrfs_item_ptr(leaf, path->slots[0],
 585                                      struct btrfs_shared_data_ref);
 586                 if (ret == 0) {
 587                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
 588                 } else {
 589                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 590                         num_refs += refs_to_add;
 591                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 592                 }
 593         } else {
 594                 struct btrfs_extent_data_ref *ref;
 595                 while (ret == -EEXIST) {
 596                         ref = btrfs_item_ptr(leaf, path->slots[0],
 597                                              struct btrfs_extent_data_ref);
 598                         if (match_extent_data_ref(leaf, ref, root_objectid,
 599                                                   owner, offset))
 600                                 break;
 601                         btrfs_release_path(path);
 602                         key.offset++;
 603                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 604                                                       size);
 605                         if (ret && ret != -EEXIST)
 606                                 goto fail;
 607
 608                         leaf = path->nodes[0];
 609                 }
 610                 ref = btrfs_item_ptr(leaf, path->slots[0],
 611                                      struct btrfs_extent_data_ref);
 612                 if (ret == 0) {
 613                         btrfs_set_extent_data_ref_root(leaf, ref,
 614                                                        root_objectid);
 615                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 616                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 617                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
 618                 } else {
 619                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
 620                         num_refs += refs_to_add;
 621                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 622                 }
 623         }
 624         btrfs_mark_buffer_dirty(leaf);
 625         ret = 0;
 626 fail:
 627         btrfs_release_path(path);
 628         return ret;
 629 }
 630
 631 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
 632                                            struct btrfs_path *path,
 633                                            int refs_to_drop, int *last_ref)
 634 {
 635         struct btrfs_key key;
 636         struct btrfs_extent_data_ref *ref1 = NULL;
 637         struct btrfs_shared_data_ref *ref2 = NULL;
 638         struct extent_buffer *leaf;
 639         u32 num_refs = 0;
 640         int ret = 0;
 641
 642         leaf = path->nodes[0];
 643         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 644
 645         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 646                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 647                                       struct btrfs_extent_data_ref);
 648                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 649         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 650                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 651                                       struct btrfs_shared_data_ref);
 652                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 653         } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
 654                 btrfs_print_v0_err(trans->fs_info);
 655                 btrfs_abort_transaction(trans, -EINVAL);
 656                 return -EINVAL;
 657         } else {
 658                 BUG();
 659         }
 660
 661         BUG_ON(num_refs < refs_to_drop);
 662         num_refs -= refs_to_drop;
 663
 664         if (num_refs == 0) {
 665                 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
 666                 *last_ref = 1;
 667         } else {
 668                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
 669                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
 670                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
 671                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
 672                 btrfs_mark_buffer_dirty(leaf);
 673         }
 674         return ret;
 675 }
 676
 677 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
 678                                           struct btrfs_extent_inline_ref *iref)
 679 {
 680         struct btrfs_key key;
 681         struct extent_buffer *leaf;
 682         struct btrfs_extent_data_ref *ref1;
 683         struct btrfs_shared_data_ref *ref2;
 684         u32 num_refs = 0;
 685         int type;
 686
 687         leaf = path->nodes[0];
 688         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 689
 690         BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
 691         if (iref) {
 692                 /*
 693                  * If type is invalid, we should have bailed out earlier than
 694                  * this call.
 695                  */
 696                 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
 697                 ASSERT(type != BTRFS_REF_TYPE_INVALID);
 698                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
 699                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
 700                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 701                 } else {
 702                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
 703                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 704                 }
 705         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 706                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 707                                       struct btrfs_extent_data_ref);
 708                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 709         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 710                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 711                                       struct btrfs_shared_data_ref);
 712                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 713         } else {
 714                 WARN_ON(1);
 715         }
 716         return num_refs;
 717 }
 718
 719 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
 720                                           struct btrfs_path *path,
 721                                           u64 bytenr, u64 parent,
 722                                           u64 root_objectid)
 723 {
 724         struct btrfs_root *root = trans->fs_info->extent_root;
 725         struct btrfs_key key;
 726         int ret;
 727
 728         key.objectid = bytenr;
 729         if (parent) {
 730                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 731                 key.offset = parent;
 732         } else {
 733                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 734                 key.offset = root_objectid;
 735         }
 736
 737         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 738         if (ret > 0)
 739                 ret = -ENOENT;
 740         return ret;
 741 }
 742
 743 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
 744                                           struct btrfs_path *path,
 745                                           u64 bytenr, u64 parent,
 746                                           u64 root_objectid)
 747 {
 748         struct btrfs_key key;
 749         int ret;
 750
 751         key.objectid = bytenr;
 752         if (parent) {
 753                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 754                 key.offset = parent;
 755         } else {
 756                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 757                 key.offset = root_objectid;
 758         }
 759
 760         ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
 761                                       path, &key, 0);
 762         btrfs_release_path(path);
 763         return ret;
 764 }
 765
 766 static inline int extent_ref_type(u64 parent, u64 owner)
 767 {
 768         int type;
 769         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 770                 if (parent > 0)
 771                         type = BTRFS_SHARED_BLOCK_REF_KEY;
 772                 else
 773                         type = BTRFS_TREE_BLOCK_REF_KEY;
 774         } else {
 775                 if (parent > 0)
 776                         type = BTRFS_SHARED_DATA_REF_KEY;
 777                 else
 778                         type = BTRFS_EXTENT_DATA_REF_KEY;
 779         }
 780         return type;
 781 }
 782
 783 static int find_next_key(struct btrfs_path *path, int level,
 784                          struct btrfs_key *key)
 785
 786 {
 787         for (; level < BTRFS_MAX_LEVEL; level++) {
 788                 if (!path->nodes[level])
 789                         break;
 790                 if (path->slots[level] + 1 >=
 791                     btrfs_header_nritems(path->nodes[level]))
 792                         continue;
 793                 if (level == 0)
 794                         btrfs_item_key_to_cpu(path->nodes[level], key,
 795                                               path->slots[level] + 1);
 796                 else
 797                         btrfs_node_key_to_cpu(path->nodes[level], key,
 798                                               path->slots[level] + 1);
 799                 return 0;
 800         }
 801         return 1;
 802 }
 803
 804 /*
 805  * look for inline back ref. if back ref is found, *ref_ret is set
 806  * to the address of inline back ref, and 0 is returned.
 807  *
 808  * if back ref isn't found, *ref_ret is set to the address where it
 809  * should be inserted, and -ENOENT is returned.
 810  *
 811  * if insert is true and there are too many inline back refs, the path
 812  * points to the extent item, and -EAGAIN is returned.
 813  *
 814  * NOTE: inline back refs are ordered in the same way that back ref
 815  *       items in the tree are ordered.
 816  */
 817 static noinline_for_stack
 818 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
 819                                  struct btrfs_path *path,
 820                                  struct btrfs_extent_inline_ref **ref_ret,
 821                                  u64 bytenr, u64 num_bytes,
 822                                  u64 parent, u64 root_objectid,
 823                                  u64 owner, u64 offset, int insert)
 824 {
 825         struct btrfs_fs_info *fs_info = trans->fs_info;
 826         struct btrfs_root *root = fs_info->extent_root;
 827         struct btrfs_key key;
 828         struct extent_buffer *leaf;
 829         struct btrfs_extent_item *ei;
 830         struct btrfs_extent_inline_ref *iref;
 831         u64 flags;
 832         u64 item_size;
 833         unsigned long ptr;
 834         unsigned long end;
 835         int extra_size;
 836         int type;
 837         int want;
 838         int ret;
 839         int err = 0;
 840         bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
 841         int needed;
 842
 843         key.objectid = bytenr;
 844         key.type = BTRFS_EXTENT_ITEM_KEY;
 845         key.offset = num_bytes;
 846
 847         want = extent_ref_type(parent, owner);
 848         if (insert) {
 849                 extra_size = btrfs_extent_inline_ref_size(want);
 850                 path->keep_locks = 1;
 851         } else
 852                 extra_size = -1;
 853
 854         /*
 855          * Owner is our level, so we can just add one to get the level for the
 856          * block we are interested in.
 857          */
 858         if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
 859                 key.type = BTRFS_METADATA_ITEM_KEY;
 860                 key.offset = owner;
 861         }
 862
 863 again:
 864         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
 865         if (ret < 0) {
 866                 err = ret;
 867                 goto out;
 868         }
 869
 870         /*
 871          * We may be a newly converted file system which still has the old fat
 872          * extent entries for metadata, so try and see if we have one of those.
 873          */
 874         if (ret > 0 && skinny_metadata) {
 875                 skinny_metadata = false;
 876                 if (path->slots[0]) {
 877                         path->slots[0]--;
 878                         btrfs_item_key_to_cpu(path->nodes[0], &key,
 879                                               path->slots[0]);
 880                         if (key.objectid == bytenr &&
 881                             key.type == BTRFS_EXTENT_ITEM_KEY &&
 882                             key.offset == num_bytes)
 883                                 ret = 0;
 884                 }
 885                 if (ret) {
 886                         key.objectid = bytenr;
 887                         key.type = BTRFS_EXTENT_ITEM_KEY;
 888                         key.offset = num_bytes;
 889                         btrfs_release_path(path);
 890                         goto again;
 891                 }
 892         }
 893
 894         if (ret && !insert) {
 895                 err = -ENOENT;
 896                 goto out;
 897         } else if (WARN_ON(ret)) {
 898                 err = -EIO;
 899                 goto out;
 900         }
 901
 902         leaf = path->nodes[0];
 903         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 904         if (unlikely(item_size < sizeof(*ei))) {
 905                 err = -EINVAL;
 906                 btrfs_print_v0_err(fs_info);
 907                 btrfs_abort_transaction(trans, err);
 908                 goto out;
 909         }
 910
 911         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 912         flags = btrfs_extent_flags(leaf, ei);
 913
 914         ptr = (unsigned long)(ei + 1);
 915         end = (unsigned long)ei + item_size;
 916
 917         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
 918                 ptr += sizeof(struct btrfs_tree_block_info);
 919                 BUG_ON(ptr > end);
 920         }
 921
 922         if (owner >= BTRFS_FIRST_FREE_OBJECTID)
 923                 needed = BTRFS_REF_TYPE_DATA;
 924         else
 925                 needed = BTRFS_REF_TYPE_BLOCK;
 926
 927         err = -ENOENT;
 928         while (1) {
 929                 if (ptr >= end) {
 930                         WARN_ON(ptr > end);
 931                         break;
 932                 }
 933                 iref = (struct btrfs_extent_inline_ref *)ptr;
 934                 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
 935                 if (type == BTRFS_REF_TYPE_INVALID) {
 936                         err = -EUCLEAN;
 937                         goto out;
 938                 }
 939
 940                 if (want < type)
 941                         break;
 942                 if (want > type) {
 943                         ptr += btrfs_extent_inline_ref_size(type);
 944                         continue;
 945                 }
 946
 947                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
 948                         struct btrfs_extent_data_ref *dref;
 949                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 950                         if (match_extent_data_ref(leaf, dref, root_objectid,
 951                                                   owner, offset)) {
 952                                 err = 0;
 953                                 break;
 954                         }
 955                         if (hash_extent_data_ref_item(leaf, dref) <
 956                             hash_extent_data_ref(root_objectid, owner, offset))
 957                                 break;
 958                 } else {
 959                         u64 ref_offset;
 960                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
 961                         if (parent > 0) {
 962                                 if (parent == ref_offset) {
 963                                         err = 0;
 964                                         break;
 965                                 }
 966                                 if (ref_offset < parent)
 967                                         break;
 968                         } else {
 969                                 if (root_objectid == ref_offset) {
 970                                         err = 0;
 971                                         break;
 972                                 }
 973                                 if (ref_offset < root_objectid)
 974                                         break;
 975                         }
 976                 }
 977                 ptr += btrfs_extent_inline_ref_size(type);
 978         }
 979         if (err == -ENOENT && insert) {
 980                 if (item_size + extra_size >=
 981                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
 982                         err = -EAGAIN;
 983                         goto out;
 984                 }
 985                 /*
 986                  * To add new inline back ref, we have to make sure
 987                  * there is no corresponding back ref item.
 988                  * For simplicity, we just do not add new inline back
 989                  * ref if there is any kind of item for this block
 990                  */
 991                 if (find_next_key(path, 0, &key) == 0 &&
 992                     key.objectid == bytenr &&
 993                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
 994                         err = -EAGAIN;
 995                         goto out;
 996                 }
 997         }
 998         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
 999 out:
1000         if (insert) {
1001                 path->keep_locks = 0;
1002                 btrfs_unlock_up_safe(path, 1);
1003         }
1004         return err;
1005 }
1006
1007 /*
1008  * helper to add new inline back ref
1009  */
1010 static noinline_for_stack
1011 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1012                                  struct btrfs_path *path,
1013                                  struct btrfs_extent_inline_ref *iref,
1014                                  u64 parent, u64 root_objectid,
1015                                  u64 owner, u64 offset, int refs_to_add,
1016                                  struct btrfs_delayed_extent_op *extent_op)
1017 {
1018         struct extent_buffer *leaf;
1019         struct btrfs_extent_item *ei;
1020         unsigned long ptr;
1021         unsigned long end;
1022         unsigned long item_offset;
1023         u64 refs;
1024         int size;
1025         int type;
1026
1027         leaf = path->nodes[0];
1028         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1029         item_offset = (unsigned long)iref - (unsigned long)ei;
1030
1031         type = extent_ref_type(parent, owner);
1032         size = btrfs_extent_inline_ref_size(type);
1033
1034         btrfs_extend_item(path, size);
1035
1036         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1037         refs = btrfs_extent_refs(leaf, ei);
1038         refs += refs_to_add;
1039         btrfs_set_extent_refs(leaf, ei, refs);
1040         if (extent_op)
1041                 __run_delayed_extent_op(extent_op, leaf, ei);
1042
1043         ptr = (unsigned long)ei + item_offset;
1044         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1045         if (ptr < end - size)
1046                 memmove_extent_buffer(leaf, ptr + size, ptr,
1047                                       end - size - ptr);
1048
1049         iref = (struct btrfs_extent_inline_ref *)ptr;
1050         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1051         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1052                 struct btrfs_extent_data_ref *dref;
1053                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1054                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1055                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1056                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1057                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1058         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1059                 struct btrfs_shared_data_ref *sref;
1060                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1061                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1062                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1063         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1064                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1065         } else {
1066                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1067         }
1068         btrfs_mark_buffer_dirty(leaf);
1069 }
1070
1071 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1072                                  struct btrfs_path *path,
1073                                  struct btrfs_extent_inline_ref **ref_ret,
1074                                  u64 bytenr, u64 num_bytes, u64 parent,
1075                                  u64 root_objectid, u64 owner, u64 offset)
1076 {
1077         int ret;
1078
1079         ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1080                                            num_bytes, parent, root_objectid,
1081                                            owner, offset, 0);
1082         if (ret != -ENOENT)
1083                 return ret;
1084
1085         btrfs_release_path(path);
1086         *ref_ret = NULL;
1087
1088         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1089                 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1090                                             root_objectid);
1091         } else {
1092                 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1093                                              root_objectid, owner, offset);
1094         }
1095         return ret;
1096 }
1097
1098 /*
1099  * helper to update/remove inline back ref
1100  */
1101 static noinline_for_stack
1102 void update_inline_extent_backref(struct btrfs_path *path,
1103                                   struct btrfs_extent_inline_ref *iref,
1104                                   int refs_to_mod,
1105                                   struct btrfs_delayed_extent_op *extent_op,
1106                                   int *last_ref)
1107 {
1108         struct extent_buffer *leaf = path->nodes[0];
1109         struct btrfs_extent_item *ei;
1110         struct btrfs_extent_data_ref *dref = NULL;
1111         struct btrfs_shared_data_ref *sref = NULL;
1112         unsigned long ptr;
1113         unsigned long end;
1114         u32 item_size;
1115         int size;
1116         int type;
1117         u64 refs;
1118
1119         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1120         refs = btrfs_extent_refs(leaf, ei);
1121         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1122         refs += refs_to_mod;
1123         btrfs_set_extent_refs(leaf, ei, refs);
1124         if (extent_op)
1125                 __run_delayed_extent_op(extent_op, leaf, ei);
1126
1127         /*
1128          * If type is invalid, we should have bailed out after
1129          * lookup_inline_extent_backref().
1130          */
1131         type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1132         ASSERT(type != BTRFS_REF_TYPE_INVALID);
1133
1134         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1135                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1136                 refs = btrfs_extent_data_ref_count(leaf, dref);
1137         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1138                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1139                 refs = btrfs_shared_data_ref_count(leaf, sref);
1140         } else {
1141                 refs = 1;
1142                 BUG_ON(refs_to_mod != -1);
1143         }
1144
1145         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1146         refs += refs_to_mod;
1147
1148         if (refs > 0) {
1149                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1150                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1151                 else
1152                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1153         } else {
1154                 *last_ref = 1;
1155                 size =  btrfs_extent_inline_ref_size(type);
1156                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1157                 ptr = (unsigned long)iref;
1158                 end = (unsigned long)ei + item_size;
1159                 if (ptr + size < end)
1160                         memmove_extent_buffer(leaf, ptr, ptr + size,
1161                                               end - ptr - size);
1162                 item_size -= size;
1163                 btrfs_truncate_item(path, item_size, 1);
1164         }
1165         btrfs_mark_buffer_dirty(leaf);
1166 }
1167
1168 static noinline_for_stack
1169 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1170                                  struct btrfs_path *path,
1171                                  u64 bytenr, u64 num_bytes, u64 parent,
1172                                  u64 root_objectid, u64 owner,
1173                                  u64 offset, int refs_to_add,
1174                                  struct btrfs_delayed_extent_op *extent_op)
1175 {
1176         struct btrfs_extent_inline_ref *iref;
1177         int ret;
1178
1179         ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1180                                            num_bytes, parent, root_objectid,
1181                                            owner, offset, 1);
1182         if (ret == 0) {
1183                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1184                 update_inline_extent_backref(path, iref, refs_to_add,
1185                                              extent_op, NULL);
1186         } else if (ret == -ENOENT) {
1187                 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1188                                             root_objectid, owner, offset,
1189                                             refs_to_add, extent_op);
1190                 ret = 0;
1191         }
1192         return ret;
1193 }
1194
1195 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1196                                  struct btrfs_path *path,
1197                                  u64 bytenr, u64 parent, u64 root_objectid,
1198                                  u64 owner, u64 offset, int refs_to_add)
1199 {
1200         int ret;
1201         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1202                 BUG_ON(refs_to_add != 1);
1203                 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1204                                             root_objectid);
1205         } else {
1206                 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1207                                              root_objectid, owner, offset,
1208                                              refs_to_add);
1209         }
1210         return ret;
1211 }
1212
1213 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1214                                  struct btrfs_path *path,
1215                                  struct btrfs_extent_inline_ref *iref,
1216                                  int refs_to_drop, int is_data, int *last_ref)
1217 {
1218         int ret = 0;
1219
1220         BUG_ON(!is_data && refs_to_drop != 1);
1221         if (iref) {
1222                 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1223                                              last_ref);
1224         } else if (is_data) {
1225                 ret = remove_extent_data_ref(trans, path, refs_to_drop,
1226                                              last_ref);
1227         } else {
1228                 *last_ref = 1;
1229                 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1230         }
1231         return ret;
1232 }
1233
1234 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1235                                u64 *discarded_bytes)
1236 {
1237         int j, ret = 0;
1238         u64 bytes_left, end;
1239         u64 aligned_start = ALIGN(start, 1 << 9);
1240
1241         if (WARN_ON(start != aligned_start)) {
1242                 len -= aligned_start - start;
1243                 len = round_down(len, 1 << 9);
1244                 start = aligned_start;
1245         }
1246
1247         *discarded_bytes = 0;
1248
1249         if (!len)
1250                 return 0;
1251
1252         end = start + len;
1253         bytes_left = len;
1254
1255         /* Skip any superblocks on this device. */
1256         for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1257                 u64 sb_start = btrfs_sb_offset(j);
1258                 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1259                 u64 size = sb_start - start;
1260
1261                 if (!in_range(sb_start, start, bytes_left) &&
1262                     !in_range(sb_end, start, bytes_left) &&
1263                     !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1264                         continue;
1265
1266                 /*
1267                  * Superblock spans beginning of range.  Adjust start and
1268                  * try again.
1269                  */
1270                 if (sb_start <= start) {
1271                         start += sb_end - start;
1272                         if (start > end) {
1273                                 bytes_left = 0;
1274                                 break;
1275                         }
1276                         bytes_left = end - start;
1277                         continue;
1278                 }
1279
1280                 if (size) {
1281                         ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1282                                                    GFP_NOFS, 0);
1283                         if (!ret)
1284                                 *discarded_bytes += size;
1285                         else if (ret != -EOPNOTSUPP)
1286                                 return ret;
1287                 }
1288
1289                 start = sb_end;
1290                 if (start > end) {
1291                         bytes_left = 0;
1292                         break;
1293                 }
1294                 bytes_left = end - start;
1295         }
1296
1297         if (bytes_left) {
1298                 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1299                                            GFP_NOFS, 0);
1300                 if (!ret)
1301                         *discarded_bytes += bytes_left;
1302         }
1303         return ret;
1304 }
1305
1306 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1307                          u64 num_bytes, u64 *actual_bytes)
1308 {
1309         int ret = 0;
1310         u64 discarded_bytes = 0;
1311         u64 end = bytenr + num_bytes;
1312         u64 cur = bytenr;
1313         struct btrfs_bio *bbio = NULL;
1314
1315
1316         /*
1317          * Avoid races with device replace and make sure our bbio has devices
1318          * associated to its stripes that don't go away while we are discarding.
1319          */
1320         btrfs_bio_counter_inc_blocked(fs_info);
1321         while (cur < end) {
1322                 struct btrfs_bio_stripe *stripe;
1323                 int i;
1324
1325                 num_bytes = end - cur;
1326                 /* Tell the block device(s) that the sectors can be discarded */
1327                 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur,
1328                                       &num_bytes, &bbio, 0);
1329                 /*
1330                  * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or
1331                  * -EOPNOTSUPP. For any such error, @num_bytes is not updated,
1332                  * thus we can't continue anyway.
1333                  */
1334                 if (ret < 0)
1335                         goto out;
1336
1337                 stripe = bbio->stripes;
1338                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1339                         u64 bytes;
1340                         struct request_queue *req_q;
1341
1342                         if (!stripe->dev->bdev) {
1343                                 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1344                                 continue;
1345                         }
1346                         req_q = bdev_get_queue(stripe->dev->bdev);
1347                         if (!blk_queue_discard(req_q))
1348                                 continue;
1349
1350                         ret = btrfs_issue_discard(stripe->dev->bdev,
1351                                                   stripe->physical,
1352                                                   stripe->length,
1353                                                   &bytes);
1354                         if (!ret) {
1355                                 discarded_bytes += bytes;
1356                         } else if (ret != -EOPNOTSUPP) {
1357                                 /*
1358                                  * Logic errors or -ENOMEM, or -EIO, but
1359                                  * unlikely to happen.
1360                                  *
1361                                  * And since there are two loops, explicitly
1362                                  * go to out to avoid confusion.
1363                                  */
1364                                 btrfs_put_bbio(bbio);
1365                                 goto out;
1366                         }
1367
1368                         /*
1369                          * Just in case we get back EOPNOTSUPP for some reason,
1370                          * just ignore the return value so we don't screw up
1371                          * people calling discard_extent.
1372                          */
1373                         ret = 0;
1374                 }
1375                 btrfs_put_bbio(bbio);
1376                 cur += num_bytes;
1377         }
1378 out:
1379         btrfs_bio_counter_dec(fs_info);
1380
1381         if (actual_bytes)
1382                 *actual_bytes = discarded_bytes;
1383
1384
1385         if (ret == -EOPNOTSUPP)
1386                 ret = 0;
1387         return ret;
1388 }
1389
1390 /* Can return -ENOMEM */
1391 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1392                          struct btrfs_ref *generic_ref)
1393 {
1394         struct btrfs_fs_info *fs_info = trans->fs_info;
1395         int old_ref_mod, new_ref_mod;
1396         int ret;
1397
1398         ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1399                generic_ref->action);
1400         BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1401                generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID);
1402
1403         if (generic_ref->type == BTRFS_REF_METADATA)
1404                 ret = btrfs_add_delayed_tree_ref(trans, generic_ref,
1405                                 NULL, &old_ref_mod, &new_ref_mod);
1406         else
1407                 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0,
1408                                                  &old_ref_mod, &new_ref_mod);
1409
1410         btrfs_ref_tree_mod(fs_info, generic_ref);
1411
1412         if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
1413                 sub_pinned_bytes(fs_info, generic_ref);
1414
1415         return ret;
1416 }
1417
1418 /*
1419  * __btrfs_inc_extent_ref - insert backreference for a given extent
1420  *
1421  * @trans:          Handle of transaction
1422  *
1423  * @node:           The delayed ref node used to get the bytenr/length for
1424  *                  extent whose references are incremented.
1425  *
1426  * @parent:         If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1427  *                  BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1428  *                  bytenr of the parent block. Since new extents are always
1429  *                  created with indirect references, this will only be the case
1430  *                  when relocating a shared extent. In that case, root_objectid
1431  *                  will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
1432  *                  be 0
1433  *
1434  * @root_objectid:  The id of the root where this modification has originated,
1435  *                  this can be either one of the well-known metadata trees or
1436  *                  the subvolume id which references this extent.
1437  *
1438  * @owner:          For data extents it is the inode number of the owning file.
1439  *                  For metadata extents this parameter holds the level in the
1440  *                  tree of the extent.
1441  *
1442  * @offset:         For metadata extents the offset is ignored and is currently
1443  *                  always passed as 0. For data extents it is the fileoffset
1444  *                  this extent belongs to.
1445  *
1446  * @refs_to_add     Number of references to add
1447  *
1448  * @extent_op       Pointer to a structure, holding information necessary when
1449  *                  updating a tree block's flags
1450  *
1451  */
1452 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1453                                   struct btrfs_delayed_ref_node *node,
1454                                   u64 parent, u64 root_objectid,
1455                                   u64 owner, u64 offset, int refs_to_add,
1456                                   struct btrfs_delayed_extent_op *extent_op)
1457 {
1458         struct btrfs_path *path;
1459         struct extent_buffer *leaf;
1460         struct btrfs_extent_item *item;
1461         struct btrfs_key key;
1462         u64 bytenr = node->bytenr;
1463         u64 num_bytes = node->num_bytes;
1464         u64 refs;
1465         int ret;
1466
1467         path = btrfs_alloc_path();
1468         if (!path)
1469                 return -ENOMEM;
1470
1471         path->reada = READA_FORWARD;
1472         path->leave_spinning = 1;
1473         /* this will setup the path even if it fails to insert the back ref */
1474         ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1475                                            parent, root_objectid, owner,
1476                                            offset, refs_to_add, extent_op);
1477         if ((ret < 0 && ret != -EAGAIN) || !ret)
1478                 goto out;
1479
1480         /*
1481          * Ok we had -EAGAIN which means we didn't have space to insert and
1482          * inline extent ref, so just update the reference count and add a
1483          * normal backref.
1484          */
1485         leaf = path->nodes[0];
1486         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1487         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1488         refs = btrfs_extent_refs(leaf, item);
1489         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1490         if (extent_op)
1491                 __run_delayed_extent_op(extent_op, leaf, item);
1492
1493         btrfs_mark_buffer_dirty(leaf);
1494         btrfs_release_path(path);
1495
1496         path->reada = READA_FORWARD;
1497         path->leave_spinning = 1;
1498         /* now insert the actual backref */
1499         ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid,
1500                                     owner, offset, refs_to_add);
1501         if (ret)
1502                 btrfs_abort_transaction(trans, ret);
1503 out:
1504         btrfs_free_path(path);
1505         return ret;
1506 }
1507
1508 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1509                                 struct btrfs_delayed_ref_node *node,
1510                                 struct btrfs_delayed_extent_op *extent_op,
1511                                 int insert_reserved)
1512 {
1513         int ret = 0;
1514         struct btrfs_delayed_data_ref *ref;
1515         struct btrfs_key ins;
1516         u64 parent = 0;
1517         u64 ref_root = 0;
1518         u64 flags = 0;
1519
1520         ins.objectid = node->bytenr;
1521         ins.offset = node->num_bytes;
1522         ins.type = BTRFS_EXTENT_ITEM_KEY;
1523
1524         ref = btrfs_delayed_node_to_data_ref(node);
1525         trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1526
1527         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1528                 parent = ref->parent;
1529         ref_root = ref->root;
1530
1531         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1532                 if (extent_op)
1533                         flags |= extent_op->flags_to_set;
1534                 ret = alloc_reserved_file_extent(trans, parent, ref_root,
1535                                                  flags, ref->objectid,
1536                                                  ref->offset, &ins,
1537                                                  node->ref_mod);
1538         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1539                 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1540                                              ref->objectid, ref->offset,
1541                                              node->ref_mod, extent_op);
1542         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1543                 ret = __btrfs_free_extent(trans, node, parent,
1544                                           ref_root, ref->objectid,
1545                                           ref->offset, node->ref_mod,
1546                                           extent_op);
1547         } else {
1548                 BUG();
1549         }
1550         return ret;
1551 }
1552
1553 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1554                                     struct extent_buffer *leaf,
1555                                     struct btrfs_extent_item *ei)
1556 {
1557         u64 flags = btrfs_extent_flags(leaf, ei);
1558         if (extent_op->update_flags) {
1559                 flags |= extent_op->flags_to_set;
1560                 btrfs_set_extent_flags(leaf, ei, flags);
1561         }
1562
1563         if (extent_op->update_key) {
1564                 struct btrfs_tree_block_info *bi;
1565                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1566                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1567                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1568         }
1569 }
1570
1571 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1572                                  struct btrfs_delayed_ref_head *head,
1573                                  struct btrfs_delayed_extent_op *extent_op)
1574 {
1575         struct btrfs_fs_info *fs_info = trans->fs_info;
1576         struct btrfs_key key;
1577         struct btrfs_path *path;
1578         struct btrfs_extent_item *ei;
1579         struct extent_buffer *leaf;
1580         u32 item_size;
1581         int ret;
1582         int err = 0;
1583         int metadata = !extent_op->is_data;
1584
1585         if (trans->aborted)
1586                 return 0;
1587
1588         if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1589                 metadata = 0;
1590
1591         path = btrfs_alloc_path();
1592         if (!path)
1593                 return -ENOMEM;
1594
1595         key.objectid = head->bytenr;
1596
1597         if (metadata) {
1598                 key.type = BTRFS_METADATA_ITEM_KEY;
1599                 key.offset = extent_op->level;
1600         } else {
1601                 key.type = BTRFS_EXTENT_ITEM_KEY;
1602                 key.offset = head->num_bytes;
1603         }
1604
1605 again:
1606         path->reada = READA_FORWARD;
1607         path->leave_spinning = 1;
1608         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
1609         if (ret < 0) {
1610                 err = ret;
1611                 goto out;
1612         }
1613         if (ret > 0) {
1614                 if (metadata) {
1615                         if (path->slots[0] > 0) {
1616                                 path->slots[0]--;
1617                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
1618                                                       path->slots[0]);
1619                                 if (key.objectid == head->bytenr &&
1620                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
1621                                     key.offset == head->num_bytes)
1622                                         ret = 0;
1623                         }
1624                         if (ret > 0) {
1625                                 btrfs_release_path(path);
1626                                 metadata = 0;
1627
1628                                 key.objectid = head->bytenr;
1629                                 key.offset = head->num_bytes;
1630                                 key.type = BTRFS_EXTENT_ITEM_KEY;
1631                                 goto again;
1632                         }
1633                 } else {
1634                         err = -EIO;
1635                         goto out;
1636                 }
1637         }
1638
1639         leaf = path->nodes[0];
1640         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1641
1642         if (unlikely(item_size < sizeof(*ei))) {
1643                 err = -EINVAL;
1644                 btrfs_print_v0_err(fs_info);
1645                 btrfs_abort_transaction(trans, err);
1646                 goto out;
1647         }
1648
1649         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1650         __run_delayed_extent_op(extent_op, leaf, ei);
1651
1652         btrfs_mark_buffer_dirty(leaf);
1653 out:
1654         btrfs_free_path(path);
1655         return err;
1656 }
1657
1658 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1659                                 struct btrfs_delayed_ref_node *node,
1660                                 struct btrfs_delayed_extent_op *extent_op,
1661                                 int insert_reserved)
1662 {
1663         int ret = 0;
1664         struct btrfs_delayed_tree_ref *ref;
1665         u64 parent = 0;
1666         u64 ref_root = 0;
1667
1668         ref = btrfs_delayed_node_to_tree_ref(node);
1669         trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1670
1671         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1672                 parent = ref->parent;
1673         ref_root = ref->root;
1674
1675         if (node->ref_mod != 1) {
1676                 btrfs_err(trans->fs_info,
1677         "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
1678                           node->bytenr, node->ref_mod, node->action, ref_root,
1679                           parent);
1680                 return -EIO;
1681         }
1682         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1683                 BUG_ON(!extent_op || !extent_op->update_flags);
1684                 ret = alloc_reserved_tree_block(trans, node, extent_op);
1685         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1686                 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1687                                              ref->level, 0, 1, extent_op);
1688         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1689                 ret = __btrfs_free_extent(trans, node, parent, ref_root,
1690                                           ref->level, 0, 1, extent_op);
1691         } else {
1692                 BUG();
1693         }
1694         return ret;
1695 }
1696
1697 /* helper function to actually process a single delayed ref entry */
1698 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1699                                struct btrfs_delayed_ref_node *node,
1700                                struct btrfs_delayed_extent_op *extent_op,
1701                                int insert_reserved)
1702 {
1703         int ret = 0;
1704
1705         if (trans->aborted) {
1706                 if (insert_reserved)
1707                         btrfs_pin_extent(trans->fs_info, node->bytenr,
1708                                          node->num_bytes, 1);
1709                 return 0;
1710         }
1711
1712         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1713             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1714                 ret = run_delayed_tree_ref(trans, node, extent_op,
1715                                            insert_reserved);
1716         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1717                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1718                 ret = run_delayed_data_ref(trans, node, extent_op,
1719                                            insert_reserved);
1720         else
1721                 BUG();
1722         if (ret && insert_reserved)
1723                 btrfs_pin_extent(trans->fs_info, node->bytenr,
1724                                  node->num_bytes, 1);
1725         return ret;
1726 }
1727
1728 static inline struct btrfs_delayed_ref_node *
1729 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1730 {
1731         struct btrfs_delayed_ref_node *ref;
1732
1733         if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1734                 return NULL;
1735
1736         /*
1737          * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1738          * This is to prevent a ref count from going down to zero, which deletes
1739          * the extent item from the extent tree, when there still are references
1740          * to add, which would fail because they would not find the extent item.
1741          */
1742         if (!list_empty(&head->ref_add_list))
1743                 return list_first_entry(&head->ref_add_list,
1744                                 struct btrfs_delayed_ref_node, add_list);
1745
1746         ref = rb_entry(rb_first_cached(&head->ref_tree),
1747                        struct btrfs_delayed_ref_node, ref_node);
1748         ASSERT(list_empty(&ref->add_list));
1749         return ref;
1750 }
1751
1752 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1753                                       struct btrfs_delayed_ref_head *head)
1754 {
1755         spin_lock(&delayed_refs->lock);
1756         head->processing = 0;
1757         delayed_refs->num_heads_ready++;
1758         spin_unlock(&delayed_refs->lock);
1759         btrfs_delayed_ref_unlock(head);
1760 }
1761
1762 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1763                                 struct btrfs_delayed_ref_head *head)
1764 {
1765         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1766
1767         if (!extent_op)
1768                 return NULL;
1769
1770         if (head->must_insert_reserved) {
1771                 head->extent_op = NULL;
1772                 btrfs_free_delayed_extent_op(extent_op);
1773                 return NULL;
1774         }
1775         return extent_op;
1776 }
1777
1778 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1779                                      struct btrfs_delayed_ref_head *head)
1780 {
1781         struct btrfs_delayed_extent_op *extent_op;
1782         int ret;
1783
1784         extent_op = cleanup_extent_op(head);
1785         if (!extent_op)
1786                 return 0;
1787         head->extent_op = NULL;
1788         spin_unlock(&head->lock);
1789         ret = run_delayed_extent_op(trans, head, extent_op);
1790         btrfs_free_delayed_extent_op(extent_op);
1791         return ret ? ret : 1;
1792 }
1793
1794 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1795                                   struct btrfs_delayed_ref_root *delayed_refs,
1796                                   struct btrfs_delayed_ref_head *head)
1797 {
1798         int nr_items = 1;       /* Dropping this ref head update. */
1799
1800         if (head->total_ref_mod < 0) {
1801                 struct btrfs_space_info *space_info;
1802                 u64 flags;
1803
1804                 if (head->is_data)
1805                         flags = BTRFS_BLOCK_GROUP_DATA;
1806                 else if (head->is_system)
1807                         flags = BTRFS_BLOCK_GROUP_SYSTEM;
1808                 else
1809                         flags = BTRFS_BLOCK_GROUP_METADATA;
1810                 space_info = btrfs_find_space_info(fs_info, flags);
1811                 ASSERT(space_info);
1812                 percpu_counter_add_batch(&space_info->total_bytes_pinned,
1813                                    -head->num_bytes,
1814                                    BTRFS_TOTAL_BYTES_PINNED_BATCH);
1815
1816                 /*
1817                  * We had csum deletions accounted for in our delayed refs rsv,
1818                  * we need to drop the csum leaves for this update from our
1819                  * delayed_refs_rsv.
1820                  */
1821                 if (head->is_data) {
1822                         spin_lock(&delayed_refs->lock);
1823                         delayed_refs->pending_csums -= head->num_bytes;
1824                         spin_unlock(&delayed_refs->lock);
1825                         nr_items += btrfs_csum_bytes_to_leaves(fs_info,
1826                                 head->num_bytes);
1827                 }
1828         }
1829
1830         btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1831 }
1832
1833 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1834                             struct btrfs_delayed_ref_head *head)
1835 {
1836
1837         struct btrfs_fs_info *fs_info = trans->fs_info;
1838         struct btrfs_delayed_ref_root *delayed_refs;
1839         int ret;
1840
1841         delayed_refs = &trans->transaction->delayed_refs;
1842
1843         ret = run_and_cleanup_extent_op(trans, head);
1844         if (ret < 0) {
1845                 unselect_delayed_ref_head(delayed_refs, head);
1846                 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1847                 return ret;
1848         } else if (ret) {
1849                 return ret;
1850         }
1851
1852         /*
1853          * Need to drop our head ref lock and re-acquire the delayed ref lock
1854          * and then re-check to make sure nobody got added.
1855          */
1856         spin_unlock(&head->lock);
1857         spin_lock(&delayed_refs->lock);
1858         spin_lock(&head->lock);
1859         if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1860                 spin_unlock(&head->lock);
1861                 spin_unlock(&delayed_refs->lock);
1862                 return 1;
1863         }
1864         btrfs_delete_ref_head(delayed_refs, head);
1865         spin_unlock(&head->lock);
1866         spin_unlock(&delayed_refs->lock);
1867
1868         if (head->must_insert_reserved) {
1869                 btrfs_pin_extent(fs_info, head->bytenr,
1870                                  head->num_bytes, 1);
1871                 if (head->is_data) {
1872                         ret = btrfs_del_csums(trans, fs_info->csum_root,
1873                                               head->bytenr, head->num_bytes);
1874                 }
1875         }
1876
1877         btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1878
1879         trace_run_delayed_ref_head(fs_info, head, 0);
1880         btrfs_delayed_ref_unlock(head);
1881         btrfs_put_delayed_ref_head(head);
1882         return 0;
1883 }
1884
1885 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1886                                         struct btrfs_trans_handle *trans)
1887 {
1888         struct btrfs_delayed_ref_root *delayed_refs =
1889                 &trans->transaction->delayed_refs;
1890         struct btrfs_delayed_ref_head *head = NULL;
1891         int ret;
1892
1893         spin_lock(&delayed_refs->lock);
1894         head = btrfs_select_ref_head(delayed_refs);
1895         if (!head) {
1896                 spin_unlock(&delayed_refs->lock);
1897                 return head;
1898         }
1899
1900         /*
1901          * Grab the lock that says we are going to process all the refs for
1902          * this head
1903          */
1904         ret = btrfs_delayed_ref_lock(delayed_refs, head);
1905         spin_unlock(&delayed_refs->lock);
1906
1907         /*
1908          * We may have dropped the spin lock to get the head mutex lock, and
1909          * that might have given someone else time to free the head.  If that's
1910          * true, it has been removed from our list and we can move on.
1911          */
1912         if (ret == -EAGAIN)
1913                 head = ERR_PTR(-EAGAIN);
1914
1915         return head;
1916 }
1917
1918 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1919                                     struct btrfs_delayed_ref_head *locked_ref,
1920                                     unsigned long *run_refs)
1921 {
1922         struct btrfs_fs_info *fs_info = trans->fs_info;
1923         struct btrfs_delayed_ref_root *delayed_refs;
1924         struct btrfs_delayed_extent_op *extent_op;
1925         struct btrfs_delayed_ref_node *ref;
1926         int must_insert_reserved = 0;
1927         int ret;
1928
1929         delayed_refs = &trans->transaction->delayed_refs;
1930
1931         lockdep_assert_held(&locked_ref->mutex);
1932         lockdep_assert_held(&locked_ref->lock);
1933
1934         while ((ref = select_delayed_ref(locked_ref))) {
1935                 if (ref->seq &&
1936                     btrfs_check_delayed_seq(fs_info, ref->seq)) {
1937                         spin_unlock(&locked_ref->lock);
1938                         unselect_delayed_ref_head(delayed_refs, locked_ref);
1939                         return -EAGAIN;
1940                 }
1941
1942                 (*run_refs)++;
1943                 ref->in_tree = 0;
1944                 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1945                 RB_CLEAR_NODE(&ref->ref_node);
1946                 if (!list_empty(&ref->add_list))
1947                         list_del(&ref->add_list);
1948                 /*
1949                  * When we play the delayed ref, also correct the ref_mod on
1950                  * head
1951                  */
1952                 switch (ref->action) {
1953                 case BTRFS_ADD_DELAYED_REF:
1954                 case BTRFS_ADD_DELAYED_EXTENT:
1955                         locked_ref->ref_mod -= ref->ref_mod;
1956                         break;
1957                 case BTRFS_DROP_DELAYED_REF:
1958                         locked_ref->ref_mod += ref->ref_mod;
1959                         break;
1960                 default:
1961                         WARN_ON(1);
1962                 }
1963                 atomic_dec(&delayed_refs->num_entries);
1964
1965                 /*
1966                  * Record the must_insert_reserved flag before we drop the
1967                  * spin lock.
1968                  */
1969                 must_insert_reserved = locked_ref->must_insert_reserved;
1970                 locked_ref->must_insert_reserved = 0;
1971
1972                 extent_op = locked_ref->extent_op;
1973                 locked_ref->extent_op = NULL;
1974                 spin_unlock(&locked_ref->lock);
1975
1976                 ret = run_one_delayed_ref(trans, ref, extent_op,
1977                                           must_insert_reserved);
1978
1979                 btrfs_free_delayed_extent_op(extent_op);
1980                 if (ret) {
1981                         unselect_delayed_ref_head(delayed_refs, locked_ref);
1982                         btrfs_put_delayed_ref(ref);
1983                         btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
1984                                     ret);
1985                         return ret;
1986                 }
1987
1988                 btrfs_put_delayed_ref(ref);
1989                 cond_resched();
1990
1991                 spin_lock(&locked_ref->lock);
1992                 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
1993         }
1994
1995         return 0;
1996 }
1997
1998 /*
1999  * Returns 0 on success or if called with an already aborted transaction.
2000  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2001  */
2002 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2003                                              unsigned long nr)
2004 {
2005         struct btrfs_fs_info *fs_info = trans->fs_info;
2006         struct btrfs_delayed_ref_root *delayed_refs;
2007         struct btrfs_delayed_ref_head *locked_ref = NULL;
2008         ktime_t start = ktime_get();
2009         int ret;
2010         unsigned long count = 0;
2011         unsigned long actual_count = 0;
2012
2013         delayed_refs = &trans->transaction->delayed_refs;
2014         do {
2015                 if (!locked_ref) {
2016                         locked_ref = btrfs_obtain_ref_head(trans);
2017                         if (IS_ERR_OR_NULL(locked_ref)) {
2018                                 if (PTR_ERR(locked_ref) == -EAGAIN) {
2019                                         continue;
2020                                 } else {
2021                                         break;
2022                                 }
2023                         }
2024                         count++;
2025                 }
2026                 /*
2027                  * We need to try and merge add/drops of the same ref since we
2028                  * can run into issues with relocate dropping the implicit ref
2029                  * and then it being added back again before the drop can
2030                  * finish.  If we merged anything we need to re-loop so we can
2031                  * get a good ref.
2032                  * Or we can get node references of the same type that weren't
2033                  * merged when created due to bumps in the tree mod seq, and
2034                  * we need to merge them to prevent adding an inline extent
2035                  * backref before dropping it (triggering a BUG_ON at
2036                  * insert_inline_extent_backref()).
2037                  */
2038                 spin_lock(&locked_ref->lock);
2039                 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2040
2041                 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2042                                                       &actual_count);
2043                 if (ret < 0 && ret != -EAGAIN) {
2044                         /*
2045                          * Error, btrfs_run_delayed_refs_for_head already
2046                          * unlocked everything so just bail out
2047                          */
2048                         return ret;
2049                 } else if (!ret) {
2050                         /*
2051                          * Success, perform the usual cleanup of a processed
2052                          * head
2053                          */
2054                         ret = cleanup_ref_head(trans, locked_ref);
2055                         if (ret > 0 ) {
2056                                 /* We dropped our lock, we need to loop. */
2057                                 ret = 0;
2058                                 continue;
2059                         } else if (ret) {
2060                                 return ret;
2061                         }
2062                 }
2063
2064                 /*
2065                  * Either success case or btrfs_run_delayed_refs_for_head
2066                  * returned -EAGAIN, meaning we need to select another head
2067                  */
2068
2069                 locked_ref = NULL;
2070                 cond_resched();
2071         } while ((nr != -1 && count < nr) || locked_ref);
2072
2073         /*
2074          * We don't want to include ref heads since we can have empty ref heads
2075          * and those will drastically skew our runtime down since we just do
2076          * accounting, no actual extent tree updates.
2077          */
2078         if (actual_count > 0) {
2079                 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2080                 u64 avg;
2081
2082                 /*
2083                  * We weigh the current average higher than our current runtime
2084                  * to avoid large swings in the average.
2085                  */
2086                 spin_lock(&delayed_refs->lock);
2087                 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2088                 fs_info->avg_delayed_ref_runtime = avg >> 2;    /* div by 4 */
2089                 spin_unlock(&delayed_refs->lock);
2090         }
2091         return 0;
2092 }
2093
2094 #ifdef SCRAMBLE_DELAYED_REFS
2095 /*
2096  * Normally delayed refs get processed in ascending bytenr order. This
2097  * correlates in most cases to the order added. To expose dependencies on this
2098  * order, we start to process the tree in the middle instead of the beginning
2099  */
2100 static u64 find_middle(struct rb_root *root)
2101 {
2102         struct rb_node *n = root->rb_node;
2103         struct btrfs_delayed_ref_node *entry;
2104         int alt = 1;
2105         u64 middle;
2106         u64 first = 0, last = 0;
2107
2108         n = rb_first(root);
2109         if (n) {
2110                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2111                 first = entry->bytenr;
2112         }
2113         n = rb_last(root);
2114         if (n) {
2115                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2116                 last = entry->bytenr;
2117         }
2118         n = root->rb_node;
2119
2120         while (n) {
2121                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2122                 WARN_ON(!entry->in_tree);
2123
2124                 middle = entry->bytenr;
2125
2126                 if (alt)
2127                         n = n->rb_left;
2128                 else
2129                         n = n->rb_right;
2130
2131                 alt = 1 - alt;
2132         }
2133         return middle;
2134 }
2135 #endif
2136
2137 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2138 {
2139         u64 num_bytes;
2140
2141         num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2142                              sizeof(struct btrfs_extent_inline_ref));
2143         if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2144                 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2145
2146         /*
2147          * We don't ever fill up leaves all the way so multiply by 2 just to be
2148          * closer to what we're really going to want to use.
2149          */
2150         return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2151 }
2152
2153 /*
2154  * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2155  * would require to store the csums for that many bytes.
2156  */
2157 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2158 {
2159         u64 csum_size;
2160         u64 num_csums_per_leaf;
2161         u64 num_csums;
2162
2163         csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2164         num_csums_per_leaf = div64_u64(csum_size,
2165                         (u64)btrfs_super_csum_size(fs_info->super_copy));
2166         num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2167         num_csums += num_csums_per_leaf - 1;
2168         num_csums = div64_u64(num_csums, num_csums_per_leaf);
2169         return num_csums;
2170 }
2171
2172 /*
2173  * this starts processing the delayed reference count updates and
2174  * extent insertions we have queued up so far.  count can be
2175  * 0, which means to process everything in the tree at the start
2176  * of the run (but not newly added entries), or it can be some target
2177  * number you'd like to process.
2178  *
2179  * Returns 0 on success or if called with an aborted transaction
2180  * Returns <0 on error and aborts the transaction
2181  */
2182 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2183                            unsigned long count)
2184 {
2185         struct btrfs_fs_info *fs_info = trans->fs_info;
2186         struct rb_node *node;
2187         struct btrfs_delayed_ref_root *delayed_refs;
2188         struct btrfs_delayed_ref_head *head;
2189         int ret;
2190         int run_all = count == (unsigned long)-1;
2191
2192         /* We'll clean this up in btrfs_cleanup_transaction */
2193         if (trans->aborted)
2194                 return 0;
2195
2196         if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2197                 return 0;
2198
2199         delayed_refs = &trans->transaction->delayed_refs;
2200         if (count == 0)
2201                 count = atomic_read(&delayed_refs->num_entries) * 2;
2202
2203 again:
2204 #ifdef SCRAMBLE_DELAYED_REFS
2205         delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2206 #endif
2207         ret = __btrfs_run_delayed_refs(trans, count);
2208         if (ret < 0) {
2209                 btrfs_abort_transaction(trans, ret);
2210                 return ret;
2211         }
2212
2213         if (run_all) {
2214                 btrfs_create_pending_block_groups(trans);
2215
2216                 spin_lock(&delayed_refs->lock);
2217                 node = rb_first_cached(&delayed_refs->href_root);
2218                 if (!node) {
2219                         spin_unlock(&delayed_refs->lock);
2220                         goto out;
2221                 }
2222                 head = rb_entry(node, struct btrfs_delayed_ref_head,
2223                                 href_node);
2224                 refcount_inc(&head->refs);
2225                 spin_unlock(&delayed_refs->lock);
2226
2227                 /* Mutex was contended, block until it's released and retry. */
2228                 mutex_lock(&head->mutex);
2229                 mutex_unlock(&head->mutex);
2230
2231                 btrfs_put_delayed_ref_head(head);
2232                 cond_resched();
2233                 goto again;
2234         }
2235 out:
2236         return 0;
2237 }
2238
2239 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2240                                 u64 bytenr, u64 num_bytes, u64 flags,
2241                                 int level, int is_data)
2242 {
2243         struct btrfs_delayed_extent_op *extent_op;
2244         int ret;
2245
2246         extent_op = btrfs_alloc_delayed_extent_op();
2247         if (!extent_op)
2248                 return -ENOMEM;
2249
2250         extent_op->flags_to_set = flags;
2251         extent_op->update_flags = true;
2252         extent_op->update_key = false;
2253         extent_op->is_data = is_data ? true : false;
2254         extent_op->level = level;
2255
2256         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2257         if (ret)
2258                 btrfs_free_delayed_extent_op(extent_op);
2259         return ret;
2260 }
2261
2262 static noinline int check_delayed_ref(struct btrfs_root *root,
2263                                       struct btrfs_path *path,
2264                                       u64 objectid, u64 offset, u64 bytenr)
2265 {
2266         struct btrfs_delayed_ref_head *head;
2267         struct btrfs_delayed_ref_node *ref;
2268         struct btrfs_delayed_data_ref *data_ref;
2269         struct btrfs_delayed_ref_root *delayed_refs;
2270         struct btrfs_transaction *cur_trans;
2271         struct rb_node *node;
2272         int ret = 0;
2273
2274         spin_lock(&root->fs_info->trans_lock);
2275         cur_trans = root->fs_info->running_transaction;
2276         if (cur_trans)
2277                 refcount_inc(&cur_trans->use_count);
2278         spin_unlock(&root->fs_info->trans_lock);
2279         if (!cur_trans)
2280                 return 0;
2281
2282         delayed_refs = &cur_trans->delayed_refs;
2283         spin_lock(&delayed_refs->lock);
2284         head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2285         if (!head) {
2286                 spin_unlock(&delayed_refs->lock);
2287                 btrfs_put_transaction(cur_trans);
2288                 return 0;
2289         }
2290
2291         if (!mutex_trylock(&head->mutex)) {
2292                 refcount_inc(&head->refs);
2293                 spin_unlock(&delayed_refs->lock);
2294
2295                 btrfs_release_path(path);
2296
2297                 /*
2298                  * Mutex was contended, block until it's released and let
2299                  * caller try again
2300                  */
2301                 mutex_lock(&head->mutex);
2302                 mutex_unlock(&head->mutex);
2303                 btrfs_put_delayed_ref_head(head);
2304                 btrfs_put_transaction(cur_trans);
2305                 return -EAGAIN;
2306         }
2307         spin_unlock(&delayed_refs->lock);
2308
2309         spin_lock(&head->lock);
2310         /*
2311          * XXX: We should replace this with a proper search function in the
2312          * future.
2313          */
2314         for (node = rb_first_cached(&head->ref_tree); node;
2315              node = rb_next(node)) {
2316                 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2317                 /* If it's a shared ref we know a cross reference exists */
2318                 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2319                         ret = 1;
2320                         break;
2321                 }
2322
2323                 data_ref = btrfs_delayed_node_to_data_ref(ref);
2324
2325                 /*
2326                  * If our ref doesn't match the one we're currently looking at
2327                  * then we have a cross reference.
2328                  */
2329                 if (data_ref->root != root->root_key.objectid ||
2330                     data_ref->objectid != objectid ||
2331                     data_ref->offset != offset) {
2332                         ret = 1;
2333                         break;
2334                 }
2335         }
2336         spin_unlock(&head->lock);
2337         mutex_unlock(&head->mutex);
2338         btrfs_put_transaction(cur_trans);
2339         return ret;
2340 }
2341
2342 static noinline int check_committed_ref(struct btrfs_root *root,
2343                                         struct btrfs_path *path,
2344                                         u64 objectid, u64 offset, u64 bytenr)
2345 {
2346         struct btrfs_fs_info *fs_info = root->fs_info;
2347         struct btrfs_root *extent_root = fs_info->extent_root;
2348         struct extent_buffer *leaf;
2349         struct btrfs_extent_data_ref *ref;
2350         struct btrfs_extent_inline_ref *iref;
2351         struct btrfs_extent_item *ei;
2352         struct btrfs_key key;
2353         u32 item_size;
2354         int type;
2355         int ret;
2356
2357         key.objectid = bytenr;
2358         key.offset = (u64)-1;
2359         key.type = BTRFS_EXTENT_ITEM_KEY;
2360
2361         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2362         if (ret < 0)
2363                 goto out;
2364         BUG_ON(ret == 0); /* Corruption */
2365
2366         ret = -ENOENT;
2367         if (path->slots[0] == 0)
2368                 goto out;
2369
2370         path->slots[0]--;
2371         leaf = path->nodes[0];
2372         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2373
2374         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2375                 goto out;
2376
2377         ret = 1;
2378         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2379         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2380
2381         /* If extent item has more than 1 inline ref then it's shared */
2382         if (item_size != sizeof(*ei) +
2383             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2384                 goto out;
2385
2386         /* If extent created before last snapshot => it's definitely shared */
2387         if (btrfs_extent_generation(leaf, ei) <=
2388             btrfs_root_last_snapshot(&root->root_item))
2389                 goto out;
2390
2391         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2392
2393         /* If this extent has SHARED_DATA_REF then it's shared */
2394         type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2395         if (type != BTRFS_EXTENT_DATA_REF_KEY)
2396                 goto out;
2397
2398         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2399         if (btrfs_extent_refs(leaf, ei) !=
2400             btrfs_extent_data_ref_count(leaf, ref) ||
2401             btrfs_extent_data_ref_root(leaf, ref) !=
2402             root->root_key.objectid ||
2403             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2404             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2405                 goto out;
2406
2407         ret = 0;
2408 out:
2409         return ret;
2410 }
2411
2412 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2413                           u64 bytenr)
2414 {
2415         struct btrfs_path *path;
2416         int ret;
2417
2418         path = btrfs_alloc_path();
2419         if (!path)
2420                 return -ENOMEM;
2421
2422         do {
2423                 ret = check_committed_ref(root, path, objectid,
2424                                           offset, bytenr);
2425                 if (ret && ret != -ENOENT)
2426                         goto out;
2427
2428                 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2429         } while (ret == -EAGAIN);
2430
2431 out:
2432         btrfs_free_path(path);
2433         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2434                 WARN_ON(ret > 0);
2435         return ret;
2436 }
2437
2438 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2439                            struct btrfs_root *root,
2440                            struct extent_buffer *buf,
2441                            int full_backref, int inc)
2442 {
2443         struct btrfs_fs_info *fs_info = root->fs_info;
2444         u64 bytenr;
2445         u64 num_bytes;
2446         u64 parent;
2447         u64 ref_root;
2448         u32 nritems;
2449         struct btrfs_key key;
2450         struct btrfs_file_extent_item *fi;
2451         struct btrfs_ref generic_ref = { 0 };
2452         bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2453         int i;
2454         int action;
2455         int level;
2456         int ret = 0;
2457
2458         if (btrfs_is_testing(fs_info))
2459                 return 0;
2460
2461         ref_root = btrfs_header_owner(buf);
2462         nritems = btrfs_header_nritems(buf);
2463         level = btrfs_header_level(buf);
2464
2465         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
2466                 return 0;
2467
2468         if (full_backref)
2469                 parent = buf->start;
2470         else
2471                 parent = 0;
2472         if (inc)
2473                 action = BTRFS_ADD_DELAYED_REF;
2474         else
2475                 action = BTRFS_DROP_DELAYED_REF;
2476
2477         for (i = 0; i < nritems; i++) {
2478                 if (level == 0) {
2479                         btrfs_item_key_to_cpu(buf, &key, i);
2480                         if (key.type != BTRFS_EXTENT_DATA_KEY)
2481                                 continue;
2482                         fi = btrfs_item_ptr(buf, i,
2483                                             struct btrfs_file_extent_item);
2484                         if (btrfs_file_extent_type(buf, fi) ==
2485                             BTRFS_FILE_EXTENT_INLINE)
2486                                 continue;
2487                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2488                         if (bytenr == 0)
2489                                 continue;
2490
2491                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2492                         key.offset -= btrfs_file_extent_offset(buf, fi);
2493                         btrfs_init_generic_ref(&generic_ref, action, bytenr,
2494                                                num_bytes, parent);
2495                         generic_ref.real_root = root->root_key.objectid;
2496                         btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2497                                             key.offset);
2498                         generic_ref.skip_qgroup = for_reloc;
2499                         if (inc)
2500                                 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2501                         else
2502                                 ret = btrfs_free_extent(trans, &generic_ref);
2503                         if (ret)
2504                                 goto fail;
2505                 } else {
2506                         bytenr = btrfs_node_blockptr(buf, i);
2507                         num_bytes = fs_info->nodesize;
2508                         btrfs_init_generic_ref(&generic_ref, action, bytenr,
2509                                                num_bytes, parent);
2510                         generic_ref.real_root = root->root_key.objectid;
2511                         btrfs_init_tree_ref(&generic_ref, level - 1, ref_root);
2512                         generic_ref.skip_qgroup = for_reloc;
2513                         if (inc)
2514                                 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2515                         else
2516                                 ret = btrfs_free_extent(trans, &generic_ref);
2517                         if (ret)
2518                                 goto fail;
2519                 }
2520         }
2521         return 0;
2522 fail:
2523         return ret;
2524 }
2525
2526 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2527                   struct extent_buffer *buf, int full_backref)
2528 {
2529         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2530 }
2531
2532 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2533                   struct extent_buffer *buf, int full_backref)
2534 {
2535         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2536 }
2537
2538 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
2539 {
2540         struct btrfs_block_group *block_group;
2541         int readonly = 0;
2542
2543         block_group = btrfs_lookup_block_group(fs_info, bytenr);
2544         if (!block_group || block_group->ro)
2545                 readonly = 1;
2546         if (block_group)
2547                 btrfs_put_block_group(block_group);
2548         return readonly;
2549 }
2550
2551 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2552 {
2553         struct btrfs_fs_info *fs_info = root->fs_info;
2554         u64 flags;
2555         u64 ret;
2556
2557         if (data)
2558                 flags = BTRFS_BLOCK_GROUP_DATA;
2559         else if (root == fs_info->chunk_root)
2560                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2561         else
2562                 flags = BTRFS_BLOCK_GROUP_METADATA;
2563
2564         ret = btrfs_get_alloc_profile(fs_info, flags);
2565         return ret;
2566 }
2567
2568 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
2569 {
2570         struct btrfs_block_group *cache;
2571         u64 bytenr;
2572
2573         spin_lock(&fs_info->block_group_cache_lock);
2574         bytenr = fs_info->first_logical_byte;
2575         spin_unlock(&fs_info->block_group_cache_lock);
2576
2577         if (bytenr < (u64)-1)
2578                 return bytenr;
2579
2580         cache = btrfs_lookup_first_block_group(fs_info, search_start);
2581         if (!cache)
2582                 return 0;
2583
2584         bytenr = cache->start;
2585         btrfs_put_block_group(cache);
2586
2587         return bytenr;
2588 }
2589
2590 static int pin_down_extent(struct btrfs_block_group *cache,
2591                            u64 bytenr, u64 num_bytes, int reserved)
2592 {
2593         struct btrfs_fs_info *fs_info = cache->fs_info;
2594
2595         spin_lock(&cache->space_info->lock);
2596         spin_lock(&cache->lock);
2597         cache->pinned += num_bytes;
2598         btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2599                                              num_bytes);
2600         if (reserved) {
2601                 cache->reserved -= num_bytes;
2602                 cache->space_info->bytes_reserved -= num_bytes;
2603         }
2604         spin_unlock(&cache->lock);
2605         spin_unlock(&cache->space_info->lock);
2606
2607         percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
2608                     num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
2609         set_extent_dirty(fs_info->pinned_extents, bytenr,
2610                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
2611         return 0;
2612 }
2613
2614 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2615                      u64 bytenr, u64 num_bytes, int reserved)
2616 {
2617         struct btrfs_block_group *cache;
2618
2619         ASSERT(fs_info->running_transaction);
2620
2621         cache = btrfs_lookup_block_group(fs_info, bytenr);
2622         BUG_ON(!cache); /* Logic error */
2623
2624         pin_down_extent(cache, bytenr, num_bytes, reserved);
2625
2626         btrfs_put_block_group(cache);
2627         return 0;
2628 }
2629
2630 /*
2631  * this function must be called within transaction
2632  */
2633 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
2634                                     u64 bytenr, u64 num_bytes)
2635 {
2636         struct btrfs_block_group *cache;
2637         int ret;
2638
2639         cache = btrfs_lookup_block_group(fs_info, bytenr);
2640         if (!cache)
2641                 return -EINVAL;
2642
2643         /*
2644          * pull in the free space cache (if any) so that our pin
2645          * removes the free space from the cache.  We have load_only set
2646          * to one because the slow code to read in the free extents does check
2647          * the pinned extents.
2648          */
2649         btrfs_cache_block_group(cache, 1);
2650
2651         pin_down_extent(cache, bytenr, num_bytes, 0);
2652
2653         /* remove us from the free space cache (if we're there at all) */
2654         ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2655         btrfs_put_block_group(cache);
2656         return ret;
2657 }
2658
2659 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2660                                    u64 start, u64 num_bytes)
2661 {
2662         int ret;
2663         struct btrfs_block_group *block_group;
2664         struct btrfs_caching_control *caching_ctl;
2665
2666         block_group = btrfs_lookup_block_group(fs_info, start);
2667         if (!block_group)
2668                 return -EINVAL;
2669
2670         btrfs_cache_block_group(block_group, 0);
2671         caching_ctl = btrfs_get_caching_control(block_group);
2672
2673         if (!caching_ctl) {
2674                 /* Logic error */
2675                 BUG_ON(!btrfs_block_group_done(block_group));
2676                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2677         } else {
2678                 mutex_lock(&caching_ctl->mutex);
2679
2680                 if (start >= caching_ctl->progress) {
2681                         ret = btrfs_add_excluded_extent(fs_info, start,
2682                                                         num_bytes);
2683                 } else if (start + num_bytes <= caching_ctl->progress) {
2684                         ret = btrfs_remove_free_space(block_group,
2685                                                       start, num_bytes);
2686                 } else {
2687                         num_bytes = caching_ctl->progress - start;
2688                         ret = btrfs_remove_free_space(block_group,
2689                                                       start, num_bytes);
2690                         if (ret)
2691                                 goto out_lock;
2692
2693                         num_bytes = (start + num_bytes) -
2694                                 caching_ctl->progress;
2695                         start = caching_ctl->progress;
2696                         ret = btrfs_add_excluded_extent(fs_info, start,
2697                                                         num_bytes);
2698                 }
2699 out_lock:
2700                 mutex_unlock(&caching_ctl->mutex);
2701                 btrfs_put_caching_control(caching_ctl);
2702         }
2703         btrfs_put_block_group(block_group);
2704         return ret;
2705 }
2706
2707 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2708 {
2709         struct btrfs_fs_info *fs_info = eb->fs_info;
2710         struct btrfs_file_extent_item *item;
2711         struct btrfs_key key;
2712         int found_type;
2713         int i;
2714         int ret = 0;
2715
2716         if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2717                 return 0;
2718
2719         for (i = 0; i < btrfs_header_nritems(eb); i++) {
2720                 btrfs_item_key_to_cpu(eb, &key, i);
2721                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2722                         continue;
2723                 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2724                 found_type = btrfs_file_extent_type(eb, item);
2725                 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2726                         continue;
2727                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2728                         continue;
2729                 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2730                 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2731                 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2732                 if (ret)
2733                         break;
2734         }
2735
2736         return ret;
2737 }
2738
2739 static void
2740 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2741 {
2742         atomic_inc(&bg->reservations);
2743 }
2744
2745 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
2746 {
2747         struct btrfs_caching_control *next;
2748         struct btrfs_caching_control *caching_ctl;
2749         struct btrfs_block_group *cache;
2750
2751         down_write(&fs_info->commit_root_sem);
2752
2753         list_for_each_entry_safe(caching_ctl, next,
2754                                  &fs_info->caching_block_groups, list) {
2755                 cache = caching_ctl->block_group;
2756                 if (btrfs_block_group_done(cache)) {
2757                         cache->last_byte_to_unpin = (u64)-1;
2758                         list_del_init(&caching_ctl->list);
2759                         btrfs_put_caching_control(caching_ctl);
2760                 } else {
2761                         cache->last_byte_to_unpin = caching_ctl->progress;
2762                 }
2763         }
2764
2765         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
2766                 fs_info->pinned_extents = &fs_info->freed_extents[1];
2767         else
2768                 fs_info->pinned_extents = &fs_info->freed_extents[0];
2769
2770         up_write(&fs_info->commit_root_sem);
2771
2772         btrfs_update_global_block_rsv(fs_info);
2773 }
2774
2775 /*
2776  * Returns the free cluster for the given space info and sets empty_cluster to
2777  * what it should be based on the mount options.
2778  */
2779 static struct btrfs_free_cluster *
2780 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2781                    struct btrfs_space_info *space_info, u64 *empty_cluster)
2782 {
2783         struct btrfs_free_cluster *ret = NULL;
2784
2785         *empty_cluster = 0;
2786         if (btrfs_mixed_space_info(space_info))
2787                 return ret;
2788
2789         if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2790                 ret = &fs_info->meta_alloc_cluster;
2791                 if (btrfs_test_opt(fs_info, SSD))
2792                         *empty_cluster = SZ_2M;
2793                 else
2794                         *empty_cluster = SZ_64K;
2795         } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2796                    btrfs_test_opt(fs_info, SSD_SPREAD)) {
2797                 *empty_cluster = SZ_2M;
2798                 ret = &fs_info->data_alloc_cluster;
2799         }
2800
2801         return ret;
2802 }
2803
2804 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2805                               u64 start, u64 end,
2806                               const bool return_free_space)
2807 {
2808         struct btrfs_block_group *cache = NULL;
2809         struct btrfs_space_info *space_info;
2810         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2811         struct btrfs_free_cluster *cluster = NULL;
2812         u64 len;
2813         u64 total_unpinned = 0;
2814         u64 empty_cluster = 0;
2815         bool readonly;
2816
2817         while (start <= end) {
2818                 readonly = false;
2819                 if (!cache ||
2820                     start >= cache->start + cache->length) {
2821                         if (cache)
2822                                 btrfs_put_block_group(cache);
2823                         total_unpinned = 0;
2824                         cache = btrfs_lookup_block_group(fs_info, start);
2825                         BUG_ON(!cache); /* Logic error */
2826
2827                         cluster = fetch_cluster_info(fs_info,
2828                                                      cache->space_info,
2829                                                      &empty_cluster);
2830                         empty_cluster <<= 1;
2831                 }
2832
2833                 len = cache->start + cache->length - start;
2834                 len = min(len, end + 1 - start);
2835
2836                 if (start < cache->last_byte_to_unpin) {
2837                         len = min(len, cache->last_byte_to_unpin - start);
2838                         if (return_free_space)
2839                                 btrfs_add_free_space(cache, start, len);
2840                 }
2841
2842                 start += len;
2843                 total_unpinned += len;
2844                 space_info = cache->space_info;
2845
2846                 /*
2847                  * If this space cluster has been marked as fragmented and we've
2848                  * unpinned enough in this block group to potentially allow a
2849                  * cluster to be created inside of it go ahead and clear the
2850                  * fragmented check.
2851                  */
2852                 if (cluster && cluster->fragmented &&
2853                     total_unpinned > empty_cluster) {
2854                         spin_lock(&cluster->lock);
2855                         cluster->fragmented = 0;
2856                         spin_unlock(&cluster->lock);
2857                 }
2858
2859                 spin_lock(&space_info->lock);
2860                 spin_lock(&cache->lock);
2861                 cache->pinned -= len;
2862                 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2863                 space_info->max_extent_size = 0;
2864                 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2865                             -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
2866                 if (cache->ro) {
2867                         space_info->bytes_readonly += len;
2868                         readonly = true;
2869                 }
2870                 spin_unlock(&cache->lock);
2871                 if (!readonly && return_free_space &&
2872                     global_rsv->space_info == space_info) {
2873                         u64 to_add = len;
2874
2875                         spin_lock(&global_rsv->lock);
2876                         if (!global_rsv->full) {
2877                                 to_add = min(len, global_rsv->size -
2878                                              global_rsv->reserved);
2879                                 global_rsv->reserved += to_add;
2880                                 btrfs_space_info_update_bytes_may_use(fs_info,
2881                                                 space_info, to_add);
2882                                 if (global_rsv->reserved >= global_rsv->size)
2883                                         global_rsv->full = 1;
2884                                 len -= to_add;
2885                         }
2886                         spin_unlock(&global_rsv->lock);
2887                         /* Add to any tickets we may have */
2888                         if (len)
2889                                 btrfs_try_granting_tickets(fs_info,
2890                                                            space_info);
2891                 }
2892                 spin_unlock(&space_info->lock);
2893         }
2894
2895         if (cache)
2896                 btrfs_put_block_group(cache);
2897         return 0;
2898 }
2899
2900 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2901 {
2902         struct btrfs_fs_info *fs_info = trans->fs_info;
2903         struct btrfs_block_group *block_group, *tmp;
2904         struct list_head *deleted_bgs;
2905         struct extent_io_tree *unpin;
2906         u64 start;
2907         u64 end;
2908         int ret;
2909
2910         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
2911                 unpin = &fs_info->freed_extents[1];
2912         else
2913                 unpin = &fs_info->freed_extents[0];
2914
2915         while (!trans->aborted) {
2916                 struct extent_state *cached_state = NULL;
2917
2918                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2919                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2920                                             EXTENT_DIRTY, &cached_state);
2921                 if (ret) {
2922                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2923                         break;
2924                 }
2925
2926                 if (btrfs_test_opt(fs_info, DISCARD))
2927                         ret = btrfs_discard_extent(fs_info, start,
2928                                                    end + 1 - start, NULL);
2929
2930                 clear_extent_dirty(unpin, start, end, &cached_state);
2931                 unpin_extent_range(fs_info, start, end, true);
2932                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2933                 free_extent_state(cached_state);
2934                 cond_resched();
2935         }
2936
2937         /*
2938          * Transaction is finished.  We don't need the lock anymore.  We
2939          * do need to clean up the block groups in case of a transaction
2940          * abort.
2941          */
2942         deleted_bgs = &trans->transaction->deleted_bgs;
2943         list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2944                 u64 trimmed = 0;
2945
2946                 ret = -EROFS;
2947                 if (!trans->aborted)
2948                         ret = btrfs_discard_extent(fs_info,
2949                                                    block_group->start,
2950                                                    block_group->length,
2951                                                    &trimmed);
2952
2953                 list_del_init(&block_group->bg_list);
2954                 btrfs_put_block_group_trimming(block_group);
2955                 btrfs_put_block_group(block_group);
2956
2957                 if (ret) {
2958                         const char *errstr = btrfs_decode_error(ret);
2959                         btrfs_warn(fs_info,
2960                            "discard failed while removing blockgroup: errno=%d %s",
2961                                    ret, errstr);
2962                 }
2963         }
2964
2965         return 0;
2966 }
2967
2968 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2969                                struct btrfs_delayed_ref_node *node, u64 parent,
2970                                u64 root_objectid, u64 owner_objectid,
2971                                u64 owner_offset, int refs_to_drop,
2972                                struct btrfs_delayed_extent_op *extent_op)
2973 {
2974         struct btrfs_fs_info *info = trans->fs_info;
2975         struct btrfs_key key;
2976         struct btrfs_path *path;
2977         struct btrfs_root *extent_root = info->extent_root;
2978         struct extent_buffer *leaf;
2979         struct btrfs_extent_item *ei;
2980         struct btrfs_extent_inline_ref *iref;
2981         int ret;
2982         int is_data;
2983         int extent_slot = 0;
2984         int found_extent = 0;
2985         int num_to_del = 1;
2986         u32 item_size;
2987         u64 refs;
2988         u64 bytenr = node->bytenr;
2989         u64 num_bytes = node->num_bytes;
2990         int last_ref = 0;
2991         bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2992
2993         path = btrfs_alloc_path();
2994         if (!path)
2995                 return -ENOMEM;
2996
2997         path->reada = READA_FORWARD;
2998         path->leave_spinning = 1;
2999
3000         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3001         BUG_ON(!is_data && refs_to_drop != 1);
3002
3003         if (is_data)
3004                 skinny_metadata = false;
3005
3006         ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
3007                                     parent, root_objectid, owner_objectid,
3008                                     owner_offset);
3009         if (ret == 0) {
3010                 extent_slot = path->slots[0];
3011                 while (extent_slot >= 0) {
3012                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3013                                               extent_slot);
3014                         if (key.objectid != bytenr)
3015                                 break;
3016                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3017                             key.offset == num_bytes) {
3018                                 found_extent = 1;
3019                                 break;
3020                         }
3021                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
3022                             key.offset == owner_objectid) {
3023                                 found_extent = 1;
3024                                 break;
3025                         }
3026                         if (path->slots[0] - extent_slot > 5)
3027                                 break;
3028                         extent_slot--;
3029                 }
3030
3031                 if (!found_extent) {
3032                         BUG_ON(iref);
3033                         ret = remove_extent_backref(trans, path, NULL,
3034                                                     refs_to_drop,
3035                                                     is_data, &last_ref);
3036                         if (ret) {
3037                                 btrfs_abort_transaction(trans, ret);
3038                                 goto out;
3039                         }
3040                         btrfs_release_path(path);
3041                         path->leave_spinning = 1;
3042
3043                         key.objectid = bytenr;
3044                         key.type = BTRFS_EXTENT_ITEM_KEY;
3045                         key.offset = num_bytes;
3046
3047                         if (!is_data && skinny_metadata) {
3048                                 key.type = BTRFS_METADATA_ITEM_KEY;
3049                                 key.offset = owner_objectid;
3050                         }
3051
3052                         ret = btrfs_search_slot(trans, extent_root,
3053                                                 &key, path, -1, 1);
3054                         if (ret > 0 && skinny_metadata && path->slots[0]) {
3055                                 /*
3056                                  * Couldn't find our skinny metadata item,
3057                                  * see if we have ye olde extent item.
3058                                  */
3059                                 path->slots[0]--;
3060                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
3061                                                       path->slots[0]);
3062                                 if (key.objectid == bytenr &&
3063                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
3064                                     key.offset == num_bytes)
3065                                         ret = 0;
3066                         }
3067
3068                         if (ret > 0 && skinny_metadata) {
3069                                 skinny_metadata = false;
3070                                 key.objectid = bytenr;
3071                                 key.type = BTRFS_EXTENT_ITEM_KEY;
3072                                 key.offset = num_bytes;
3073                                 btrfs_release_path(path);
3074                                 ret = btrfs_search_slot(trans, extent_root,
3075                                                         &key, path, -1, 1);
3076                         }
3077
3078                         if (ret) {
3079                                 btrfs_err(info,
3080                                           "umm, got %d back from search, was looking for %llu",
3081                                           ret, bytenr);
3082                                 if (ret > 0)
3083                                         btrfs_print_leaf(path->nodes[0]);
3084                         }
3085                         if (ret < 0) {
3086                                 btrfs_abort_transaction(trans, ret);
3087                                 goto out;
3088                         }
3089                         extent_slot = path->slots[0];
3090                 }
3091         } else if (WARN_ON(ret == -ENOENT)) {
3092                 btrfs_print_leaf(path->nodes[0]);
3093                 btrfs_err(info,
3094                         "unable to find ref byte nr %llu parent %llu root %llu  owner %llu offset %llu",
3095                         bytenr, parent, root_objectid, owner_objectid,
3096                         owner_offset);
3097                 btrfs_abort_transaction(trans, ret);
3098                 goto out;
3099         } else {
3100                 btrfs_abort_transaction(trans, ret);
3101                 goto out;
3102         }
3103
3104         leaf = path->nodes[0];
3105         item_size = btrfs_item_size_nr(leaf, extent_slot);
3106         if (unlikely(item_size < sizeof(*ei))) {
3107                 ret = -EINVAL;
3108                 btrfs_print_v0_err(info);
3109                 btrfs_abort_transaction(trans, ret);
3110                 goto out;
3111         }
3112         ei = btrfs_item_ptr(leaf, extent_slot,
3113                             struct btrfs_extent_item);
3114         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3115             key.type == BTRFS_EXTENT_ITEM_KEY) {
3116                 struct btrfs_tree_block_info *bi;
3117                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3118                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3119                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3120         }
3121
3122         refs = btrfs_extent_refs(leaf, ei);
3123         if (refs < refs_to_drop) {
3124                 btrfs_err(info,
3125                           "trying to drop %d refs but we only have %Lu for bytenr %Lu",
3126                           refs_to_drop, refs, bytenr);
3127                 ret = -EINVAL;
3128                 btrfs_abort_transaction(trans, ret);
3129                 goto out;
3130         }
3131         refs -= refs_to_drop;
3132
3133         if (refs > 0) {
3134                 if (extent_op)
3135                         __run_delayed_extent_op(extent_op, leaf, ei);
3136                 /*
3137                  * In the case of inline back ref, reference count will
3138                  * be updated by remove_extent_backref
3139                  */
3140                 if (iref) {
3141                         BUG_ON(!found_extent);
3142                 } else {
3143                         btrfs_set_extent_refs(leaf, ei, refs);
3144                         btrfs_mark_buffer_dirty(leaf);
3145                 }
3146                 if (found_extent) {
3147                         ret = remove_extent_backref(trans, path, iref,
3148                                                     refs_to_drop, is_data,
3149                                                     &last_ref);
3150                         if (ret) {
3151                                 btrfs_abort_transaction(trans, ret);
3152                                 goto out;
3153                         }
3154                 }
3155         } else {
3156                 if (found_extent) {
3157                         BUG_ON(is_data && refs_to_drop !=
3158                                extent_data_ref_count(path, iref));
3159                         if (iref) {
3160                                 BUG_ON(path->slots[0] != extent_slot);
3161                         } else {
3162                                 BUG_ON(path->slots[0] != extent_slot + 1);
3163                                 path->slots[0] = extent_slot;
3164                                 num_to_del = 2;
3165                         }
3166                 }
3167
3168                 last_ref = 1;
3169                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3170                                       num_to_del);
3171                 if (ret) {
3172                         btrfs_abort_transaction(trans, ret);
3173                         goto out;
3174                 }
3175                 btrfs_release_path(path);
3176
3177                 if (is_data) {
3178                         ret = btrfs_del_csums(trans, info->csum_root, bytenr,
3179                                               num_bytes);
3180                         if (ret) {
3181                                 btrfs_abort_transaction(trans, ret);
3182                                 goto out;
3183                         }
3184                 }
3185
3186                 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
3187                 if (ret) {
3188                         btrfs_abort_transaction(trans, ret);
3189                         goto out;
3190                 }
3191
3192                 ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0);
3193                 if (ret) {
3194                         btrfs_abort_transaction(trans, ret);
3195                         goto out;
3196                 }
3197         }
3198         btrfs_release_path(path);
3199
3200 out:
3201         btrfs_free_path(path);
3202         return ret;
3203 }
3204
3205 /*
3206  * when we free an block, it is possible (and likely) that we free the last
3207  * delayed ref for that extent as well.  This searches the delayed ref tree for
3208  * a given extent, and if there are no other delayed refs to be processed, it
3209  * removes it from the tree.
3210  */
3211 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3212                                       u64 bytenr)
3213 {
3214         struct btrfs_delayed_ref_head *head;
3215         struct btrfs_delayed_ref_root *delayed_refs;
3216         int ret = 0;
3217
3218         delayed_refs = &trans->transaction->delayed_refs;
3219         spin_lock(&delayed_refs->lock);
3220         head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3221         if (!head)
3222                 goto out_delayed_unlock;
3223
3224         spin_lock(&head->lock);
3225         if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3226                 goto out;
3227
3228         if (cleanup_extent_op(head) != NULL)
3229                 goto out;
3230
3231         /*
3232          * waiting for the lock here would deadlock.  If someone else has it
3233          * locked they are already in the process of dropping it anyway
3234          */
3235         if (!mutex_trylock(&head->mutex))
3236                 goto out;
3237
3238         btrfs_delete_ref_head(delayed_refs, head);
3239         head->processing = 0;
3240
3241         spin_unlock(&head->lock);
3242         spin_unlock(&delayed_refs->lock);
3243
3244         BUG_ON(head->extent_op);
3245         if (head->must_insert_reserved)
3246                 ret = 1;
3247
3248         btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3249         mutex_unlock(&head->mutex);
3250         btrfs_put_delayed_ref_head(head);
3251         return ret;
3252 out:
3253         spin_unlock(&head->lock);
3254
3255 out_delayed_unlock:
3256         spin_unlock(&delayed_refs->lock);
3257         return 0;
3258 }
3259
3260 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3261                            struct btrfs_root *root,
3262                            struct extent_buffer *buf,
3263                            u64 parent, int last_ref)
3264 {
3265         struct btrfs_fs_info *fs_info = root->fs_info;
3266         struct btrfs_ref generic_ref = { 0 };
3267         int pin = 1;
3268         int ret;
3269
3270         btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3271                                buf->start, buf->len, parent);
3272         btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3273                             root->root_key.objectid);
3274
3275         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3276                 int old_ref_mod, new_ref_mod;
3277
3278                 btrfs_ref_tree_mod(fs_info, &generic_ref);
3279                 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL,
3280                                                  &old_ref_mod, &new_ref_mod);
3281                 BUG_ON(ret); /* -ENOMEM */
3282                 pin = old_ref_mod >= 0 && new_ref_mod < 0;
3283         }
3284
3285         if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3286                 struct btrfs_block_group *cache;
3287
3288                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3289                         ret = check_ref_cleanup(trans, buf->start);
3290                         if (!ret)
3291                                 goto out;
3292                 }
3293
3294                 pin = 0;
3295                 cache = btrfs_lookup_block_group(fs_info, buf->start);
3296
3297                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3298                         pin_down_extent(cache, buf->start, buf->len, 1);
3299                         btrfs_put_block_group(cache);
3300                         goto out;
3301                 }
3302
3303                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3304
3305                 btrfs_add_free_space(cache, buf->start, buf->len);
3306                 btrfs_free_reserved_bytes(cache, buf->len, 0);
3307                 btrfs_put_block_group(cache);
3308                 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3309         }
3310 out:
3311         if (pin)
3312                 add_pinned_bytes(fs_info, &generic_ref);
3313
3314         if (last_ref) {
3315                 /*
3316                  * Deleting the buffer, clear the corrupt flag since it doesn't
3317                  * matter anymore.
3318                  */
3319                 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3320         }
3321 }
3322
3323 /* Can return -ENOMEM */
3324 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3325 {
3326         struct btrfs_fs_info *fs_info = trans->fs_info;
3327         int old_ref_mod, new_ref_mod;
3328         int ret;
3329
3330         if (btrfs_is_testing(fs_info))
3331                 return 0;
3332
3333         /*
3334          * tree log blocks never actually go into the extent allocation
3335          * tree, just update pinning info and exit early.
3336          */
3337         if ((ref->type == BTRFS_REF_METADATA &&
3338              ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3339             (ref->type == BTRFS_REF_DATA &&
3340              ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) {
3341                 /* unlocks the pinned mutex */
3342                 btrfs_pin_extent(fs_info, ref->bytenr, ref->len, 1);
3343                 old_ref_mod = new_ref_mod = 0;
3344                 ret = 0;
3345         } else if (ref->type == BTRFS_REF_METADATA) {
3346                 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL,
3347                                                  &old_ref_mod, &new_ref_mod);
3348         } else {
3349                 ret = btrfs_add_delayed_data_ref(trans, ref, 0,
3350                                                  &old_ref_mod, &new_ref_mod);
3351         }
3352
3353         if (!((ref->type == BTRFS_REF_METADATA &&
3354                ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3355               (ref->type == BTRFS_REF_DATA &&
3356                ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)))
3357                 btrfs_ref_tree_mod(fs_info, ref);
3358
3359         if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
3360                 add_pinned_bytes(fs_info, ref);
3361
3362         return ret;
3363 }
3364
3365 enum btrfs_loop_type {
3366         LOOP_CACHING_NOWAIT,
3367         LOOP_CACHING_WAIT,
3368         LOOP_ALLOC_CHUNK,
3369         LOOP_NO_EMPTY_SIZE,
3370 };
3371
3372 static inline void
3373 btrfs_lock_block_group(struct btrfs_block_group *cache,
3374                        int delalloc)
3375 {
3376         if (delalloc)
3377                 down_read(&cache->data_rwsem);
3378 }
3379
3380 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3381                        int delalloc)
3382 {
3383         btrfs_get_block_group(cache);
3384         if (delalloc)
3385                 down_read(&cache->data_rwsem);
3386 }
3387
3388 static struct btrfs_block_group *btrfs_lock_cluster(
3389                    struct btrfs_block_group *block_group,
3390                    struct btrfs_free_cluster *cluster,
3391                    int delalloc)
3392 {
3393         struct btrfs_block_group *used_bg = NULL;
3394
3395         spin_lock(&cluster->refill_lock);
3396         while (1) {
3397                 used_bg = cluster->block_group;
3398                 if (!used_bg)
3399                         return NULL;
3400
3401                 if (used_bg == block_group)
3402                         return used_bg;
3403
3404                 btrfs_get_block_group(used_bg);
3405
3406                 if (!delalloc)
3407                         return used_bg;
3408
3409                 if (down_read_trylock(&used_bg->data_rwsem))
3410                         return used_bg;
3411
3412                 spin_unlock(&cluster->refill_lock);
3413
3414                 /* We should only have one-level nested. */
3415                 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3416
3417                 spin_lock(&cluster->refill_lock);
3418                 if (used_bg == cluster->block_group)
3419                         return used_bg;
3420
3421                 up_read(&used_bg->data_rwsem);
3422                 btrfs_put_block_group(used_bg);
3423         }
3424 }
3425
3426 static inline void
3427 btrfs_release_block_group(struct btrfs_block_group *cache,
3428                          int delalloc)
3429 {
3430         if (delalloc)
3431                 up_read(&cache->data_rwsem);
3432         btrfs_put_block_group(cache);
3433 }
3434
3435 /*
3436  * Structure used internally for find_free_extent() function.  Wraps needed
3437  * parameters.
3438  */
3439 struct find_free_extent_ctl {
3440         /* Basic allocation info */
3441         u64 ram_bytes;
3442         u64 num_bytes;
3443         u64 empty_size;
3444         u64 flags;
3445         int delalloc;
3446
3447         /* Where to start the search inside the bg */
3448         u64 search_start;
3449
3450         /* For clustered allocation */
3451         u64 empty_cluster;
3452
3453         bool have_caching_bg;
3454         bool orig_have_caching_bg;
3455
3456         /* RAID index, converted from flags */
3457         int index;
3458
3459         /*
3460          * Current loop number, check find_free_extent_update_loop() for details
3461          */
3462         int loop;
3463
3464         /*
3465          * Whether we're refilling a cluster, if true we need to re-search
3466          * current block group but don't try to refill the cluster again.
3467          */
3468         bool retry_clustered;
3469
3470         /*
3471          * Whether we're updating free space cache, if true we need to re-search
3472          * current block group but don't try updating free space cache again.
3473          */
3474         bool retry_unclustered;
3475
3476         /* If current block group is cached */
3477         int cached;
3478
3479         /* Max contiguous hole found */
3480         u64 max_extent_size;
3481
3482         /* Total free space from free space cache, not always contiguous */
3483         u64 total_free_space;
3484
3485         /* Found result */
3486         u64 found_offset;
3487 };
3488
3489
3490 /*
3491  * Helper function for find_free_extent().
3492  *
3493  * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3494  * Return -EAGAIN to inform caller that we need to re-search this block group
3495  * Return >0 to inform caller that we find nothing
3496  * Return 0 means we have found a location and set ffe_ctl->found_offset.
3497  */
3498 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3499                 struct btrfs_free_cluster *last_ptr,
3500                 struct find_free_extent_ctl *ffe_ctl,
3501                 struct btrfs_block_group **cluster_bg_ret)
3502 {
3503         struct btrfs_block_group *cluster_bg;
3504         u64 aligned_cluster;
3505         u64 offset;
3506         int ret;
3507
3508         cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3509         if (!cluster_bg)
3510                 goto refill_cluster;
3511         if (cluster_bg != bg && (cluster_bg->ro ||
3512             !block_group_bits(cluster_bg, ffe_ctl->flags)))
3513                 goto release_cluster;
3514
3515         offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3516                         ffe_ctl->num_bytes, cluster_bg->start,
3517                         &ffe_ctl->max_extent_size);
3518         if (offset) {
3519                 /* We have a block, we're done */
3520                 spin_unlock(&last_ptr->refill_lock);
3521                 trace_btrfs_reserve_extent_cluster(cluster_bg,
3522                                 ffe_ctl->search_start, ffe_ctl->num_bytes);
3523                 *cluster_bg_ret = cluster_bg;
3524                 ffe_ctl->found_offset = offset;
3525                 return 0;
3526         }
3527         WARN_ON(last_ptr->block_group != cluster_bg);
3528
3529 release_cluster:
3530         /*
3531          * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3532          * lets just skip it and let the allocator find whatever block it can
3533          * find. If we reach this point, we will have tried the cluster
3534          * allocator plenty of times and not have found anything, so we are
3535          * likely way too fragmented for the clustering stuff to find anything.
3536          *
3537          * However, if the cluster is taken from the current block group,
3538          * release the cluster first, so that we stand a better chance of
3539          * succeeding in the unclustered allocation.
3540          */
3541         if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3542                 spin_unlock(&last_ptr->refill_lock);
3543                 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3544                 return -ENOENT;
3545         }
3546
3547         /* This cluster didn't work out, free it and start over */
3548         btrfs_return_cluster_to_free_space(NULL, last_ptr);
3549
3550         if (cluster_bg != bg)
3551                 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3552
3553 refill_cluster:
3554         if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3555                 spin_unlock(&last_ptr->refill_lock);
3556                 return -ENOENT;
3557         }
3558
3559         aligned_cluster = max_t(u64,
3560                         ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3561                         bg->full_stripe_len);
3562         ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3563                         ffe_ctl->num_bytes, aligned_cluster);
3564         if (ret == 0) {
3565                 /* Now pull our allocation out of this cluster */
3566                 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3567                                 ffe_ctl->num_bytes, ffe_ctl->search_start,
3568                                 &ffe_ctl->max_extent_size);
3569                 if (offset) {
3570                         /* We found one, proceed */
3571                         spin_unlock(&last_ptr->refill_lock);
3572                         trace_btrfs_reserve_extent_cluster(bg,
3573                                         ffe_ctl->search_start,
3574                                         ffe_ctl->num_bytes);
3575                         ffe_ctl->found_offset = offset;
3576                         return 0;
3577                 }
3578         } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3579                    !ffe_ctl->retry_clustered) {
3580                 spin_unlock(&last_ptr->refill_lock);
3581
3582                 ffe_ctl->retry_clustered = true;
3583                 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3584                                 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3585                 return -EAGAIN;
3586         }
3587         /*
3588          * At this point we either didn't find a cluster or we weren't able to
3589          * allocate a block from our cluster.  Free the cluster we've been
3590          * trying to use, and go to the next block group.
3591          */
3592         btrfs_return_cluster_to_free_space(NULL, last_ptr);
3593         spin_unlock(&last_ptr->refill_lock);
3594         return 1;
3595 }
3596
3597 /*
3598  * Return >0 to inform caller that we find nothing
3599  * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3600  * Return -EAGAIN to inform caller that we need to re-search this block group
3601  */
3602 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3603                 struct btrfs_free_cluster *last_ptr,
3604                 struct find_free_extent_ctl *ffe_ctl)
3605 {
3606         u64 offset;
3607
3608         /*
3609          * We are doing an unclustered allocation, set the fragmented flag so
3610          * we don't bother trying to setup a cluster again until we get more
3611          * space.
3612          */
3613         if (unlikely(last_ptr)) {
3614                 spin_lock(&last_ptr->lock);
3615                 last_ptr->fragmented = 1;
3616                 spin_unlock(&last_ptr->lock);
3617         }
3618         if (ffe_ctl->cached) {
3619                 struct btrfs_free_space_ctl *free_space_ctl;
3620
3621                 free_space_ctl = bg->free_space_ctl;
3622                 spin_lock(&free_space_ctl->tree_lock);
3623                 if (free_space_ctl->free_space <
3624                     ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3625                     ffe_ctl->empty_size) {
3626                         ffe_ctl->total_free_space = max_t(u64,
3627                                         ffe_ctl->total_free_space,
3628                                         free_space_ctl->free_space);
3629                         spin_unlock(&free_space_ctl->tree_lock);
3630                         return 1;
3631                 }
3632                 spin_unlock(&free_space_ctl->tree_lock);
3633         }
3634
3635         offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3636                         ffe_ctl->num_bytes, ffe_ctl->empty_size,
3637                         &ffe_ctl->max_extent_size);
3638
3639         /*
3640          * If we didn't find a chunk, and we haven't failed on this block group
3641          * before, and this block group is in the middle of caching and we are
3642          * ok with waiting, then go ahead and wait for progress to be made, and
3643          * set @retry_unclustered to true.
3644          *
3645          * If @retry_unclustered is true then we've already waited on this
3646          * block group once and should move on to the next block group.
3647          */
3648         if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3649             ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
3650                 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3651                                                       ffe_ctl->empty_size);
3652                 ffe_ctl->retry_unclustered = true;
3653                 return -EAGAIN;
3654         } else if (!offset) {
3655                 return 1;
3656         }
3657         ffe_ctl->found_offset = offset;
3658         return 0;
3659 }
3660
3661 /*
3662  * Return >0 means caller needs to re-search for free extent
3663  * Return 0 means we have the needed free extent.
3664  * Return <0 means we failed to locate any free extent.
3665  */
3666 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
3667                                         struct btrfs_free_cluster *last_ptr,
3668                                         struct btrfs_key *ins,
3669                                         struct find_free_extent_ctl *ffe_ctl,
3670                                         int full_search, bool use_cluster)
3671 {
3672         struct btrfs_root *root = fs_info->extent_root;
3673         int ret;
3674
3675         if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
3676             ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
3677                 ffe_ctl->orig_have_caching_bg = true;
3678
3679         if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
3680             ffe_ctl->have_caching_bg)
3681                 return 1;
3682
3683         if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
3684                 return 1;
3685
3686         if (ins->objectid) {
3687                 if (!use_cluster && last_ptr) {
3688                         spin_lock(&last_ptr->lock);
3689                         last_ptr->window_start = ins->objectid;
3690                         spin_unlock(&last_ptr->lock);
3691                 }
3692                 return 0;
3693         }
3694
3695         /*
3696          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
3697          *                      caching kthreads as we move along
3698          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3699          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3700          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3701          *                     again
3702          */
3703         if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
3704                 ffe_ctl->index = 0;
3705                 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
3706                         /*
3707                          * We want to skip the LOOP_CACHING_WAIT step if we
3708                          * don't have any uncached bgs and we've already done a
3709                          * full search through.
3710                          */
3711                         if (ffe_ctl->orig_have_caching_bg || !full_search)
3712                                 ffe_ctl->loop = LOOP_CACHING_WAIT;
3713                         else
3714                                 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
3715                 } else {
3716                         ffe_ctl->loop++;
3717                 }
3718
3719                 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
3720                         struct btrfs_trans_handle *trans;
3721                         int exist = 0;
3722
3723                         trans = current->journal_info;
3724                         if (trans)
3725                                 exist = 1;
3726                         else
3727                                 trans = btrfs_join_transaction(root);
3728
3729                         if (IS_ERR(trans)) {
3730                                 ret = PTR_ERR(trans);
3731                                 return ret;
3732                         }
3733
3734                         ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
3735                                                 CHUNK_ALLOC_FORCE);
3736
3737                         /*
3738                          * If we can't allocate a new chunk we've already looped
3739                          * through at least once, move on to the NO_EMPTY_SIZE
3740                          * case.
3741                          */
3742                         if (ret == -ENOSPC)
3743                                 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
3744
3745                         /* Do not bail out on ENOSPC since we can do more. */
3746                         if (ret < 0 && ret != -ENOSPC)
3747                                 btrfs_abort_transaction(trans, ret);
3748                         else
3749                                 ret = 0;
3750                         if (!exist)
3751                                 btrfs_end_transaction(trans);
3752                         if (ret)
3753                                 return ret;
3754                 }
3755
3756                 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
3757                         /*
3758                          * Don't loop again if we already have no empty_size and
3759                          * no empty_cluster.
3760                          */
3761                         if (ffe_ctl->empty_size == 0 &&
3762                             ffe_ctl->empty_cluster == 0)
3763                                 return -ENOSPC;
3764                         ffe_ctl->empty_size = 0;
3765                         ffe_ctl->empty_cluster = 0;
3766                 }
3767                 return 1;
3768         }
3769         return -ENOSPC;
3770 }
3771
3772 /*
3773  * walks the btree of allocated extents and find a hole of a given size.
3774  * The key ins is changed to record the hole:
3775  * ins->objectid == start position
3776  * ins->flags = BTRFS_EXTENT_ITEM_KEY
3777  * ins->offset == the size of the hole.
3778  * Any available blocks before search_start are skipped.
3779  *
3780  * If there is no suitable free space, we will record the max size of
3781  * the free space extent currently.
3782  *
3783  * The overall logic and call chain:
3784  *
3785  * find_free_extent()
3786  * |- Iterate through all block groups
3787  * |  |- Get a valid block group
3788  * |  |- Try to do clustered allocation in that block group
3789  * |  |- Try to do unclustered allocation in that block group
3790  * |  |- Check if the result is valid
3791  * |  |  |- If valid, then exit
3792  * |  |- Jump to next block group
3793  * |
3794  * |- Push harder to find free extents
3795  *    |- If not found, re-iterate all block groups
3796  */
3797 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
3798                                 u64 ram_bytes, u64 num_bytes, u64 empty_size,
3799                                 u64 hint_byte, struct btrfs_key *ins,
3800                                 u64 flags, int delalloc)
3801 {
3802         int ret = 0;
3803         int cache_block_group_error = 0;
3804         struct btrfs_free_cluster *last_ptr = NULL;
3805         struct btrfs_block_group *block_group = NULL;
3806         struct find_free_extent_ctl ffe_ctl = {0};
3807         struct btrfs_space_info *space_info;
3808         bool use_cluster = true;
3809         bool full_search = false;
3810
3811         WARN_ON(num_bytes < fs_info->sectorsize);
3812
3813         ffe_ctl.ram_bytes = ram_bytes;
3814         ffe_ctl.num_bytes = num_bytes;
3815         ffe_ctl.empty_size = empty_size;
3816         ffe_ctl.flags = flags;
3817         ffe_ctl.search_start = 0;
3818         ffe_ctl.retry_clustered = false;
3819         ffe_ctl.retry_unclustered = false;
3820         ffe_ctl.delalloc = delalloc;
3821         ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
3822         ffe_ctl.have_caching_bg = false;
3823         ffe_ctl.orig_have_caching_bg = false;
3824         ffe_ctl.found_offset = 0;
3825
3826         ins->type = BTRFS_EXTENT_ITEM_KEY;
3827         ins->objectid = 0;
3828         ins->offset = 0;
3829
3830         trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
3831
3832         space_info = btrfs_find_space_info(fs_info, flags);
3833         if (!space_info) {
3834                 btrfs_err(fs_info, "No space info for %llu", flags);
3835                 return -ENOSPC;
3836         }
3837
3838         /*
3839          * If our free space is heavily fragmented we may not be able to make
3840          * big contiguous allocations, so instead of doing the expensive search
3841          * for free space, simply return ENOSPC with our max_extent_size so we
3842          * can go ahead and search for a more manageable chunk.
3843          *
3844          * If our max_extent_size is large enough for our allocation simply
3845          * disable clustering since we will likely not be able to find enough
3846          * space to create a cluster and induce latency trying.
3847          */
3848         if (unlikely(space_info->max_extent_size)) {
3849                 spin_lock(&space_info->lock);
3850                 if (space_info->max_extent_size &&
3851                     num_bytes > space_info->max_extent_size) {
3852                         ins->offset = space_info->max_extent_size;
3853                         spin_unlock(&space_info->lock);
3854                         return -ENOSPC;
3855                 } else if (space_info->max_extent_size) {
3856                         use_cluster = false;
3857                 }
3858                 spin_unlock(&space_info->lock);
3859         }
3860
3861         last_ptr = fetch_cluster_info(fs_info, space_info,
3862                                       &ffe_ctl.empty_cluster);
3863         if (last_ptr) {
3864                 spin_lock(&last_ptr->lock);
3865                 if (last_ptr->block_group)
3866                         hint_byte = last_ptr->window_start;
3867                 if (last_ptr->fragmented) {
3868                         /*
3869                          * We still set window_start so we can keep track of the
3870                          * last place we found an allocation to try and save
3871                          * some time.
3872                          */
3873                         hint_byte = last_ptr->window_start;
3874                         use_cluster = false;
3875                 }
3876                 spin_unlock(&last_ptr->lock);
3877         }
3878
3879         ffe_ctl.search_start = max(ffe_ctl.search_start,
3880                                    first_logical_byte(fs_info, 0));
3881         ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
3882         if (ffe_ctl.search_start == hint_byte) {
3883                 block_group = btrfs_lookup_block_group(fs_info,
3884                                                        ffe_ctl.search_start);
3885                 /*
3886                  * we don't want to use the block group if it doesn't match our
3887                  * allocation bits, or if its not cached.
3888                  *
3889                  * However if we are re-searching with an ideal block group
3890                  * picked out then we don't care that the block group is cached.
3891                  */
3892                 if (block_group && block_group_bits(block_group, flags) &&
3893                     block_group->cached != BTRFS_CACHE_NO) {
3894                         down_read(&space_info->groups_sem);
3895                         if (list_empty(&block_group->list) ||
3896                             block_group->ro) {
3897                                 /*
3898                                  * someone is removing this block group,
3899                                  * we can't jump into the have_block_group
3900                                  * target because our list pointers are not
3901                                  * valid
3902                                  */
3903                                 btrfs_put_block_group(block_group);
3904                                 up_read(&space_info->groups_sem);
3905                         } else {
3906                                 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
3907                                                 block_group->flags);
3908                                 btrfs_lock_block_group(block_group, delalloc);
3909                                 goto have_block_group;
3910                         }
3911                 } else if (block_group) {
3912                         btrfs_put_block_group(block_group);
3913                 }
3914         }
3915 search:
3916         ffe_ctl.have_caching_bg = false;
3917         if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
3918             ffe_ctl.index == 0)
3919                 full_search = true;
3920         down_read(&space_info->groups_sem);
3921         list_for_each_entry(block_group,
3922                             &space_info->block_groups[ffe_ctl.index], list) {
3923                 /* If the block group is read-only, we can skip it entirely. */
3924                 if (unlikely(block_group->ro))
3925                         continue;
3926
3927                 btrfs_grab_block_group(block_group, delalloc);
3928                 ffe_ctl.search_start = block_group->start;
3929
3930                 /*
3931                  * this can happen if we end up cycling through all the
3932                  * raid types, but we want to make sure we only allocate
3933                  * for the proper type.
3934                  */
3935                 if (!block_group_bits(block_group, flags)) {
3936                         u64 extra = BTRFS_BLOCK_GROUP_DUP |
3937                                 BTRFS_BLOCK_GROUP_RAID1_MASK |
3938                                 BTRFS_BLOCK_GROUP_RAID56_MASK |
3939                                 BTRFS_BLOCK_GROUP_RAID10;
3940
3941                         /*
3942                          * if they asked for extra copies and this block group
3943                          * doesn't provide them, bail.  This does allow us to
3944                          * fill raid0 from raid1.
3945                          */
3946                         if ((flags & extra) && !(block_group->flags & extra))
3947                                 goto loop;
3948
3949                         /*
3950                          * This block group has different flags than we want.
3951                          * It's possible that we have MIXED_GROUP flag but no
3952                          * block group is mixed.  Just skip such block group.
3953                          */
3954                         btrfs_release_block_group(block_group, delalloc);
3955                         continue;
3956                 }
3957
3958 have_block_group:
3959                 ffe_ctl.cached = btrfs_block_group_done(block_group);
3960                 if (unlikely(!ffe_ctl.cached)) {
3961                         ffe_ctl.have_caching_bg = true;
3962                         ret = btrfs_cache_block_group(block_group, 0);
3963
3964                         /*
3965                          * If we get ENOMEM here or something else we want to
3966                          * try other block groups, because it may not be fatal.
3967                          * However if we can't find anything else we need to
3968                          * save our return here so that we return the actual
3969                          * error that caused problems, not ENOSPC.
3970                          */
3971                         if (ret < 0) {
3972                                 if (!cache_block_group_error)
3973                                         cache_block_group_error = ret;
3974                                 ret = 0;
3975                                 goto loop;
3976                         }
3977                         ret = 0;
3978                 }
3979
3980                 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
3981                         goto loop;
3982
3983                 /*
3984                  * Ok we want to try and use the cluster allocator, so
3985                  * lets look there
3986                  */
3987                 if (last_ptr && use_cluster) {
3988                         struct btrfs_block_group *cluster_bg = NULL;
3989
3990                         ret = find_free_extent_clustered(block_group, last_ptr,
3991                                                          &ffe_ctl, &cluster_bg);
3992
3993                         if (ret == 0) {
3994                                 if (cluster_bg && cluster_bg != block_group) {
3995                                         btrfs_release_block_group(block_group,
3996                                                                   delalloc);
3997                                         block_group = cluster_bg;
3998                                 }
3999                                 goto checks;
4000                         } else if (ret == -EAGAIN) {
4001                                 goto have_block_group;
4002                         } else if (ret > 0) {
4003                                 goto loop;
4004                         }
4005                         /* ret == -ENOENT case falls through */
4006                 }
4007
4008                 ret = find_free_extent_unclustered(block_group, last_ptr,
4009                                                    &ffe_ctl);
4010                 if (ret == -EAGAIN)
4011                         goto have_block_group;
4012                 else if (ret > 0)
4013                         goto loop;
4014                 /* ret == 0 case falls through */
4015 checks:
4016                 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
4017                                              fs_info->stripesize);
4018
4019                 /* move on to the next group */
4020                 if (ffe_ctl.search_start + num_bytes >
4021                     block_group->start + block_group->length) {
4022                         btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4023                                              num_bytes);
4024                         goto loop;
4025                 }
4026
4027                 if (ffe_ctl.found_offset < ffe_ctl.search_start)
4028                         btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4029                                 ffe_ctl.search_start - ffe_ctl.found_offset);
4030
4031                 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
4032                                 num_bytes, delalloc);
4033                 if (ret == -EAGAIN) {
4034                         btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4035                                              num_bytes);
4036                         goto loop;
4037                 }
4038                 btrfs_inc_block_group_reservations(block_group);
4039
4040                 /* we are all good, lets return */
4041                 ins->objectid = ffe_ctl.search_start;
4042                 ins->offset = num_bytes;
4043
4044                 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
4045                                            num_bytes);
4046                 btrfs_release_block_group(block_group, delalloc);
4047                 break;
4048 loop:
4049                 ffe_ctl.retry_clustered = false;
4050                 ffe_ctl.retry_unclustered = false;
4051                 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
4052                        ffe_ctl.index);
4053                 btrfs_release_block_group(block_group, delalloc);
4054                 cond_resched();
4055         }
4056         up_read(&space_info->groups_sem);
4057
4058         ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
4059                                            full_search, use_cluster);
4060         if (ret > 0)
4061                 goto search;
4062
4063         if (ret == -ENOSPC && !cache_block_group_error) {
4064                 /*
4065                  * Use ffe_ctl->total_free_space as fallback if we can't find
4066                  * any contiguous hole.
4067                  */
4068                 if (!ffe_ctl.max_extent_size)
4069                         ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
4070                 spin_lock(&space_info->lock);
4071                 space_info->max_extent_size = ffe_ctl.max_extent_size;
4072                 spin_unlock(&space_info->lock);
4073                 ins->offset = ffe_ctl.max_extent_size;
4074         } else if (ret == -ENOSPC) {
4075                 ret = cache_block_group_error;
4076         }
4077         return ret;
4078 }
4079
4080 /*
4081  * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4082  *                        hole that is at least as big as @num_bytes.
4083  *
4084  * @root           -    The root that will contain this extent
4085  *
4086  * @ram_bytes      -    The amount of space in ram that @num_bytes take. This
4087  *                      is used for accounting purposes. This value differs
4088  *                      from @num_bytes only in the case of compressed extents.
4089  *
4090  * @num_bytes      -    Number of bytes to allocate on-disk.
4091  *
4092  * @min_alloc_size -    Indicates the minimum amount of space that the
4093  *                      allocator should try to satisfy. In some cases
4094  *                      @num_bytes may be larger than what is required and if
4095  *                      the filesystem is fragmented then allocation fails.
4096  *                      However, the presence of @min_alloc_size gives a
4097  *                      chance to try and satisfy the smaller allocation.
4098  *
4099  * @empty_size     -    A hint that you plan on doing more COW. This is the
4100  *                      size in bytes the allocator should try to find free
4101  *                      next to the block it returns.  This is just a hint and
4102  *                      may be ignored by the allocator.
4103  *
4104  * @hint_byte      -    Hint to the allocator to start searching above the byte
4105  *                      address passed. It might be ignored.
4106  *
4107  * @ins            -    This key is modified to record the found hole. It will
4108  *                      have the following values:
4109  *                      ins->objectid == start position
4110  *                      ins->flags = BTRFS_EXTENT_ITEM_KEY
4111  *                      ins->offset == the size of the hole.
4112  *
4113  * @is_data        -    Boolean flag indicating whether an extent is
4114  *                      allocated for data (true) or metadata (false)
4115  *
4116  * @delalloc       -    Boolean flag indicating whether this allocation is for
4117  *                      delalloc or not. If 'true' data_rwsem of block groups
4118  *                      is going to be acquired.
4119  *
4120  *
4121  * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4122  * case -ENOSPC is returned then @ins->offset will contain the size of the
4123  * largest available hole the allocator managed to find.
4124  */
4125 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4126                          u64 num_bytes, u64 min_alloc_size,
4127                          u64 empty_size, u64 hint_byte,
4128                          struct btrfs_key *ins, int is_data, int delalloc)
4129 {
4130         struct btrfs_fs_info *fs_info = root->fs_info;
4131         bool final_tried = num_bytes == min_alloc_size;
4132         u64 flags;
4133         int ret;
4134
4135         flags = get_alloc_profile_by_root(root, is_data);
4136 again:
4137         WARN_ON(num_bytes < fs_info->sectorsize);
4138         ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
4139                                hint_byte, ins, flags, delalloc);
4140         if (!ret && !is_data) {
4141                 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4142         } else if (ret == -ENOSPC) {
4143                 if (!final_tried && ins->offset) {
4144                         num_bytes = min(num_bytes >> 1, ins->offset);
4145                         num_bytes = round_down(num_bytes,
4146                                                fs_info->sectorsize);
4147                         num_bytes = max(num_bytes, min_alloc_size);
4148                         ram_bytes = num_bytes;
4149                         if (num_bytes == min_alloc_size)
4150                                 final_tried = true;
4151                         goto again;
4152                 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4153                         struct btrfs_space_info *sinfo;
4154
4155                         sinfo = btrfs_find_space_info(fs_info, flags);
4156                         btrfs_err(fs_info,
4157                                   "allocation failed flags %llu, wanted %llu",
4158                                   flags, num_bytes);
4159                         if (sinfo)
4160                                 btrfs_dump_space_info(fs_info, sinfo,
4161                                                       num_bytes, 1);
4162                 }
4163         }
4164
4165         return ret;
4166 }
4167
4168 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4169                                         u64 start, u64 len,
4170                                         int pin, int delalloc)
4171 {
4172         struct btrfs_block_group *cache;
4173         int ret = 0;
4174
4175         cache = btrfs_lookup_block_group(fs_info, start);
4176         if (!cache) {
4177                 btrfs_err(fs_info, "Unable to find block group for %llu",
4178                           start);
4179                 return -ENOSPC;
4180         }
4181
4182         if (pin)
4183                 pin_down_extent(cache, start, len, 1);
4184         else {
4185                 if (btrfs_test_opt(fs_info, DISCARD))
4186                         ret = btrfs_discard_extent(fs_info, start, len, NULL);
4187                 btrfs_add_free_space(cache, start, len);
4188                 btrfs_free_reserved_bytes(cache, len, delalloc);
4189                 trace_btrfs_reserved_extent_free(fs_info, start, len);
4190         }
4191
4192         btrfs_put_block_group(cache);
4193         return ret;
4194 }
4195
4196 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4197                                u64 start, u64 len, int delalloc)
4198 {
4199         return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
4200 }
4201
4202 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
4203                                        u64 start, u64 len)
4204 {
4205         return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
4206 }
4207
4208 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4209                                       u64 parent, u64 root_objectid,
4210                                       u64 flags, u64 owner, u64 offset,
4211                                       struct btrfs_key *ins, int ref_mod)
4212 {
4213         struct btrfs_fs_info *fs_info = trans->fs_info;
4214         int ret;
4215         struct btrfs_extent_item *extent_item;
4216         struct btrfs_extent_inline_ref *iref;
4217         struct btrfs_path *path;
4218         struct extent_buffer *leaf;
4219         int type;
4220         u32 size;
4221
4222         if (parent > 0)
4223                 type = BTRFS_SHARED_DATA_REF_KEY;
4224         else
4225                 type = BTRFS_EXTENT_DATA_REF_KEY;
4226
4227         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4228
4229         path = btrfs_alloc_path();
4230         if (!path)
4231                 return -ENOMEM;
4232
4233         path->leave_spinning = 1;
4234         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4235                                       ins, size);
4236         if (ret) {
4237                 btrfs_free_path(path);
4238                 return ret;
4239         }
4240
4241         leaf = path->nodes[0];
4242         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4243                                      struct btrfs_extent_item);
4244         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4245         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4246         btrfs_set_extent_flags(leaf, extent_item,
4247                                flags | BTRFS_EXTENT_FLAG_DATA);
4248
4249         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4250         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4251         if (parent > 0) {
4252                 struct btrfs_shared_data_ref *ref;
4253                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4254                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4255                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4256         } else {
4257                 struct btrfs_extent_data_ref *ref;
4258                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4259                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4260                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4261                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4262                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4263         }
4264
4265         btrfs_mark_buffer_dirty(path->nodes[0]);
4266         btrfs_free_path(path);
4267
4268         ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
4269         if (ret)
4270                 return ret;
4271
4272         ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1);
4273         if (ret) { /* -ENOENT, logic error */
4274                 btrfs_err(fs_info, "update block group failed for %llu %llu",
4275                         ins->objectid, ins->offset);
4276                 BUG();
4277         }
4278         trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
4279         return ret;
4280 }
4281
4282 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4283                                      struct btrfs_delayed_ref_node *node,
4284                                      struct btrfs_delayed_extent_op *extent_op)
4285 {
4286         struct btrfs_fs_info *fs_info = trans->fs_info;
4287         int ret;
4288         struct btrfs_extent_item *extent_item;
4289         struct btrfs_key extent_key;
4290         struct btrfs_tree_block_info *block_info;
4291         struct btrfs_extent_inline_ref *iref;
4292         struct btrfs_path *path;
4293         struct extent_buffer *leaf;
4294         struct btrfs_delayed_tree_ref *ref;
4295         u32 size = sizeof(*extent_item) + sizeof(*iref);
4296         u64 num_bytes;
4297         u64 flags = extent_op->flags_to_set;
4298         bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4299
4300         ref = btrfs_delayed_node_to_tree_ref(node);
4301
4302         extent_key.objectid = node->bytenr;
4303         if (skinny_metadata) {
4304                 extent_key.offset = ref->level;
4305                 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4306                 num_bytes = fs_info->nodesize;
4307         } else {
4308                 extent_key.offset = node->num_bytes;
4309                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4310                 size += sizeof(*block_info);
4311                 num_bytes = node->num_bytes;
4312         }
4313
4314         path = btrfs_alloc_path();
4315         if (!path)
4316                 return -ENOMEM;
4317
4318         path->leave_spinning = 1;
4319         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4320                                       &extent_key, size);
4321         if (ret) {
4322                 btrfs_free_path(path);
4323                 return ret;
4324         }
4325
4326         leaf = path->nodes[0];
4327         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4328                                      struct btrfs_extent_item);
4329         btrfs_set_extent_refs(leaf, extent_item, 1);
4330         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4331         btrfs_set_extent_flags(leaf, extent_item,
4332                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4333
4334         if (skinny_metadata) {
4335                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4336         } else {
4337                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4338                 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4339                 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4340                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4341         }
4342
4343         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4344                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4345                 btrfs_set_extent_inline_ref_type(leaf, iref,
4346                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4347                 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4348         } else {
4349                 btrfs_set_extent_inline_ref_type(leaf, iref,
4350                                                  BTRFS_TREE_BLOCK_REF_KEY);
4351                 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4352         }
4353
4354         btrfs_mark_buffer_dirty(leaf);
4355         btrfs_free_path(path);
4356
4357         ret = remove_from_free_space_tree(trans, extent_key.objectid,
4358                                           num_bytes);
4359         if (ret)
4360                 return ret;
4361
4362         ret = btrfs_update_block_group(trans, extent_key.objectid,
4363                                        fs_info->nodesize, 1);
4364         if (ret) { /* -ENOENT, logic error */
4365                 btrfs_err(fs_info, "update block group failed for %llu %llu",
4366                         extent_key.objectid, extent_key.offset);
4367                 BUG();
4368         }
4369
4370         trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
4371                                           fs_info->nodesize);
4372         return ret;
4373 }
4374
4375 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4376                                      struct btrfs_root *root, u64 owner,
4377                                      u64 offset, u64 ram_bytes,
4378                                      struct btrfs_key *ins)
4379 {
4380         struct btrfs_ref generic_ref = { 0 };
4381         int ret;
4382
4383         BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4384
4385         btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4386                                ins->objectid, ins->offset, 0);
4387         btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset);
4388         btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4389         ret = btrfs_add_delayed_data_ref(trans, &generic_ref,
4390                                          ram_bytes, NULL, NULL);
4391         return ret;
4392 }
4393
4394 /*
4395  * this is used by the tree logging recovery code.  It records that
4396  * an extent has been allocated and makes sure to clear the free
4397  * space cache bits as well
4398  */
4399 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4400                                    u64 root_objectid, u64 owner, u64 offset,
4401                                    struct btrfs_key *ins)
4402 {
4403         struct btrfs_fs_info *fs_info = trans->fs_info;
4404         int ret;
4405         struct btrfs_block_group *block_group;
4406         struct btrfs_space_info *space_info;
4407
4408         /*
4409          * Mixed block groups will exclude before processing the log so we only
4410          * need to do the exclude dance if this fs isn't mixed.
4411          */
4412         if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4413                 ret = __exclude_logged_extent(fs_info, ins->objectid,
4414                                               ins->offset);
4415                 if (ret)
4416                         return ret;
4417         }
4418
4419         block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4420         if (!block_group)
4421                 return -EINVAL;
4422
4423         space_info = block_group->space_info;
4424         spin_lock(&space_info->lock);
4425         spin_lock(&block_group->lock);
4426         space_info->bytes_reserved += ins->offset;
4427         block_group->reserved += ins->offset;
4428         spin_unlock(&block_group->lock);
4429         spin_unlock(&space_info->lock);
4430
4431         ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4432                                          offset, ins, 1);
4433         btrfs_put_block_group(block_group);
4434         return ret;
4435 }
4436
4437 static struct extent_buffer *
4438 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4439                       u64 bytenr, int level, u64 owner)
4440 {
4441         struct btrfs_fs_info *fs_info = root->fs_info;
4442         struct extent_buffer *buf;
4443
4444         buf = btrfs_find_create_tree_block(fs_info, bytenr);
4445         if (IS_ERR(buf))
4446                 return buf;
4447
4448         /*
4449          * Extra safety check in case the extent tree is corrupted and extent
4450          * allocator chooses to use a tree block which is already used and
4451          * locked.
4452          */
4453         if (buf->lock_owner == current->pid) {
4454                 btrfs_err_rl(fs_info,
4455 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4456                         buf->start, btrfs_header_owner(buf), current->pid);
4457                 free_extent_buffer(buf);
4458                 return ERR_PTR(-EUCLEAN);
4459         }
4460
4461         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
4462         btrfs_tree_lock(buf);
4463         btrfs_clean_tree_block(buf);
4464         clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4465
4466         btrfs_set_lock_blocking_write(buf);
4467         set_extent_buffer_uptodate(buf);
4468
4469         memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4470         btrfs_set_header_level(buf, level);
4471         btrfs_set_header_bytenr(buf, buf->start);
4472         btrfs_set_header_generation(buf, trans->transid);
4473         btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4474         btrfs_set_header_owner(buf, owner);
4475         write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4476         write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4477         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4478                 buf->log_index = root->log_transid % 2;
4479                 /*
4480                  * we allow two log transactions at a time, use different
4481                  * EXTENT bit to differentiate dirty pages.
4482                  */
4483                 if (buf->log_index == 0)
4484                         set_extent_dirty(&root->dirty_log_pages, buf->start,
4485                                         buf->start + buf->len - 1, GFP_NOFS);
4486                 else
4487                         set_extent_new(&root->dirty_log_pages, buf->start,
4488                                         buf->start + buf->len - 1);
4489         } else {
4490                 buf->log_index = -1;
4491                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4492                          buf->start + buf->len - 1, GFP_NOFS);
4493         }
4494         trans->dirty = true;
4495         /* this returns a buffer locked for blocking */
4496         return buf;
4497 }
4498
4499 /*
4500  * finds a free extent and does all the dirty work required for allocation
4501  * returns the tree buffer or an ERR_PTR on error.
4502  */
4503 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4504                                              struct btrfs_root *root,
4505                                              u64 parent, u64 root_objectid,
4506                                              const struct btrfs_disk_key *key,
4507                                              int level, u64 hint,
4508                                              u64 empty_size)
4509 {
4510         struct btrfs_fs_info *fs_info = root->fs_info;
4511         struct btrfs_key ins;
4512         struct btrfs_block_rsv *block_rsv;
4513         struct extent_buffer *buf;
4514         struct btrfs_delayed_extent_op *extent_op;
4515         struct btrfs_ref generic_ref = { 0 };
4516         u64 flags = 0;
4517         int ret;
4518         u32 blocksize = fs_info->nodesize;
4519         bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4520
4521 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4522         if (btrfs_is_testing(fs_info)) {
4523                 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4524                                             level, root_objectid);
4525                 if (!IS_ERR(buf))
4526                         root->alloc_bytenr += blocksize;
4527                 return buf;
4528         }
4529 #endif
4530
4531         block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4532         if (IS_ERR(block_rsv))
4533                 return ERR_CAST(block_rsv);
4534
4535         ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4536                                    empty_size, hint, &ins, 0, 0);
4537         if (ret)
4538                 goto out_unuse;
4539
4540         buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4541                                     root_objectid);
4542         if (IS_ERR(buf)) {
4543                 ret = PTR_ERR(buf);
4544                 goto out_free_reserved;
4545         }
4546
4547         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4548                 if (parent == 0)
4549                         parent = ins.objectid;
4550                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4551         } else
4552                 BUG_ON(parent > 0);
4553
4554         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4555                 extent_op = btrfs_alloc_delayed_extent_op();
4556                 if (!extent_op) {
4557                         ret = -ENOMEM;
4558                         goto out_free_buf;
4559                 }
4560                 if (key)
4561                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4562                 else
4563                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4564                 extent_op->flags_to_set = flags;
4565                 extent_op->update_key = skinny_metadata ? false : true;
4566                 extent_op->update_flags = true;
4567                 extent_op->is_data = false;
4568                 extent_op->level = level;
4569
4570                 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4571                                        ins.objectid, ins.offset, parent);
4572                 generic_ref.real_root = root->root_key.objectid;
4573                 btrfs_init_tree_ref(&generic_ref, level, root_objectid);
4574                 btrfs_ref_tree_mod(fs_info, &generic_ref);
4575                 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref,
4576                                                  extent_op, NULL, NULL);
4577                 if (ret)
4578                         goto out_free_delayed;
4579         }
4580         return buf;
4581
4582 out_free_delayed:
4583         btrfs_free_delayed_extent_op(extent_op);
4584 out_free_buf:
4585         free_extent_buffer(buf);
4586 out_free_reserved:
4587         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
4588 out_unuse:
4589         btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
4590         return ERR_PTR(ret);
4591 }
4592
4593 struct walk_control {
4594         u64 refs[BTRFS_MAX_LEVEL];
4595         u64 flags[BTRFS_MAX_LEVEL];
4596         struct btrfs_key update_progress;
4597         struct btrfs_key drop_progress;
4598         int drop_level;
4599         int stage;
4600         int level;
4601         int shared_level;
4602         int update_ref;
4603         int keep_locks;
4604         int reada_slot;
4605         int reada_count;
4606         int restarted;
4607 };
4608
4609 #define DROP_REFERENCE  1
4610 #define UPDATE_BACKREF  2
4611
4612 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4613                                      struct btrfs_root *root,
4614                                      struct walk_control *wc,
4615                                      struct btrfs_path *path)
4616 {
4617         struct btrfs_fs_info *fs_info = root->fs_info;
4618         u64 bytenr;
4619         u64 generation;
4620         u64 refs;
4621         u64 flags;
4622         u32 nritems;
4623         struct btrfs_key key;
4624         struct extent_buffer *eb;
4625         int ret;
4626         int slot;
4627         int nread = 0;
4628
4629         if (path->slots[wc->level] < wc->reada_slot) {
4630                 wc->reada_count = wc->reada_count * 2 / 3;
4631                 wc->reada_count = max(wc->reada_count, 2);
4632         } else {
4633                 wc->reada_count = wc->reada_count * 3 / 2;
4634                 wc->reada_count = min_t(int, wc->reada_count,
4635                                         BTRFS_NODEPTRS_PER_BLOCK(fs_info));
4636         }
4637
4638         eb = path->nodes[wc->level];
4639         nritems = btrfs_header_nritems(eb);
4640
4641         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4642                 if (nread >= wc->reada_count)
4643                         break;
4644
4645                 cond_resched();
4646                 bytenr = btrfs_node_blockptr(eb, slot);
4647                 generation = btrfs_node_ptr_generation(eb, slot);
4648
4649                 if (slot == path->slots[wc->level])
4650                         goto reada;
4651
4652                 if (wc->stage == UPDATE_BACKREF &&
4653                     generation <= root->root_key.offset)
4654                         continue;
4655
4656                 /* We don't lock the tree block, it's OK to be racy here */
4657                 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
4658                                                wc->level - 1, 1, &refs,
4659                                                &flags);
4660                 /* We don't care about errors in readahead. */
4661                 if (ret < 0)
4662                         continue;
4663                 BUG_ON(refs == 0);
4664
4665                 if (wc->stage == DROP_REFERENCE) {
4666                         if (refs == 1)
4667                                 goto reada;
4668
4669                         if (wc->level == 1 &&
4670                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4671                                 continue;
4672                         if (!wc->update_ref ||
4673                             generation <= root->root_key.offset)
4674                                 continue;
4675                         btrfs_node_key_to_cpu(eb, &key, slot);
4676                         ret = btrfs_comp_cpu_keys(&key,
4677                                                   &wc->update_progress);
4678                         if (ret < 0)
4679                                 continue;
4680                 } else {
4681                         if (wc->level == 1 &&
4682                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4683                                 continue;
4684                 }
4685 reada:
4686                 readahead_tree_block(fs_info, bytenr);
4687                 nread++;
4688         }
4689         wc->reada_slot = slot;
4690 }
4691
4692 /*
4693  * helper to process tree block while walking down the tree.
4694  *
4695  * when wc->stage == UPDATE_BACKREF, this function updates
4696  * back refs for pointers in the block.
4697  *
4698  * NOTE: return value 1 means we should stop walking down.
4699  */
4700 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4701                                    struct btrfs_root *root,
4702                                    struct btrfs_path *path,
4703                                    struct walk_control *wc, int lookup_info)
4704 {
4705         struct btrfs_fs_info *fs_info = root->fs_info;
4706         int level = wc->level;
4707         struct extent_buffer *eb = path->nodes[level];
4708         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4709         int ret;
4710
4711         if (wc->stage == UPDATE_BACKREF &&
4712             btrfs_header_owner(eb) != root->root_key.objectid)
4713                 return 1;
4714
4715         /*
4716          * when reference count of tree block is 1, it won't increase
4717          * again. once full backref flag is set, we never clear it.
4718          */
4719         if (lookup_info &&
4720             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4721              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
4722                 BUG_ON(!path->locks[level]);
4723                 ret = btrfs_lookup_extent_info(trans, fs_info,
4724                                                eb->start, level, 1,
4725                                                &wc->refs[level],
4726                                                &wc->flags[level]);
4727                 BUG_ON(ret == -ENOMEM);
4728                 if (ret)
4729                         return ret;
4730                 BUG_ON(wc->refs[level] == 0);
4731         }
4732
4733         if (wc->stage == DROP_REFERENCE) {
4734                 if (wc->refs[level] > 1)
4735                         return 1;
4736
4737                 if (path->locks[level] && !wc->keep_locks) {
4738                         btrfs_tree_unlock_rw(eb, path->locks[level]);
4739                         path->locks[level] = 0;
4740                 }
4741                 return 0;
4742         }
4743
4744         /* wc->stage == UPDATE_BACKREF */
4745         if (!(wc->flags[level] & flag)) {
4746                 BUG_ON(!path->locks[level]);
4747                 ret = btrfs_inc_ref(trans, root, eb, 1);
4748                 BUG_ON(ret); /* -ENOMEM */
4749                 ret = btrfs_dec_ref(trans, root, eb, 0);
4750                 BUG_ON(ret); /* -ENOMEM */
4751                 ret = btrfs_set_disk_extent_flags(trans, eb->start,
4752                                                   eb->len, flag,
4753                                                   btrfs_header_level(eb), 0);
4754                 BUG_ON(ret); /* -ENOMEM */
4755                 wc->flags[level] |= flag;
4756         }
4757
4758         /*
4759          * the block is shared by multiple trees, so it's not good to
4760          * keep the tree lock
4761          */
4762         if (path->locks[level] && level > 0) {
4763                 btrfs_tree_unlock_rw(eb, path->locks[level]);
4764                 path->locks[level] = 0;
4765         }
4766         return 0;
4767 }
4768
4769 /*
4770  * This is used to verify a ref exists for this root to deal with a bug where we
4771  * would have a drop_progress key that hadn't been updated properly.
4772  */
4773 static int check_ref_exists(struct btrfs_trans_handle *trans,
4774                             struct btrfs_root *root, u64 bytenr, u64 parent,
4775                             int level)
4776 {
4777         struct btrfs_path *path;
4778         struct btrfs_extent_inline_ref *iref;
4779         int ret;
4780
4781         path = btrfs_alloc_path();
4782         if (!path)
4783                 return -ENOMEM;
4784
4785         ret = lookup_extent_backref(trans, path, &iref, bytenr,
4786                                     root->fs_info->nodesize, parent,
4787                                     root->root_key.objectid, level, 0);
4788         btrfs_free_path(path);
4789         if (ret == -ENOENT)
4790                 return 0;
4791         if (ret < 0)
4792                 return ret;
4793         return 1;
4794 }
4795
4796 /*
4797  * helper to process tree block pointer.
4798  *
4799  * when wc->stage == DROP_REFERENCE, this function checks
4800  * reference count of the block pointed to. if the block
4801  * is shared and we need update back refs for the subtree
4802  * rooted at the block, this function changes wc->stage to
4803  * UPDATE_BACKREF. if the block is shared and there is no
4804  * need to update back, this function drops the reference
4805  * to the block.
4806  *
4807  * NOTE: return value 1 means we should stop walking down.
4808  */
4809 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4810                                  struct btrfs_root *root,
4811                                  struct btrfs_path *path,
4812                                  struct walk_control *wc, int *lookup_info)
4813 {
4814         struct btrfs_fs_info *fs_info = root->fs_info;
4815         u64 bytenr;
4816         u64 generation;
4817         u64 parent;
4818         struct btrfs_key key;
4819         struct btrfs_key first_key;
4820         struct btrfs_ref ref = { 0 };
4821         struct extent_buffer *next;
4822         int level = wc->level;
4823         int reada = 0;
4824         int ret = 0;
4825         bool need_account = false;
4826
4827         generation = btrfs_node_ptr_generation(path->nodes[level],
4828                                                path->slots[level]);
4829         /*
4830          * if the lower level block was created before the snapshot
4831          * was created, we know there is no need to update back refs
4832          * for the subtree
4833          */
4834         if (wc->stage == UPDATE_BACKREF &&
4835             generation <= root->root_key.offset) {
4836                 *lookup_info = 1;
4837                 return 1;
4838         }
4839
4840         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
4841         btrfs_node_key_to_cpu(path->nodes[level], &first_key,
4842                               path->slots[level]);
4843
4844         next = find_extent_buffer(fs_info, bytenr);
4845         if (!next) {
4846                 next = btrfs_find_create_tree_block(fs_info, bytenr);
4847                 if (IS_ERR(next))
4848                         return PTR_ERR(next);
4849
4850                 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
4851                                                level - 1);
4852                 reada = 1;
4853         }
4854         btrfs_tree_lock(next);
4855         btrfs_set_lock_blocking_write(next);
4856
4857         ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
4858                                        &wc->refs[level - 1],
4859                                        &wc->flags[level - 1]);
4860         if (ret < 0)
4861                 goto out_unlock;
4862
4863         if (unlikely(wc->refs[level - 1] == 0)) {
4864                 btrfs_err(fs_info, "Missing references.");
4865                 ret = -EIO;
4866                 goto out_unlock;
4867         }
4868         *lookup_info = 0;
4869
4870         if (wc->stage == DROP_REFERENCE) {
4871                 if (wc->refs[level - 1] > 1) {
4872                         need_account = true;
4873                         if (level == 1 &&
4874                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4875                                 goto skip;
4876
4877                         if (!wc->update_ref ||
4878                             generation <= root->root_key.offset)
4879                                 goto skip;
4880
4881                         btrfs_node_key_to_cpu(path->nodes[level], &key,
4882                                               path->slots[level]);
4883                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
4884                         if (ret < 0)
4885                                 goto skip;
4886
4887                         wc->stage = UPDATE_BACKREF;
4888                         wc->shared_level = level - 1;
4889                 }
4890         } else {
4891                 if (level == 1 &&
4892                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4893                         goto skip;
4894         }
4895
4896         if (!btrfs_buffer_uptodate(next, generation, 0)) {
4897                 btrfs_tree_unlock(next);
4898                 free_extent_buffer(next);
4899                 next = NULL;
4900                 *lookup_info = 1;
4901         }
4902
4903         if (!next) {
4904                 if (reada && level == 1)
4905                         reada_walk_down(trans, root, wc, path);
4906                 next = read_tree_block(fs_info, bytenr, generation, level - 1,
4907                                        &first_key);
4908                 if (IS_ERR(next)) {
4909                         return PTR_ERR(next);
4910                 } else if (!extent_buffer_uptodate(next)) {
4911                         free_extent_buffer(next);
4912                         return -EIO;
4913                 }
4914                 btrfs_tree_lock(next);
4915                 btrfs_set_lock_blocking_write(next);
4916         }
4917
4918         level--;
4919         ASSERT(level == btrfs_header_level(next));
4920         if (level != btrfs_header_level(next)) {
4921                 btrfs_err(root->fs_info, "mismatched level");
4922                 ret = -EIO;
4923                 goto out_unlock;
4924         }
4925         path->nodes[level] = next;
4926         path->slots[level] = 0;
4927         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
4928         wc->level = level;
4929         if (wc->level == 1)
4930                 wc->reada_slot = 0;
4931         return 0;
4932 skip:
4933         wc->refs[level - 1] = 0;
4934         wc->flags[level - 1] = 0;
4935         if (wc->stage == DROP_REFERENCE) {
4936                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4937                         parent = path->nodes[level]->start;
4938                 } else {
4939                         ASSERT(root->root_key.objectid ==
4940                                btrfs_header_owner(path->nodes[level]));
4941                         if (root->root_key.objectid !=
4942                             btrfs_header_owner(path->nodes[level])) {
4943                                 btrfs_err(root->fs_info,
4944                                                 "mismatched block owner");
4945                                 ret = -EIO;
4946                                 goto out_unlock;
4947                         }
4948                         parent = 0;
4949                 }
4950
4951                 /*
4952                  * If we had a drop_progress we need to verify the refs are set
4953                  * as expected.  If we find our ref then we know that from here
4954                  * on out everything should be correct, and we can clear the
4955                  * ->restarted flag.
4956                  */
4957                 if (wc->restarted) {
4958                         ret = check_ref_exists(trans, root, bytenr, parent,
4959                                                level - 1);
4960                         if (ret < 0)
4961                                 goto out_unlock;
4962                         if (ret == 0)
4963                                 goto no_delete;
4964                         ret = 0;
4965                         wc->restarted = 0;
4966                 }
4967
4968                 /*
4969                  * Reloc tree doesn't contribute to qgroup numbers, and we have
4970                  * already accounted them at merge time (replace_path),
4971                  * thus we could skip expensive subtree trace here.
4972                  */
4973                 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
4974                     need_account) {
4975                         ret = btrfs_qgroup_trace_subtree(trans, next,
4976                                                          generation, level - 1);
4977                         if (ret) {
4978                                 btrfs_err_rl(fs_info,
4979                                              "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
4980                                              ret);
4981                         }
4982                 }
4983
4984                 /*
4985                  * We need to update the next key in our walk control so we can
4986                  * update the drop_progress key accordingly.  We don't care if
4987                  * find_next_key doesn't find a key because that means we're at
4988                  * the end and are going to clean up now.
4989                  */
4990                 wc->drop_level = level;
4991                 find_next_key(path, level, &wc->drop_progress);
4992
4993                 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
4994                                        fs_info->nodesize, parent);
4995                 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid);
4996                 ret = btrfs_free_extent(trans, &ref);
4997                 if (ret)
4998                         goto out_unlock;
4999         }
5000 no_delete:
5001         *lookup_info = 1;
5002         ret = 1;
5003
5004 out_unlock:
5005         btrfs_tree_unlock(next);
5006         free_extent_buffer(next);
5007
5008         return ret;
5009 }
5010
5011 /*
5012  * helper to process tree block while walking up the tree.
5013  *
5014  * when wc->stage == DROP_REFERENCE, this function drops
5015  * reference count on the block.
5016  *
5017  * when wc->stage == UPDATE_BACKREF, this function changes
5018  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5019  * to UPDATE_BACKREF previously while processing the block.
5020  *
5021  * NOTE: return value 1 means we should stop walking up.
5022  */
5023 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5024                                  struct btrfs_root *root,
5025                                  struct btrfs_path *path,
5026                                  struct walk_control *wc)
5027 {
5028         struct btrfs_fs_info *fs_info = root->fs_info;
5029         int ret;
5030         int level = wc->level;
5031         struct extent_buffer *eb = path->nodes[level];
5032         u64 parent = 0;
5033
5034         if (wc->stage == UPDATE_BACKREF) {
5035                 BUG_ON(wc->shared_level < level);
5036                 if (level < wc->shared_level)
5037                         goto out;
5038
5039                 ret = find_next_key(path, level + 1, &wc->update_progress);
5040                 if (ret > 0)
5041                         wc->update_ref = 0;
5042
5043                 wc->stage = DROP_REFERENCE;
5044                 wc->shared_level = -1;
5045                 path->slots[level] = 0;
5046
5047                 /*
5048                  * check reference count again if the block isn't locked.
5049                  * we should start walking down the tree again if reference
5050                  * count is one.
5051                  */
5052                 if (!path->locks[level]) {
5053                         BUG_ON(level == 0);
5054                         btrfs_tree_lock(eb);
5055                         btrfs_set_lock_blocking_write(eb);
5056                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5057
5058                         ret = btrfs_lookup_extent_info(trans, fs_info,
5059                                                        eb->start, level, 1,
5060                                                        &wc->refs[level],
5061                                                        &wc->flags[level]);
5062                         if (ret < 0) {
5063                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
5064                                 path->locks[level] = 0;
5065                                 return ret;
5066                         }
5067                         BUG_ON(wc->refs[level] == 0);
5068                         if (wc->refs[level] == 1) {
5069                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
5070                                 path->locks[level] = 0;
5071                                 return 1;
5072                         }
5073                 }
5074         }
5075
5076         /* wc->stage == DROP_REFERENCE */
5077         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5078
5079         if (wc->refs[level] == 1) {
5080                 if (level == 0) {
5081                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5082                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5083                         else
5084                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5085                         BUG_ON(ret); /* -ENOMEM */
5086                         if (is_fstree(root->root_key.objectid)) {
5087                                 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5088                                 if (ret) {
5089                                         btrfs_err_rl(fs_info,
5090         "error %d accounting leaf items, quota is out of sync, rescan required",
5091                                              ret);
5092                                 }
5093                         }
5094                 }
5095                 /* make block locked assertion in btrfs_clean_tree_block happy */
5096                 if (!path->locks[level] &&
5097                     btrfs_header_generation(eb) == trans->transid) {
5098                         btrfs_tree_lock(eb);
5099                         btrfs_set_lock_blocking_write(eb);
5100                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5101                 }
5102                 btrfs_clean_tree_block(eb);
5103         }
5104
5105         if (eb == root->node) {
5106                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5107                         parent = eb->start;
5108                 else if (root->root_key.objectid != btrfs_header_owner(eb))
5109                         goto owner_mismatch;
5110         } else {
5111                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5112                         parent = path->nodes[level + 1]->start;
5113                 else if (root->root_key.objectid !=
5114                          btrfs_header_owner(path->nodes[level + 1]))
5115                         goto owner_mismatch;
5116         }
5117
5118         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5119 out:
5120         wc->refs[level] = 0;
5121         wc->flags[level] = 0;
5122         return 0;
5123
5124 owner_mismatch:
5125         btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5126                      btrfs_header_owner(eb), root->root_key.objectid);
5127         return -EUCLEAN;
5128 }
5129
5130 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5131                                    struct btrfs_root *root,
5132                                    struct btrfs_path *path,
5133                                    struct walk_control *wc)
5134 {
5135         int level = wc->level;
5136         int lookup_info = 1;
5137         int ret;
5138
5139         while (level >= 0) {
5140                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5141                 if (ret > 0)
5142                         break;
5143
5144                 if (level == 0)
5145                         break;
5146
5147                 if (path->slots[level] >=
5148                     btrfs_header_nritems(path->nodes[level]))
5149                         break;
5150
5151                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5152                 if (ret > 0) {
5153                         path->slots[level]++;
5154                         continue;
5155                 } else if (ret < 0)
5156                         return ret;
5157                 level = wc->level;
5158         }
5159         return 0;
5160 }
5161
5162 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5163                                  struct btrfs_root *root,
5164                                  struct btrfs_path *path,
5165                                  struct walk_control *wc, int max_level)
5166 {
5167         int level = wc->level;
5168         int ret;
5169
5170         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5171         while (level < max_level && path->nodes[level]) {
5172                 wc->level = level;
5173                 if (path->slots[level] + 1 <
5174                     btrfs_header_nritems(path->nodes[level])) {
5175                         path->slots[level]++;
5176                         return 0;
5177                 } else {
5178                         ret = walk_up_proc(trans, root, path, wc);
5179                         if (ret > 0)
5180                                 return 0;
5181                         if (ret < 0)
5182                                 return ret;
5183
5184                         if (path->locks[level]) {
5185                                 btrfs_tree_unlock_rw(path->nodes[level],
5186                                                      path->locks[level]);
5187                                 path->locks[level] = 0;
5188                         }
5189                         free_extent_buffer(path->nodes[level]);
5190                         path->nodes[level] = NULL;
5191                         level++;
5192                 }
5193         }
5194         return 1;
5195 }
5196
5197 /*
5198  * drop a subvolume tree.
5199  *
5200  * this function traverses the tree freeing any blocks that only
5201  * referenced by the tree.
5202  *
5203  * when a shared tree block is found. this function decreases its
5204  * reference count by one. if update_ref is true, this function
5205  * also make sure backrefs for the shared block and all lower level
5206  * blocks are properly updated.
5207  *
5208  * If called with for_reloc == 0, may exit early with -EAGAIN
5209  */
5210 int btrfs_drop_snapshot(struct btrfs_root *root,
5211                          struct btrfs_block_rsv *block_rsv, int update_ref,
5212                          int for_reloc)
5213 {
5214         struct btrfs_fs_info *fs_info = root->fs_info;
5215         struct btrfs_path *path;
5216         struct btrfs_trans_handle *trans;
5217         struct btrfs_root *tree_root = fs_info->tree_root;
5218         struct btrfs_root_item *root_item = &root->root_item;
5219         struct walk_control *wc;
5220         struct btrfs_key key;
5221         int err = 0;
5222         int ret;
5223         int level;
5224         bool root_dropped = false;
5225
5226         btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5227
5228         path = btrfs_alloc_path();
5229         if (!path) {
5230                 err = -ENOMEM;
5231                 goto out;
5232         }
5233
5234         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5235         if (!wc) {
5236                 btrfs_free_path(path);
5237                 err = -ENOMEM;
5238                 goto out;
5239         }
5240
5241         trans = btrfs_start_transaction(tree_root, 0);
5242         if (IS_ERR(trans)) {
5243                 err = PTR_ERR(trans);
5244                 goto out_free;
5245         }
5246
5247         err = btrfs_run_delayed_items(trans);
5248         if (err)
5249                 goto out_end_trans;
5250
5251         if (block_rsv)
5252                 trans->block_rsv = block_rsv;
5253
5254         /*
5255          * This will help us catch people modifying the fs tree while we're
5256          * dropping it.  It is unsafe to mess with the fs tree while it's being
5257          * dropped as we unlock the root node and parent nodes as we walk down
5258          * the tree, assuming nothing will change.  If something does change
5259          * then we'll have stale information and drop references to blocks we've
5260          * already dropped.
5261          */
5262         set_bit(BTRFS_ROOT_DELETING, &root->state);
5263         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5264                 level = btrfs_header_level(root->node);
5265                 path->nodes[level] = btrfs_lock_root_node(root);
5266                 btrfs_set_lock_blocking_write(path->nodes[level]);
5267                 path->slots[level] = 0;
5268                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5269                 memset(&wc->update_progress, 0,
5270                        sizeof(wc->update_progress));
5271         } else {
5272                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5273                 memcpy(&wc->update_progress, &key,
5274                        sizeof(wc->update_progress));
5275
5276                 level = root_item->drop_level;
5277                 BUG_ON(level == 0);
5278                 path->lowest_level = level;
5279                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5280                 path->lowest_level = 0;
5281                 if (ret < 0) {
5282                         err = ret;
5283                         goto out_end_trans;
5284                 }
5285                 WARN_ON(ret > 0);
5286
5287                 /*
5288                  * unlock our path, this is safe because only this
5289                  * function is allowed to delete this snapshot
5290                  */
5291                 btrfs_unlock_up_safe(path, 0);
5292
5293                 level = btrfs_header_level(root->node);
5294                 while (1) {
5295                         btrfs_tree_lock(path->nodes[level]);
5296                         btrfs_set_lock_blocking_write(path->nodes[level]);
5297                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5298
5299                         ret = btrfs_lookup_extent_info(trans, fs_info,
5300                                                 path->nodes[level]->start,
5301                                                 level, 1, &wc->refs[level],
5302                                                 &wc->flags[level]);
5303                         if (ret < 0) {
5304                                 err = ret;
5305                                 goto out_end_trans;
5306                         }
5307                         BUG_ON(wc->refs[level] == 0);
5308
5309                         if (level == root_item->drop_level)
5310                                 break;
5311
5312                         btrfs_tree_unlock(path->nodes[level]);
5313                         path->locks[level] = 0;
5314                         WARN_ON(wc->refs[level] != 1);
5315                         level--;
5316                 }
5317         }
5318
5319         wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5320         wc->level = level;
5321         wc->shared_level = -1;
5322         wc->stage = DROP_REFERENCE;
5323         wc->update_ref = update_ref;
5324         wc->keep_locks = 0;
5325         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5326
5327         while (1) {
5328
5329                 ret = walk_down_tree(trans, root, path, wc);
5330                 if (ret < 0) {
5331                         err = ret;
5332                         break;
5333                 }
5334
5335                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5336                 if (ret < 0) {
5337                         err = ret;
5338                         break;
5339                 }
5340
5341                 if (ret > 0) {
5342                         BUG_ON(wc->stage != DROP_REFERENCE);
5343                         break;
5344                 }
5345
5346                 if (wc->stage == DROP_REFERENCE) {
5347                         wc->drop_level = wc->level;
5348                         btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5349                                               &wc->drop_progress,
5350                                               path->slots[wc->drop_level]);
5351                 }
5352                 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5353                                       &wc->drop_progress);
5354                 root_item->drop_level = wc->drop_level;
5355
5356                 BUG_ON(wc->level == 0);
5357                 if (btrfs_should_end_transaction(trans) ||
5358                     (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5359                         ret = btrfs_update_root(trans, tree_root,
5360                                                 &root->root_key,
5361                                                 root_item);
5362                         if (ret) {
5363                                 btrfs_abort_transaction(trans, ret);
5364                                 err = ret;
5365                                 goto out_end_trans;
5366                         }
5367
5368                         btrfs_end_transaction_throttle(trans);
5369                         if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5370                                 btrfs_debug(fs_info,
5371                                             "drop snapshot early exit");
5372                                 err = -EAGAIN;
5373                                 goto out_free;
5374                         }
5375
5376                         trans = btrfs_start_transaction(tree_root, 0);
5377                         if (IS_ERR(trans)) {
5378                                 err = PTR_ERR(trans);
5379                                 goto out_free;
5380                         }
5381                         if (block_rsv)
5382                                 trans->block_rsv = block_rsv;
5383                 }
5384         }
5385         btrfs_release_path(path);
5386         if (err)
5387                 goto out_end_trans;
5388
5389         ret = btrfs_del_root(trans, &root->root_key);
5390         if (ret) {
5391                 btrfs_abort_transaction(trans, ret);
5392                 err = ret;
5393                 goto out_end_trans;
5394         }
5395
5396         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5397                 ret = btrfs_find_root(tree_root, &root->root_key, path,
5398                                       NULL, NULL);
5399                 if (ret < 0) {
5400                         btrfs_abort_transaction(trans, ret);
5401                         err = ret;
5402                         goto out_end_trans;
5403                 } else if (ret > 0) {
5404                         /* if we fail to delete the orphan item this time
5405                          * around, it'll get picked up the next time.
5406                          *
5407                          * The most common failure here is just -ENOENT.
5408                          */
5409                         btrfs_del_orphan_item(trans, tree_root,
5410                                               root->root_key.objectid);
5411                 }
5412         }
5413
5414         if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
5415                 btrfs_add_dropped_root(trans, root);
5416         } else {
5417                 free_extent_buffer(root->node);
5418                 free_extent_buffer(root->commit_root);
5419                 btrfs_put_fs_root(root);
5420         }
5421         root_dropped = true;
5422 out_end_trans:
5423         btrfs_end_transaction_throttle(trans);
5424 out_free:
5425         kfree(wc);
5426         btrfs_free_path(path);
5427 out:
5428         /*
5429          * So if we need to stop dropping the snapshot for whatever reason we
5430          * need to make sure to add it back to the dead root list so that we
5431          * keep trying to do the work later.  This also cleans up roots if we
5432          * don't have it in the radix (like when we recover after a power fail
5433          * or unmount) so we don't leak memory.
5434          */
5435         if (!for_reloc && !root_dropped)
5436                 btrfs_add_dead_root(root);
5437         if (err && err != -EAGAIN)
5438                 btrfs_handle_fs_error(fs_info, err, NULL);
5439         return err;
5440 }
5441
5442 /*
5443  * drop subtree rooted at tree block 'node'.
5444  *
5445  * NOTE: this function will unlock and release tree block 'node'
5446  * only used by relocation code
5447  */
5448 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5449                         struct btrfs_root *root,
5450                         struct extent_buffer *node,
5451                         struct extent_buffer *parent)
5452 {
5453         struct btrfs_fs_info *fs_info = root->fs_info;
5454         struct btrfs_path *path;
5455         struct walk_control *wc;
5456         int level;
5457         int parent_level;
5458         int ret = 0;
5459         int wret;
5460
5461         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5462
5463         path = btrfs_alloc_path();
5464         if (!path)
5465                 return -ENOMEM;
5466
5467         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5468         if (!wc) {
5469                 btrfs_free_path(path);
5470                 return -ENOMEM;
5471         }
5472
5473         btrfs_assert_tree_locked(parent);
5474         parent_level = btrfs_header_level(parent);
5475         atomic_inc(&parent->refs);
5476         path->nodes[parent_level] = parent;
5477         path->slots[parent_level] = btrfs_header_nritems(parent);
5478
5479         btrfs_assert_tree_locked(node);
5480         level = btrfs_header_level(node);
5481         path->nodes[level] = node;
5482         path->slots[level] = 0;
5483         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
5484
5485         wc->refs[parent_level] = 1;
5486         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5487         wc->level = level;
5488         wc->shared_level = -1;
5489         wc->stage = DROP_REFERENCE;
5490         wc->update_ref = 0;
5491         wc->keep_locks = 1;
5492         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5493
5494         while (1) {
5495                 wret = walk_down_tree(trans, root, path, wc);
5496                 if (wret < 0) {
5497                         ret = wret;
5498                         break;
5499                 }
5500
5501                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5502                 if (wret < 0)
5503                         ret = wret;
5504                 if (wret != 0)
5505                         break;
5506         }
5507
5508         kfree(wc);
5509         btrfs_free_path(path);
5510         return ret;
5511 }
5512
5513 /*
5514  * helper to account the unused space of all the readonly block group in the
5515  * space_info. takes mirrors into account.
5516  */
5517 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
5518 {
5519         struct btrfs_block_group *block_group;
5520         u64 free_bytes = 0;
5521         int factor;
5522
5523         /* It's df, we don't care if it's racy */
5524         if (list_empty(&sinfo->ro_bgs))
5525                 return 0;
5526
5527         spin_lock(&sinfo->lock);
5528         list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
5529                 spin_lock(&block_group->lock);
5530
5531                 if (!block_group->ro) {
5532                         spin_unlock(&block_group->lock);
5533                         continue;
5534                 }
5535
5536                 factor = btrfs_bg_type_to_factor(block_group->flags);
5537                 free_bytes += (block_group->length -
5538                                block_group->used) * factor;
5539
5540                 spin_unlock(&block_group->lock);
5541         }
5542         spin_unlock(&sinfo->lock);
5543
5544         return free_bytes;
5545 }
5546
5547 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5548                                    u64 start, u64 end)
5549 {
5550         return unpin_extent_range(fs_info, start, end, false);
5551 }
5552
5553 /*
5554  * It used to be that old block groups would be left around forever.
5555  * Iterating over them would be enough to trim unused space.  Since we
5556  * now automatically remove them, we also need to iterate over unallocated
5557  * space.
5558  *
5559  * We don't want a transaction for this since the discard may take a
5560  * substantial amount of time.  We don't require that a transaction be
5561  * running, but we do need to take a running transaction into account
5562  * to ensure that we're not discarding chunks that were released or
5563  * allocated in the current transaction.
5564  *
5565  * Holding the chunks lock will prevent other threads from allocating
5566  * or releasing chunks, but it won't prevent a running transaction
5567  * from committing and releasing the memory that the pending chunks
5568  * list head uses.  For that, we need to take a reference to the
5569  * transaction and hold the commit root sem.  We only need to hold
5570  * it while performing the free space search since we have already
5571  * held back allocations.
5572  */
5573 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
5574 {
5575         u64 start = SZ_1M, len = 0, end = 0;
5576         int ret;
5577
5578         *trimmed = 0;
5579
5580         /* Discard not supported = nothing to do. */
5581         if (!blk_queue_discard(bdev_get_queue(device->bdev)))
5582                 return 0;
5583
5584         /* Not writable = nothing to do. */
5585         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5586                 return 0;
5587
5588         /* No free space = nothing to do. */
5589         if (device->total_bytes <= device->bytes_used)
5590                 return 0;
5591
5592         ret = 0;
5593
5594         while (1) {
5595                 struct btrfs_fs_info *fs_info = device->fs_info;
5596                 u64 bytes;
5597
5598                 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
5599                 if (ret)
5600                         break;
5601
5602                 find_first_clear_extent_bit(&device->alloc_state, start,
5603                                             &start, &end,
5604                                             CHUNK_TRIMMED | CHUNK_ALLOCATED);
5605
5606                 /* Ensure we skip the reserved area in the first 1M */
5607                 start = max_t(u64, start, SZ_1M);
5608
5609                 /*
5610                  * If find_first_clear_extent_bit find a range that spans the
5611                  * end of the device it will set end to -1, in this case it's up
5612                  * to the caller to trim the value to the size of the device.
5613                  */
5614                 end = min(end, device->total_bytes - 1);
5615
5616                 len = end - start + 1;
5617
5618                 /* We didn't find any extents */
5619                 if (!len) {
5620                         mutex_unlock(&fs_info->chunk_mutex);
5621                         ret = 0;
5622                         break;
5623                 }
5624
5625                 ret = btrfs_issue_discard(device->bdev, start, len,
5626                                           &bytes);
5627                 if (!ret)
5628                         set_extent_bits(&device->alloc_state, start,
5629                                         start + bytes - 1,
5630                                         CHUNK_TRIMMED);
5631                 mutex_unlock(&fs_info->chunk_mutex);
5632
5633                 if (ret)
5634                         break;
5635
5636                 start += len;
5637                 *trimmed += bytes;
5638
5639                 if (fatal_signal_pending(current)) {
5640                         ret = -ERESTARTSYS;
5641                         break;
5642                 }
5643
5644                 cond_resched();
5645         }
5646
5647         return ret;
5648 }
5649
5650 /*
5651  * Trim the whole filesystem by:
5652  * 1) trimming the free space in each block group
5653  * 2) trimming the unallocated space on each device
5654  *
5655  * This will also continue trimming even if a block group or device encounters
5656  * an error.  The return value will be the last error, or 0 if nothing bad
5657  * happens.
5658  */
5659 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
5660 {
5661         struct btrfs_block_group *cache = NULL;
5662         struct btrfs_device *device;
5663         struct list_head *devices;
5664         u64 group_trimmed;
5665         u64 range_end = U64_MAX;
5666         u64 start;
5667         u64 end;
5668         u64 trimmed = 0;
5669         u64 bg_failed = 0;
5670         u64 dev_failed = 0;
5671         int bg_ret = 0;
5672         int dev_ret = 0;
5673         int ret = 0;
5674
5675         /*
5676          * Check range overflow if range->len is set.
5677          * The default range->len is U64_MAX.
5678          */
5679         if (range->len != U64_MAX &&
5680             check_add_overflow(range->start, range->len, &range_end))
5681                 return -EINVAL;
5682
5683         cache = btrfs_lookup_first_block_group(fs_info, range->start);
5684         for (; cache; cache = btrfs_next_block_group(cache)) {
5685                 if (cache->start >= range_end) {
5686                         btrfs_put_block_group(cache);
5687                         break;
5688                 }
5689
5690                 start = max(range->start, cache->start);
5691                 end = min(range_end, cache->start + cache->length);
5692
5693                 if (end - start >= range->minlen) {
5694                         if (!btrfs_block_group_done(cache)) {
5695                                 ret = btrfs_cache_block_group(cache, 0);
5696                                 if (ret) {
5697                                         bg_failed++;
5698                                         bg_ret = ret;
5699                                         continue;
5700                                 }
5701                                 ret = btrfs_wait_block_group_cache_done(cache);
5702                                 if (ret) {
5703                                         bg_failed++;
5704                                         bg_ret = ret;
5705                                         continue;
5706                                 }
5707                         }
5708                         ret = btrfs_trim_block_group(cache,
5709                                                      &group_trimmed,
5710                                                      start,
5711                                                      end,
5712                                                      range->minlen);
5713
5714                         trimmed += group_trimmed;
5715                         if (ret) {
5716                                 bg_failed++;
5717                                 bg_ret = ret;
5718                                 continue;
5719                         }
5720                 }
5721         }
5722
5723         if (bg_failed)
5724                 btrfs_warn(fs_info,
5725                         "failed to trim %llu block group(s), last error %d",
5726                         bg_failed, bg_ret);
5727         mutex_lock(&fs_info->fs_devices->device_list_mutex);
5728         devices = &fs_info->fs_devices->devices;
5729         list_for_each_entry(device, devices, dev_list) {
5730                 ret = btrfs_trim_free_extents(device, &group_trimmed);
5731                 if (ret) {
5732                         dev_failed++;
5733                         dev_ret = ret;
5734                         break;
5735                 }
5736
5737                 trimmed += group_trimmed;
5738         }
5739         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5740
5741         if (dev_failed)
5742                 btrfs_warn(fs_info,
5743                         "failed to trim %llu device(s), last error %d",
5744                         dev_failed, dev_ret);
5745         range->len = trimmed;
5746         if (bg_ret)
5747                 return bg_ret;
5748         return dev_ret;
5749 }
5750
5751 /*
5752  * btrfs_{start,end}_write_no_snapshotting() are similar to
5753  * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
5754  * data into the page cache through nocow before the subvolume is snapshoted,
5755  * but flush the data into disk after the snapshot creation, or to prevent
5756  * operations while snapshotting is ongoing and that cause the snapshot to be
5757  * inconsistent (writes followed by expanding truncates for example).
5758  */
5759 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
5760 {
5761         percpu_counter_dec(&root->subv_writers->counter);
5762         cond_wake_up(&root->subv_writers->wait);
5763 }
5764
5765 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
5766 {
5767         if (atomic_read(&root->will_be_snapshotted))
5768                 return 0;
5769
5770         percpu_counter_inc(&root->subv_writers->counter);
5771         /*
5772          * Make sure counter is updated before we check for snapshot creation.
5773          */
5774         smp_mb();
5775         if (atomic_read(&root->will_be_snapshotted)) {
5776                 btrfs_end_write_no_snapshotting(root);
5777                 return 0;
5778         }
5779         return 1;
5780 }
5781
5782 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
5783 {
5784         while (true) {
5785                 int ret;
5786
5787                 ret = btrfs_start_write_no_snapshotting(root);
5788                 if (ret)
5789                         break;
5790                 wait_var_event(&root->will_be_snapshotted,
5791                                !atomic_read(&root->will_be_snapshotted));
5792         }
5793 }