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 "ctree.h"
  20 #include "extent-tree.h"
  21 #include "transaction.h"
  22 #include "disk-io.h"
  23 #include "print-tree.h"
  24 #include "volumes.h"
  25 #include "raid56.h"
  26 #include "locking.h"
  27 #include "free-space-cache.h"
  28 #include "free-space-tree.h"
  29 #include "qgroup.h"
  30 #include "ref-verify.h"
  31 #include "space-info.h"
  32 #include "block-rsv.h"
  33 #include "discard.h"
  34 #include "zoned.h"
  35 #include "dev-replace.h"
  36 #include "fs.h"
  37 #include "accessors.h"
  38 #include "root-tree.h"
  39 #include "file-item.h"
  40 #include "orphan.h"
  41 #include "tree-checker.h"
  42 #include "raid-stripe-tree.h"
  43
  44 #undef SCRAMBLE_DELAYED_REFS
  45
  46
  47 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  48                                struct btrfs_delayed_ref_head *href,
  49                                struct btrfs_delayed_ref_node *node,
  50                                struct btrfs_delayed_extent_op *extra_op);
  51 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  52                                     struct extent_buffer *leaf,
  53                                     struct btrfs_extent_item *ei);
  54 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  55                                       u64 parent, u64 root_objectid,
  56                                       u64 flags, u64 owner, u64 offset,
  57                                       struct btrfs_key *ins, int ref_mod, u64 oref_root);
  58 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  59                                      struct btrfs_delayed_ref_node *node,
  60                                      struct btrfs_delayed_extent_op *extent_op);
  61 static int find_next_key(struct btrfs_path *path, int level,
  62                          struct btrfs_key *key);
  63
  64 static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
  65 {
  66         return (cache->flags & bits) == bits;
  67 }
  68
  69 /* simple helper to search for an existing data extent at a given offset */
  70 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
  71 {
  72         struct btrfs_root *root = btrfs_extent_root(fs_info, start);
  73         int ret;
  74         struct btrfs_key key;
  75         struct btrfs_path *path;
  76
  77         path = btrfs_alloc_path();
  78         if (!path)
  79                 return -ENOMEM;
  80
  81         key.objectid = start;
  82         key.offset = len;
  83         key.type = BTRFS_EXTENT_ITEM_KEY;
  84         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  85         btrfs_free_path(path);
  86         return ret;
  87 }
  88
  89 /*
  90  * helper function to lookup reference count and flags of a tree block.
  91  *
  92  * the head node for delayed ref is used to store the sum of all the
  93  * reference count modifications queued up in the rbtree. the head
  94  * node may also store the extent flags to set. This way you can check
  95  * to see what the reference count and extent flags would be if all of
  96  * the delayed refs are not processed.
  97  */
  98 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
  99                              struct btrfs_fs_info *fs_info, u64 bytenr,
 100                              u64 offset, int metadata, u64 *refs, u64 *flags,
 101                              u64 *owning_root)
 102 {
 103         struct btrfs_root *extent_root;
 104         struct btrfs_delayed_ref_head *head;
 105         struct btrfs_delayed_ref_root *delayed_refs;
 106         struct btrfs_path *path;
 107         struct btrfs_extent_item *ei;
 108         struct extent_buffer *leaf;
 109         struct btrfs_key key;
 110         u32 item_size;
 111         u64 num_refs;
 112         u64 extent_flags;
 113         u64 owner = 0;
 114         int ret;
 115
 116         /*
 117          * If we don't have skinny metadata, don't bother doing anything
 118          * different
 119          */
 120         if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
 121                 offset = fs_info->nodesize;
 122                 metadata = 0;
 123         }
 124
 125         path = btrfs_alloc_path();
 126         if (!path)
 127                 return -ENOMEM;
 128
 129         if (!trans) {
 130                 path->skip_locking = 1;
 131                 path->search_commit_root = 1;
 132         }
 133
 134 search_again:
 135         key.objectid = bytenr;
 136         key.offset = offset;
 137         if (metadata)
 138                 key.type = BTRFS_METADATA_ITEM_KEY;
 139         else
 140                 key.type = BTRFS_EXTENT_ITEM_KEY;
 141
 142         extent_root = btrfs_extent_root(fs_info, bytenr);
 143         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 144         if (ret < 0)
 145                 goto out_free;
 146
 147         if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
 148                 if (path->slots[0]) {
 149                         path->slots[0]--;
 150                         btrfs_item_key_to_cpu(path->nodes[0], &key,
 151                                               path->slots[0]);
 152                         if (key.objectid == bytenr &&
 153                             key.type == BTRFS_EXTENT_ITEM_KEY &&
 154                             key.offset == fs_info->nodesize)
 155                                 ret = 0;
 156                 }
 157         }
 158
 159         if (ret == 0) {
 160                 leaf = path->nodes[0];
 161                 item_size = btrfs_item_size(leaf, path->slots[0]);
 162                 if (item_size >= sizeof(*ei)) {
 163                         ei = btrfs_item_ptr(leaf, path->slots[0],
 164                                             struct btrfs_extent_item);
 165                         num_refs = btrfs_extent_refs(leaf, ei);
 166                         extent_flags = btrfs_extent_flags(leaf, ei);
 167                         owner = btrfs_get_extent_owner_root(fs_info, leaf,
 168                                                             path->slots[0]);
 169                 } else {
 170                         ret = -EUCLEAN;
 171                         btrfs_err(fs_info,
 172                         "unexpected extent item size, has %u expect >= %zu",
 173                                   item_size, sizeof(*ei));
 174                         if (trans)
 175                                 btrfs_abort_transaction(trans, ret);
 176                         else
 177                                 btrfs_handle_fs_error(fs_info, ret, NULL);
 178
 179                         goto out_free;
 180                 }
 181
 182                 BUG_ON(num_refs == 0);
 183         } else {
 184                 num_refs = 0;
 185                 extent_flags = 0;
 186                 ret = 0;
 187         }
 188
 189         if (!trans)
 190                 goto out;
 191
 192         delayed_refs = &trans->transaction->delayed_refs;
 193         spin_lock(&delayed_refs->lock);
 194         head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
 195         if (head) {
 196                 if (!mutex_trylock(&head->mutex)) {
 197                         refcount_inc(&head->refs);
 198                         spin_unlock(&delayed_refs->lock);
 199
 200                         btrfs_release_path(path);
 201
 202                         /*
 203                          * Mutex was contended, block until it's released and try
 204                          * again
 205                          */
 206                         mutex_lock(&head->mutex);
 207                         mutex_unlock(&head->mutex);
 208                         btrfs_put_delayed_ref_head(head);
 209                         goto search_again;
 210                 }
 211                 spin_lock(&head->lock);
 212                 if (head->extent_op && head->extent_op->update_flags)
 213                         extent_flags |= head->extent_op->flags_to_set;
 214                 else
 215                         BUG_ON(num_refs == 0);
 216
 217                 num_refs += head->ref_mod;
 218                 spin_unlock(&head->lock);
 219                 mutex_unlock(&head->mutex);
 220         }
 221         spin_unlock(&delayed_refs->lock);
 222 out:
 223         WARN_ON(num_refs == 0);
 224         if (refs)
 225                 *refs = num_refs;
 226         if (flags)
 227                 *flags = extent_flags;
 228         if (owning_root)
 229                 *owning_root = owner;
 230 out_free:
 231         btrfs_free_path(path);
 232         return ret;
 233 }
 234
 235 /*
 236  * Back reference rules.  Back refs have three main goals:
 237  *
 238  * 1) differentiate between all holders of references to an extent so that
 239  *    when a reference is dropped we can make sure it was a valid reference
 240  *    before freeing the extent.
 241  *
 242  * 2) Provide enough information to quickly find the holders of an extent
 243  *    if we notice a given block is corrupted or bad.
 244  *
 245  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 246  *    maintenance.  This is actually the same as #2, but with a slightly
 247  *    different use case.
 248  *
 249  * There are two kinds of back refs. The implicit back refs is optimized
 250  * for pointers in non-shared tree blocks. For a given pointer in a block,
 251  * back refs of this kind provide information about the block's owner tree
 252  * and the pointer's key. These information allow us to find the block by
 253  * b-tree searching. The full back refs is for pointers in tree blocks not
 254  * referenced by their owner trees. The location of tree block is recorded
 255  * in the back refs. Actually the full back refs is generic, and can be
 256  * used in all cases the implicit back refs is used. The major shortcoming
 257  * of the full back refs is its overhead. Every time a tree block gets
 258  * COWed, we have to update back refs entry for all pointers in it.
 259  *
 260  * For a newly allocated tree block, we use implicit back refs for
 261  * pointers in it. This means most tree related operations only involve
 262  * implicit back refs. For a tree block created in old transaction, the
 263  * only way to drop a reference to it is COW it. So we can detect the
 264  * event that tree block loses its owner tree's reference and do the
 265  * back refs conversion.
 266  *
 267  * When a tree block is COWed through a tree, there are four cases:
 268  *
 269  * The reference count of the block is one and the tree is the block's
 270  * owner tree. Nothing to do in this case.
 271  *
 272  * The reference count of the block is one and the tree is not the
 273  * block's owner tree. In this case, full back refs is used for pointers
 274  * in the block. Remove these full back refs, add implicit back refs for
 275  * every pointers in the new block.
 276  *
 277  * The reference count of the block is greater than one and the tree is
 278  * the block's owner tree. In this case, implicit back refs is used for
 279  * pointers in the block. Add full back refs for every pointers in the
 280  * block, increase lower level extents' reference counts. The original
 281  * implicit back refs are entailed to the new block.
 282  *
 283  * The reference count of the block is greater than one and the tree is
 284  * not the block's owner tree. Add implicit back refs for every pointer in
 285  * the new block, increase lower level extents' reference count.
 286  *
 287  * Back Reference Key composing:
 288  *
 289  * The key objectid corresponds to the first byte in the extent,
 290  * The key type is used to differentiate between types of back refs.
 291  * There are different meanings of the key offset for different types
 292  * of back refs.
 293  *
 294  * File extents can be referenced by:
 295  *
 296  * - multiple snapshots, subvolumes, or different generations in one subvol
 297  * - different files inside a single subvolume
 298  * - different offsets inside a file (bookend extents in file.c)
 299  *
 300  * The extent ref structure for the implicit back refs has fields for:
 301  *
 302  * - Objectid of the subvolume root
 303  * - objectid of the file holding the reference
 304  * - original offset in the file
 305  * - how many bookend extents
 306  *
 307  * The key offset for the implicit back refs is hash of the first
 308  * three fields.
 309  *
 310  * The extent ref structure for the full back refs has field for:
 311  *
 312  * - number of pointers in the tree leaf
 313  *
 314  * The key offset for the implicit back refs is the first byte of
 315  * the tree leaf
 316  *
 317  * When a file extent is allocated, The implicit back refs is used.
 318  * the fields are filled in:
 319  *
 320  *     (root_key.objectid, inode objectid, offset in file, 1)
 321  *
 322  * When a file extent is removed file truncation, we find the
 323  * corresponding implicit back refs and check the following fields:
 324  *
 325  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 326  *
 327  * Btree extents can be referenced by:
 328  *
 329  * - Different subvolumes
 330  *
 331  * Both the implicit back refs and the full back refs for tree blocks
 332  * only consist of key. The key offset for the implicit back refs is
 333  * objectid of block's owner tree. The key offset for the full back refs
 334  * is the first byte of parent block.
 335  *
 336  * When implicit back refs is used, information about the lowest key and
 337  * level of the tree block are required. These information are stored in
 338  * tree block info structure.
 339  */
 340
 341 /*
 342  * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
 343  * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
 344  * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
 345  */
 346 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
 347                                      struct btrfs_extent_inline_ref *iref,
 348                                      enum btrfs_inline_ref_type is_data)
 349 {
 350         struct btrfs_fs_info *fs_info = eb->fs_info;
 351         int type = btrfs_extent_inline_ref_type(eb, iref);
 352         u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
 353
 354         if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
 355                 ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
 356                 return type;
 357         }
 358
 359         if (type == BTRFS_TREE_BLOCK_REF_KEY ||
 360             type == BTRFS_SHARED_BLOCK_REF_KEY ||
 361             type == BTRFS_SHARED_DATA_REF_KEY ||
 362             type == BTRFS_EXTENT_DATA_REF_KEY) {
 363                 if (is_data == BTRFS_REF_TYPE_BLOCK) {
 364                         if (type == BTRFS_TREE_BLOCK_REF_KEY)
 365                                 return type;
 366                         if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
 367                                 ASSERT(fs_info);
 368                                 /*
 369                                  * Every shared one has parent tree block,
 370                                  * which must be aligned to sector size.
 371                                  */
 372                                 if (offset && IS_ALIGNED(offset, fs_info->sectorsize))
 373                                         return type;
 374                         }
 375                 } else if (is_data == BTRFS_REF_TYPE_DATA) {
 376                         if (type == BTRFS_EXTENT_DATA_REF_KEY)
 377                                 return type;
 378                         if (type == BTRFS_SHARED_DATA_REF_KEY) {
 379                                 ASSERT(fs_info);
 380                                 /*
 381                                  * Every shared one has parent tree block,
 382                                  * which must be aligned to sector size.
 383                                  */
 384                                 if (offset &&
 385                                     IS_ALIGNED(offset, fs_info->sectorsize))
 386                                         return type;
 387                         }
 388                 } else {
 389                         ASSERT(is_data == BTRFS_REF_TYPE_ANY);
 390                         return type;
 391                 }
 392         }
 393
 394         WARN_ON(1);
 395         btrfs_print_leaf(eb);
 396         btrfs_err(fs_info,
 397                   "eb %llu iref 0x%lx invalid extent inline ref type %d",
 398                   eb->start, (unsigned long)iref, type);
 399
 400         return BTRFS_REF_TYPE_INVALID;
 401 }
 402
 403 u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 404 {
 405         u32 high_crc = ~(u32)0;
 406         u32 low_crc = ~(u32)0;
 407         __le64 lenum;
 408
 409         lenum = cpu_to_le64(root_objectid);
 410         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 411         lenum = cpu_to_le64(owner);
 412         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 413         lenum = cpu_to_le64(offset);
 414         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 415
 416         return ((u64)high_crc << 31) ^ (u64)low_crc;
 417 }
 418
 419 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 420                                      struct btrfs_extent_data_ref *ref)
 421 {
 422         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 423                                     btrfs_extent_data_ref_objectid(leaf, ref),
 424                                     btrfs_extent_data_ref_offset(leaf, ref));
 425 }
 426
 427 static int match_extent_data_ref(struct extent_buffer *leaf,
 428                                  struct btrfs_extent_data_ref *ref,
 429                                  u64 root_objectid, u64 owner, u64 offset)
 430 {
 431         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 432             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 433             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 434                 return 0;
 435         return 1;
 436 }
 437
 438 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 439                                            struct btrfs_path *path,
 440                                            u64 bytenr, u64 parent,
 441                                            u64 root_objectid,
 442                                            u64 owner, u64 offset)
 443 {
 444         struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
 445         struct btrfs_key key;
 446         struct btrfs_extent_data_ref *ref;
 447         struct extent_buffer *leaf;
 448         u32 nritems;
 449         int recow;
 450         int ret;
 451
 452         key.objectid = bytenr;
 453         if (parent) {
 454                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 455                 key.offset = parent;
 456         } else {
 457                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 458                 key.offset = hash_extent_data_ref(root_objectid,
 459                                                   owner, offset);
 460         }
 461 again:
 462         recow = 0;
 463         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 464         if (ret < 0)
 465                 return ret;
 466
 467         if (parent) {
 468                 if (ret)
 469                         return -ENOENT;
 470                 return 0;
 471         }
 472
 473         ret = -ENOENT;
 474         leaf = path->nodes[0];
 475         nritems = btrfs_header_nritems(leaf);
 476         while (1) {
 477                 if (path->slots[0] >= nritems) {
 478                         ret = btrfs_next_leaf(root, path);
 479                         if (ret) {
 480                                 if (ret > 0)
 481                                         return -ENOENT;
 482                                 return ret;
 483                         }
 484
 485                         leaf = path->nodes[0];
 486                         nritems = btrfs_header_nritems(leaf);
 487                         recow = 1;
 488                 }
 489
 490                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 491                 if (key.objectid != bytenr ||
 492                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 493                         goto fail;
 494
 495                 ref = btrfs_item_ptr(leaf, path->slots[0],
 496                                      struct btrfs_extent_data_ref);
 497
 498                 if (match_extent_data_ref(leaf, ref, root_objectid,
 499                                           owner, offset)) {
 500                         if (recow) {
 501                                 btrfs_release_path(path);
 502                                 goto again;
 503                         }
 504                         ret = 0;
 505                         break;
 506                 }
 507                 path->slots[0]++;
 508         }
 509 fail:
 510         return ret;
 511 }
 512
 513 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 514                                            struct btrfs_path *path,
 515                                            struct btrfs_delayed_ref_node *node,
 516                                            u64 bytenr)
 517 {
 518         struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
 519         struct btrfs_key key;
 520         struct extent_buffer *leaf;
 521         u64 owner = btrfs_delayed_ref_owner(node);
 522         u64 offset = btrfs_delayed_ref_offset(node);
 523         u32 size;
 524         u32 num_refs;
 525         int ret;
 526
 527         key.objectid = bytenr;
 528         if (node->parent) {
 529                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 530                 key.offset = node->parent;
 531                 size = sizeof(struct btrfs_shared_data_ref);
 532         } else {
 533                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 534                 key.offset = hash_extent_data_ref(node->ref_root, owner, offset);
 535                 size = sizeof(struct btrfs_extent_data_ref);
 536         }
 537
 538         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 539         if (ret && ret != -EEXIST)
 540                 goto fail;
 541
 542         leaf = path->nodes[0];
 543         if (node->parent) {
 544                 struct btrfs_shared_data_ref *ref;
 545                 ref = btrfs_item_ptr(leaf, path->slots[0],
 546                                      struct btrfs_shared_data_ref);
 547                 if (ret == 0) {
 548                         btrfs_set_shared_data_ref_count(leaf, ref, node->ref_mod);
 549                 } else {
 550                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 551                         num_refs += node->ref_mod;
 552                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 553                 }
 554         } else {
 555                 struct btrfs_extent_data_ref *ref;
 556                 while (ret == -EEXIST) {
 557                         ref = btrfs_item_ptr(leaf, path->slots[0],
 558                                              struct btrfs_extent_data_ref);
 559                         if (match_extent_data_ref(leaf, ref, node->ref_root,
 560                                                   owner, offset))
 561                                 break;
 562                         btrfs_release_path(path);
 563                         key.offset++;
 564                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 565                                                       size);
 566                         if (ret && ret != -EEXIST)
 567                                 goto fail;
 568
 569                         leaf = path->nodes[0];
 570                 }
 571                 ref = btrfs_item_ptr(leaf, path->slots[0],
 572                                      struct btrfs_extent_data_ref);
 573                 if (ret == 0) {
 574                         btrfs_set_extent_data_ref_root(leaf, ref, node->ref_root);
 575                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 576                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 577                         btrfs_set_extent_data_ref_count(leaf, ref, node->ref_mod);
 578                 } else {
 579                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
 580                         num_refs += node->ref_mod;
 581                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 582                 }
 583         }
 584         btrfs_mark_buffer_dirty(trans, leaf);
 585         ret = 0;
 586 fail:
 587         btrfs_release_path(path);
 588         return ret;
 589 }
 590
 591 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
 592                                            struct btrfs_root *root,
 593                                            struct btrfs_path *path,
 594                                            int refs_to_drop)
 595 {
 596         struct btrfs_key key;
 597         struct btrfs_extent_data_ref *ref1 = NULL;
 598         struct btrfs_shared_data_ref *ref2 = NULL;
 599         struct extent_buffer *leaf;
 600         u32 num_refs = 0;
 601         int ret = 0;
 602
 603         leaf = path->nodes[0];
 604         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 605
 606         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 607                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 608                                       struct btrfs_extent_data_ref);
 609                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 610         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 611                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 612                                       struct btrfs_shared_data_ref);
 613                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 614         } else {
 615                 btrfs_err(trans->fs_info,
 616                           "unrecognized backref key (%llu %u %llu)",
 617                           key.objectid, key.type, key.offset);
 618                 btrfs_abort_transaction(trans, -EUCLEAN);
 619                 return -EUCLEAN;
 620         }
 621
 622         BUG_ON(num_refs < refs_to_drop);
 623         num_refs -= refs_to_drop;
 624
 625         if (num_refs == 0) {
 626                 ret = btrfs_del_item(trans, root, path);
 627         } else {
 628                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
 629                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
 630                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
 631                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
 632                 btrfs_mark_buffer_dirty(trans, leaf);
 633         }
 634         return ret;
 635 }
 636
 637 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
 638                                           struct btrfs_extent_inline_ref *iref)
 639 {
 640         struct btrfs_key key;
 641         struct extent_buffer *leaf;
 642         struct btrfs_extent_data_ref *ref1;
 643         struct btrfs_shared_data_ref *ref2;
 644         u32 num_refs = 0;
 645         int type;
 646
 647         leaf = path->nodes[0];
 648         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 649
 650         if (iref) {
 651                 /*
 652                  * If type is invalid, we should have bailed out earlier than
 653                  * this call.
 654                  */
 655                 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
 656                 ASSERT(type != BTRFS_REF_TYPE_INVALID);
 657                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
 658                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
 659                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 660                 } else {
 661                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
 662                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 663                 }
 664         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
 665                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
 666                                       struct btrfs_extent_data_ref);
 667                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
 668         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
 669                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
 670                                       struct btrfs_shared_data_ref);
 671                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
 672         } else {
 673                 WARN_ON(1);
 674         }
 675         return num_refs;
 676 }
 677
 678 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
 679                                           struct btrfs_path *path,
 680                                           u64 bytenr, u64 parent,
 681                                           u64 root_objectid)
 682 {
 683         struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
 684         struct btrfs_key key;
 685         int ret;
 686
 687         key.objectid = bytenr;
 688         if (parent) {
 689                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 690                 key.offset = parent;
 691         } else {
 692                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 693                 key.offset = root_objectid;
 694         }
 695
 696         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 697         if (ret > 0)
 698                 ret = -ENOENT;
 699         return ret;
 700 }
 701
 702 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
 703                                           struct btrfs_path *path,
 704                                           struct btrfs_delayed_ref_node *node,
 705                                           u64 bytenr)
 706 {
 707         struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
 708         struct btrfs_key key;
 709         int ret;
 710
 711         key.objectid = bytenr;
 712         if (node->parent) {
 713                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
 714                 key.offset = node->parent;
 715         } else {
 716                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
 717                 key.offset = node->ref_root;
 718         }
 719
 720         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 721         btrfs_release_path(path);
 722         return ret;
 723 }
 724
 725 static inline int extent_ref_type(u64 parent, u64 owner)
 726 {
 727         int type;
 728         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 729                 if (parent > 0)
 730                         type = BTRFS_SHARED_BLOCK_REF_KEY;
 731                 else
 732                         type = BTRFS_TREE_BLOCK_REF_KEY;
 733         } else {
 734                 if (parent > 0)
 735                         type = BTRFS_SHARED_DATA_REF_KEY;
 736                 else
 737                         type = BTRFS_EXTENT_DATA_REF_KEY;
 738         }
 739         return type;
 740 }
 741
 742 static int find_next_key(struct btrfs_path *path, int level,
 743                          struct btrfs_key *key)
 744
 745 {
 746         for (; level < BTRFS_MAX_LEVEL; level++) {
 747                 if (!path->nodes[level])
 748                         break;
 749                 if (path->slots[level] + 1 >=
 750                     btrfs_header_nritems(path->nodes[level]))
 751                         continue;
 752                 if (level == 0)
 753                         btrfs_item_key_to_cpu(path->nodes[level], key,
 754                                               path->slots[level] + 1);
 755                 else
 756                         btrfs_node_key_to_cpu(path->nodes[level], key,
 757                                               path->slots[level] + 1);
 758                 return 0;
 759         }
 760         return 1;
 761 }
 762
 763 /*
 764  * look for inline back ref. if back ref is found, *ref_ret is set
 765  * to the address of inline back ref, and 0 is returned.
 766  *
 767  * if back ref isn't found, *ref_ret is set to the address where it
 768  * should be inserted, and -ENOENT is returned.
 769  *
 770  * if insert is true and there are too many inline back refs, the path
 771  * points to the extent item, and -EAGAIN is returned.
 772  *
 773  * NOTE: inline back refs are ordered in the same way that back ref
 774  *       items in the tree are ordered.
 775  */
 776 static noinline_for_stack
 777 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
 778                                  struct btrfs_path *path,
 779                                  struct btrfs_extent_inline_ref **ref_ret,
 780                                  u64 bytenr, u64 num_bytes,
 781                                  u64 parent, u64 root_objectid,
 782                                  u64 owner, u64 offset, int insert)
 783 {
 784         struct btrfs_fs_info *fs_info = trans->fs_info;
 785         struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
 786         struct btrfs_key key;
 787         struct extent_buffer *leaf;
 788         struct btrfs_extent_item *ei;
 789         struct btrfs_extent_inline_ref *iref;
 790         u64 flags;
 791         u64 item_size;
 792         unsigned long ptr;
 793         unsigned long end;
 794         int extra_size;
 795         int type;
 796         int want;
 797         int ret;
 798         bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
 799         int needed;
 800
 801         key.objectid = bytenr;
 802         key.type = BTRFS_EXTENT_ITEM_KEY;
 803         key.offset = num_bytes;
 804
 805         want = extent_ref_type(parent, owner);
 806         if (insert) {
 807                 extra_size = btrfs_extent_inline_ref_size(want);
 808                 path->search_for_extension = 1;
 809                 path->keep_locks = 1;
 810         } else
 811                 extra_size = -1;
 812
 813         /*
 814          * Owner is our level, so we can just add one to get the level for the
 815          * block we are interested in.
 816          */
 817         if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
 818                 key.type = BTRFS_METADATA_ITEM_KEY;
 819                 key.offset = owner;
 820         }
 821
 822 again:
 823         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
 824         if (ret < 0)
 825                 goto out;
 826
 827         /*
 828          * We may be a newly converted file system which still has the old fat
 829          * extent entries for metadata, so try and see if we have one of those.
 830          */
 831         if (ret > 0 && skinny_metadata) {
 832                 skinny_metadata = false;
 833                 if (path->slots[0]) {
 834                         path->slots[0]--;
 835                         btrfs_item_key_to_cpu(path->nodes[0], &key,
 836                                               path->slots[0]);
 837                         if (key.objectid == bytenr &&
 838                             key.type == BTRFS_EXTENT_ITEM_KEY &&
 839                             key.offset == num_bytes)
 840                                 ret = 0;
 841                 }
 842                 if (ret) {
 843                         key.objectid = bytenr;
 844                         key.type = BTRFS_EXTENT_ITEM_KEY;
 845                         key.offset = num_bytes;
 846                         btrfs_release_path(path);
 847                         goto again;
 848                 }
 849         }
 850
 851         if (ret && !insert) {
 852                 ret = -ENOENT;
 853                 goto out;
 854         } else if (WARN_ON(ret)) {
 855                 btrfs_print_leaf(path->nodes[0]);
 856                 btrfs_err(fs_info,
 857 "extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu",
 858                           bytenr, num_bytes, parent, root_objectid, owner,
 859                           offset);
 860                 ret = -EUCLEAN;
 861                 goto out;
 862         }
 863
 864         leaf = path->nodes[0];
 865         item_size = btrfs_item_size(leaf, path->slots[0]);
 866         if (unlikely(item_size < sizeof(*ei))) {
 867                 ret = -EUCLEAN;
 868                 btrfs_err(fs_info,
 869                           "unexpected extent item size, has %llu expect >= %zu",
 870                           item_size, sizeof(*ei));
 871                 btrfs_abort_transaction(trans, ret);
 872                 goto out;
 873         }
 874
 875         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 876         flags = btrfs_extent_flags(leaf, ei);
 877
 878         ptr = (unsigned long)(ei + 1);
 879         end = (unsigned long)ei + item_size;
 880
 881         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
 882                 ptr += sizeof(struct btrfs_tree_block_info);
 883                 BUG_ON(ptr > end);
 884         }
 885
 886         if (owner >= BTRFS_FIRST_FREE_OBJECTID)
 887                 needed = BTRFS_REF_TYPE_DATA;
 888         else
 889                 needed = BTRFS_REF_TYPE_BLOCK;
 890
 891         ret = -ENOENT;
 892         while (ptr < end) {
 893                 iref = (struct btrfs_extent_inline_ref *)ptr;
 894                 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
 895                 if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
 896                         ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
 897                         ptr += btrfs_extent_inline_ref_size(type);
 898                         continue;
 899                 }
 900                 if (type == BTRFS_REF_TYPE_INVALID) {
 901                         ret = -EUCLEAN;
 902                         goto out;
 903                 }
 904
 905                 if (want < type)
 906                         break;
 907                 if (want > type) {
 908                         ptr += btrfs_extent_inline_ref_size(type);
 909                         continue;
 910                 }
 911
 912                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
 913                         struct btrfs_extent_data_ref *dref;
 914                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 915                         if (match_extent_data_ref(leaf, dref, root_objectid,
 916                                                   owner, offset)) {
 917                                 ret = 0;
 918                                 break;
 919                         }
 920                         if (hash_extent_data_ref_item(leaf, dref) <
 921                             hash_extent_data_ref(root_objectid, owner, offset))
 922                                 break;
 923                 } else {
 924                         u64 ref_offset;
 925                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
 926                         if (parent > 0) {
 927                                 if (parent == ref_offset) {
 928                                         ret = 0;
 929                                         break;
 930                                 }
 931                                 if (ref_offset < parent)
 932                                         break;
 933                         } else {
 934                                 if (root_objectid == ref_offset) {
 935                                         ret = 0;
 936                                         break;
 937                                 }
 938                                 if (ref_offset < root_objectid)
 939                                         break;
 940                         }
 941                 }
 942                 ptr += btrfs_extent_inline_ref_size(type);
 943         }
 944
 945         if (unlikely(ptr > end)) {
 946                 ret = -EUCLEAN;
 947                 btrfs_print_leaf(path->nodes[0]);
 948                 btrfs_crit(fs_info,
 949 "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu",
 950                            path->slots[0], root_objectid, owner, offset, parent);
 951                 goto out;
 952         }
 953
 954         if (ret == -ENOENT && insert) {
 955                 if (item_size + extra_size >=
 956                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
 957                         ret = -EAGAIN;
 958                         goto out;
 959                 }
 960                 /*
 961                  * To add new inline back ref, we have to make sure
 962                  * there is no corresponding back ref item.
 963                  * For simplicity, we just do not add new inline back
 964                  * ref if there is any kind of item for this block
 965                  */
 966                 if (find_next_key(path, 0, &key) == 0 &&
 967                     key.objectid == bytenr &&
 968                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
 969                         ret = -EAGAIN;
 970                         goto out;
 971                 }
 972         }
 973         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
 974 out:
 975         if (insert) {
 976                 path->keep_locks = 0;
 977                 path->search_for_extension = 0;
 978                 btrfs_unlock_up_safe(path, 1);
 979         }
 980         return ret;
 981 }
 982
 983 /*
 984  * helper to add new inline back ref
 985  */
 986 static noinline_for_stack
 987 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
 988                                  struct btrfs_path *path,
 989                                  struct btrfs_extent_inline_ref *iref,
 990                                  u64 parent, u64 root_objectid,
 991                                  u64 owner, u64 offset, int refs_to_add,
 992                                  struct btrfs_delayed_extent_op *extent_op)
 993 {
 994         struct extent_buffer *leaf;
 995         struct btrfs_extent_item *ei;
 996         unsigned long ptr;
 997         unsigned long end;
 998         unsigned long item_offset;
 999         u64 refs;
1000         int size;
1001         int type;
1002
1003         leaf = path->nodes[0];
1004         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1005         item_offset = (unsigned long)iref - (unsigned long)ei;
1006
1007         type = extent_ref_type(parent, owner);
1008         size = btrfs_extent_inline_ref_size(type);
1009
1010         btrfs_extend_item(trans, path, size);
1011
1012         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1013         refs = btrfs_extent_refs(leaf, ei);
1014         refs += refs_to_add;
1015         btrfs_set_extent_refs(leaf, ei, refs);
1016         if (extent_op)
1017                 __run_delayed_extent_op(extent_op, leaf, ei);
1018
1019         ptr = (unsigned long)ei + item_offset;
1020         end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]);
1021         if (ptr < end - size)
1022                 memmove_extent_buffer(leaf, ptr + size, ptr,
1023                                       end - size - ptr);
1024
1025         iref = (struct btrfs_extent_inline_ref *)ptr;
1026         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1027         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1028                 struct btrfs_extent_data_ref *dref;
1029                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1030                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1031                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1032                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1033                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1034         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1035                 struct btrfs_shared_data_ref *sref;
1036                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1037                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1038                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1039         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1040                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1041         } else {
1042                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1043         }
1044         btrfs_mark_buffer_dirty(trans, leaf);
1045 }
1046
1047 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1048                                  struct btrfs_path *path,
1049                                  struct btrfs_extent_inline_ref **ref_ret,
1050                                  u64 bytenr, u64 num_bytes, u64 parent,
1051                                  u64 root_objectid, u64 owner, u64 offset)
1052 {
1053         int ret;
1054
1055         ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1056                                            num_bytes, parent, root_objectid,
1057                                            owner, offset, 0);
1058         if (ret != -ENOENT)
1059                 return ret;
1060
1061         btrfs_release_path(path);
1062         *ref_ret = NULL;
1063
1064         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1065                 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1066                                             root_objectid);
1067         } else {
1068                 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1069                                              root_objectid, owner, offset);
1070         }
1071         return ret;
1072 }
1073
1074 /*
1075  * helper to update/remove inline back ref
1076  */
1077 static noinline_for_stack int update_inline_extent_backref(
1078                                   struct btrfs_trans_handle *trans,
1079                                   struct btrfs_path *path,
1080                                   struct btrfs_extent_inline_ref *iref,
1081                                   int refs_to_mod,
1082                                   struct btrfs_delayed_extent_op *extent_op)
1083 {
1084         struct extent_buffer *leaf = path->nodes[0];
1085         struct btrfs_fs_info *fs_info = leaf->fs_info;
1086         struct btrfs_extent_item *ei;
1087         struct btrfs_extent_data_ref *dref = NULL;
1088         struct btrfs_shared_data_ref *sref = NULL;
1089         unsigned long ptr;
1090         unsigned long end;
1091         u32 item_size;
1092         int size;
1093         int type;
1094         u64 refs;
1095
1096         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1097         refs = btrfs_extent_refs(leaf, ei);
1098         if (unlikely(refs_to_mod < 0 && refs + refs_to_mod <= 0)) {
1099                 struct btrfs_key key;
1100                 u32 extent_size;
1101
1102                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1103                 if (key.type == BTRFS_METADATA_ITEM_KEY)
1104                         extent_size = fs_info->nodesize;
1105                 else
1106                         extent_size = key.offset;
1107                 btrfs_print_leaf(leaf);
1108                 btrfs_err(fs_info,
1109         "invalid refs_to_mod for extent %llu num_bytes %u, has %d expect >= -%llu",
1110                           key.objectid, extent_size, refs_to_mod, refs);
1111                 return -EUCLEAN;
1112         }
1113         refs += refs_to_mod;
1114         btrfs_set_extent_refs(leaf, ei, refs);
1115         if (extent_op)
1116                 __run_delayed_extent_op(extent_op, leaf, ei);
1117
1118         type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1119         /*
1120          * Function btrfs_get_extent_inline_ref_type() has already printed
1121          * error messages.
1122          */
1123         if (unlikely(type == BTRFS_REF_TYPE_INVALID))
1124                 return -EUCLEAN;
1125
1126         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1127                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1128                 refs = btrfs_extent_data_ref_count(leaf, dref);
1129         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1130                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1131                 refs = btrfs_shared_data_ref_count(leaf, sref);
1132         } else {
1133                 refs = 1;
1134                 /*
1135                  * For tree blocks we can only drop one ref for it, and tree
1136                  * blocks should not have refs > 1.
1137                  *
1138                  * Furthermore if we're inserting a new inline backref, we
1139                  * won't reach this path either. That would be
1140                  * setup_inline_extent_backref().
1141                  */
1142                 if (unlikely(refs_to_mod != -1)) {
1143                         struct btrfs_key key;
1144
1145                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1146
1147                         btrfs_print_leaf(leaf);
1148                         btrfs_err(fs_info,
1149                         "invalid refs_to_mod for tree block %llu, has %d expect -1",
1150                                   key.objectid, refs_to_mod);
1151                         return -EUCLEAN;
1152                 }
1153         }
1154
1155         if (unlikely(refs_to_mod < 0 && refs < -refs_to_mod)) {
1156                 struct btrfs_key key;
1157                 u32 extent_size;
1158
1159                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1160                 if (key.type == BTRFS_METADATA_ITEM_KEY)
1161                         extent_size = fs_info->nodesize;
1162                 else
1163                         extent_size = key.offset;
1164                 btrfs_print_leaf(leaf);
1165                 btrfs_err(fs_info,
1166 "invalid refs_to_mod for backref entry, iref %lu extent %llu num_bytes %u, has %d expect >= -%llu",
1167                           (unsigned long)iref, key.objectid, extent_size,
1168                           refs_to_mod, refs);
1169                 return -EUCLEAN;
1170         }
1171         refs += refs_to_mod;
1172
1173         if (refs > 0) {
1174                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1175                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1176                 else
1177                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1178         } else {
1179                 size =  btrfs_extent_inline_ref_size(type);
1180                 item_size = btrfs_item_size(leaf, path->slots[0]);
1181                 ptr = (unsigned long)iref;
1182                 end = (unsigned long)ei + item_size;
1183                 if (ptr + size < end)
1184                         memmove_extent_buffer(leaf, ptr, ptr + size,
1185                                               end - ptr - size);
1186                 item_size -= size;
1187                 btrfs_truncate_item(trans, path, item_size, 1);
1188         }
1189         btrfs_mark_buffer_dirty(trans, leaf);
1190         return 0;
1191 }
1192
1193 static noinline_for_stack
1194 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1195                                  struct btrfs_path *path,
1196                                  u64 bytenr, u64 num_bytes, u64 parent,
1197                                  u64 root_objectid, u64 owner,
1198                                  u64 offset, int refs_to_add,
1199                                  struct btrfs_delayed_extent_op *extent_op)
1200 {
1201         struct btrfs_extent_inline_ref *iref;
1202         int ret;
1203
1204         ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1205                                            num_bytes, parent, root_objectid,
1206                                            owner, offset, 1);
1207         if (ret == 0) {
1208                 /*
1209                  * We're adding refs to a tree block we already own, this
1210                  * should not happen at all.
1211                  */
1212                 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1213                         btrfs_print_leaf(path->nodes[0]);
1214                         btrfs_crit(trans->fs_info,
1215 "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu slot %u",
1216                                    bytenr, num_bytes, root_objectid, path->slots[0]);
1217                         return -EUCLEAN;
1218                 }
1219                 ret = update_inline_extent_backref(trans, path, iref,
1220                                                    refs_to_add, extent_op);
1221         } else if (ret == -ENOENT) {
1222                 setup_inline_extent_backref(trans, path, iref, parent,
1223                                             root_objectid, owner, offset,
1224                                             refs_to_add, extent_op);
1225                 ret = 0;
1226         }
1227         return ret;
1228 }
1229
1230 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1231                                  struct btrfs_root *root,
1232                                  struct btrfs_path *path,
1233                                  struct btrfs_extent_inline_ref *iref,
1234                                  int refs_to_drop, int is_data)
1235 {
1236         int ret = 0;
1237
1238         BUG_ON(!is_data && refs_to_drop != 1);
1239         if (iref)
1240                 ret = update_inline_extent_backref(trans, path, iref,
1241                                                    -refs_to_drop, NULL);
1242         else if (is_data)
1243                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1244         else
1245                 ret = btrfs_del_item(trans, root, path);
1246         return ret;
1247 }
1248
1249 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1250                                u64 *discarded_bytes)
1251 {
1252         int j, ret = 0;
1253         u64 bytes_left, end;
1254         u64 aligned_start = ALIGN(start, 1 << SECTOR_SHIFT);
1255
1256         /* Adjust the range to be aligned to 512B sectors if necessary. */
1257         if (start != aligned_start) {
1258                 len -= aligned_start - start;
1259                 len = round_down(len, 1 << SECTOR_SHIFT);
1260                 start = aligned_start;
1261         }
1262
1263         *discarded_bytes = 0;
1264
1265         if (!len)
1266                 return 0;
1267
1268         end = start + len;
1269         bytes_left = len;
1270
1271         /* Skip any superblocks on this device. */
1272         for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1273                 u64 sb_start = btrfs_sb_offset(j);
1274                 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1275                 u64 size = sb_start - start;
1276
1277                 if (!in_range(sb_start, start, bytes_left) &&
1278                     !in_range(sb_end, start, bytes_left) &&
1279                     !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1280                         continue;
1281
1282                 /*
1283                  * Superblock spans beginning of range.  Adjust start and
1284                  * try again.
1285                  */
1286                 if (sb_start <= start) {
1287                         start += sb_end - start;
1288                         if (start > end) {
1289                                 bytes_left = 0;
1290                                 break;
1291                         }
1292                         bytes_left = end - start;
1293                         continue;
1294                 }
1295
1296                 if (size) {
1297                         ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1298                                                    size >> SECTOR_SHIFT,
1299                                                    GFP_NOFS);
1300                         if (!ret)
1301                                 *discarded_bytes += size;
1302                         else if (ret != -EOPNOTSUPP)
1303                                 return ret;
1304                 }
1305
1306                 start = sb_end;
1307                 if (start > end) {
1308                         bytes_left = 0;
1309                         break;
1310                 }
1311                 bytes_left = end - start;
1312         }
1313
1314         if (bytes_left) {
1315                 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1316                                            bytes_left >> SECTOR_SHIFT,
1317                                            GFP_NOFS);
1318                 if (!ret)
1319                         *discarded_bytes += bytes_left;
1320         }
1321         return ret;
1322 }
1323
1324 static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes)
1325 {
1326         struct btrfs_device *dev = stripe->dev;
1327         struct btrfs_fs_info *fs_info = dev->fs_info;
1328         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1329         u64 phys = stripe->physical;
1330         u64 len = stripe->length;
1331         u64 discarded = 0;
1332         int ret = 0;
1333
1334         /* Zone reset on a zoned filesystem */
1335         if (btrfs_can_zone_reset(dev, phys, len)) {
1336                 u64 src_disc;
1337
1338                 ret = btrfs_reset_device_zone(dev, phys, len, &discarded);
1339                 if (ret)
1340                         goto out;
1341
1342                 if (!btrfs_dev_replace_is_ongoing(dev_replace) ||
1343                     dev != dev_replace->srcdev)
1344                         goto out;
1345
1346                 src_disc = discarded;
1347
1348                 /* Send to replace target as well */
1349                 ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len,
1350                                               &discarded);
1351                 discarded += src_disc;
1352         } else if (bdev_max_discard_sectors(stripe->dev->bdev)) {
1353                 ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded);
1354         } else {
1355                 ret = 0;
1356                 *bytes = 0;
1357         }
1358
1359 out:
1360         *bytes = discarded;
1361         return ret;
1362 }
1363
1364 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1365                          u64 num_bytes, u64 *actual_bytes)
1366 {
1367         int ret = 0;
1368         u64 discarded_bytes = 0;
1369         u64 end = bytenr + num_bytes;
1370         u64 cur = bytenr;
1371
1372         /*
1373          * Avoid races with device replace and make sure the devices in the
1374          * stripes don't go away while we are discarding.
1375          */
1376         btrfs_bio_counter_inc_blocked(fs_info);
1377         while (cur < end) {
1378                 struct btrfs_discard_stripe *stripes;
1379                 unsigned int num_stripes;
1380                 int i;
1381
1382                 num_bytes = end - cur;
1383                 stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes);
1384                 if (IS_ERR(stripes)) {
1385                         ret = PTR_ERR(stripes);
1386                         if (ret == -EOPNOTSUPP)
1387                                 ret = 0;
1388                         break;
1389                 }
1390
1391                 for (i = 0; i < num_stripes; i++) {
1392                         struct btrfs_discard_stripe *stripe = stripes + i;
1393                         u64 bytes;
1394
1395                         if (!stripe->dev->bdev) {
1396                                 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1397                                 continue;
1398                         }
1399
1400                         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
1401                                         &stripe->dev->dev_state))
1402                                 continue;
1403
1404                         ret = do_discard_extent(stripe, &bytes);
1405                         if (ret) {
1406                                 /*
1407                                  * Keep going if discard is not supported by the
1408                                  * device.
1409                                  */
1410                                 if (ret != -EOPNOTSUPP)
1411                                         break;
1412                                 ret = 0;
1413                         } else {
1414                                 discarded_bytes += bytes;
1415                         }
1416                 }
1417                 kfree(stripes);
1418                 if (ret)
1419                         break;
1420                 cur += num_bytes;
1421         }
1422         btrfs_bio_counter_dec(fs_info);
1423         if (actual_bytes)
1424                 *actual_bytes = discarded_bytes;
1425         return ret;
1426 }
1427
1428 /* Can return -ENOMEM */
1429 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1430                          struct btrfs_ref *generic_ref)
1431 {
1432         struct btrfs_fs_info *fs_info = trans->fs_info;
1433         int ret;
1434
1435         ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1436                generic_ref->action);
1437         BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1438                generic_ref->ref_root == BTRFS_TREE_LOG_OBJECTID);
1439
1440         if (generic_ref->type == BTRFS_REF_METADATA)
1441                 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1442         else
1443                 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1444
1445         btrfs_ref_tree_mod(fs_info, generic_ref);
1446
1447         return ret;
1448 }
1449
1450 /*
1451  * Insert backreference for a given extent.
1452  *
1453  * The counterpart is in __btrfs_free_extent(), with examples and more details
1454  * how it works.
1455  *
1456  * @trans:          Handle of transaction
1457  *
1458  * @node:           The delayed ref node used to get the bytenr/length for
1459  *                  extent whose references are incremented.
1460  *
1461  * @extent_op       Pointer to a structure, holding information necessary when
1462  *                  updating a tree block's flags
1463  *
1464  */
1465 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1466                                   struct btrfs_delayed_ref_node *node,
1467                                   struct btrfs_delayed_extent_op *extent_op)
1468 {
1469         struct btrfs_path *path;
1470         struct extent_buffer *leaf;
1471         struct btrfs_extent_item *item;
1472         struct btrfs_key key;
1473         u64 bytenr = node->bytenr;
1474         u64 num_bytes = node->num_bytes;
1475         u64 owner = btrfs_delayed_ref_owner(node);
1476         u64 offset = btrfs_delayed_ref_offset(node);
1477         u64 refs;
1478         int refs_to_add = node->ref_mod;
1479         int ret;
1480
1481         path = btrfs_alloc_path();
1482         if (!path)
1483                 return -ENOMEM;
1484
1485         /* this will setup the path even if it fails to insert the back ref */
1486         ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1487                                            node->parent, node->ref_root, owner,
1488                                            offset, refs_to_add, extent_op);
1489         if ((ret < 0 && ret != -EAGAIN) || !ret)
1490                 goto out;
1491
1492         /*
1493          * Ok we had -EAGAIN which means we didn't have space to insert and
1494          * inline extent ref, so just update the reference count and add a
1495          * normal backref.
1496          */
1497         leaf = path->nodes[0];
1498         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1499         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1500         refs = btrfs_extent_refs(leaf, item);
1501         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1502         if (extent_op)
1503                 __run_delayed_extent_op(extent_op, leaf, item);
1504
1505         btrfs_mark_buffer_dirty(trans, leaf);
1506         btrfs_release_path(path);
1507
1508         /* now insert the actual backref */
1509         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1510                 ret = insert_tree_block_ref(trans, path, node, bytenr);
1511         else
1512                 ret = insert_extent_data_ref(trans, path, node, bytenr);
1513
1514         if (ret)
1515                 btrfs_abort_transaction(trans, ret);
1516 out:
1517         btrfs_free_path(path);
1518         return ret;
1519 }
1520
1521 static void free_head_ref_squota_rsv(struct btrfs_fs_info *fs_info,
1522                                      struct btrfs_delayed_ref_head *href)
1523 {
1524         u64 root = href->owning_root;
1525
1526         /*
1527          * Don't check must_insert_reserved, as this is called from contexts
1528          * where it has already been unset.
1529          */
1530         if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE ||
1531             !href->is_data || !is_fstree(root))
1532                 return;
1533
1534         btrfs_qgroup_free_refroot(fs_info, root, href->reserved_bytes,
1535                                   BTRFS_QGROUP_RSV_DATA);
1536 }
1537
1538 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1539                                 struct btrfs_delayed_ref_head *href,
1540                                 struct btrfs_delayed_ref_node *node,
1541                                 struct btrfs_delayed_extent_op *extent_op,
1542                                 bool insert_reserved)
1543 {
1544         int ret = 0;
1545         u64 parent = 0;
1546         u64 flags = 0;
1547
1548         trace_run_delayed_data_ref(trans->fs_info, node);
1549
1550         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1551                 parent = node->parent;
1552
1553         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1554                 struct btrfs_key key;
1555                 struct btrfs_squota_delta delta = {
1556                         .root = href->owning_root,
1557                         .num_bytes = node->num_bytes,
1558                         .is_data = true,
1559                         .is_inc = true,
1560                         .generation = trans->transid,
1561                 };
1562                 u64 owner = btrfs_delayed_ref_owner(node);
1563                 u64 offset = btrfs_delayed_ref_offset(node);
1564
1565                 if (extent_op)
1566                         flags |= extent_op->flags_to_set;
1567
1568                 key.objectid = node->bytenr;
1569                 key.type = BTRFS_EXTENT_ITEM_KEY;
1570                 key.offset = node->num_bytes;
1571
1572                 ret = alloc_reserved_file_extent(trans, parent, node->ref_root,
1573                                                  flags, owner, offset, &key,
1574                                                  node->ref_mod,
1575                                                  href->owning_root);
1576                 free_head_ref_squota_rsv(trans->fs_info, href);
1577                 if (!ret)
1578                         ret = btrfs_record_squota_delta(trans->fs_info, &delta);
1579         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1580                 ret = __btrfs_inc_extent_ref(trans, node, extent_op);
1581         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1582                 ret = __btrfs_free_extent(trans, href, node, extent_op);
1583         } else {
1584                 BUG();
1585         }
1586         return ret;
1587 }
1588
1589 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1590                                     struct extent_buffer *leaf,
1591                                     struct btrfs_extent_item *ei)
1592 {
1593         u64 flags = btrfs_extent_flags(leaf, ei);
1594         if (extent_op->update_flags) {
1595                 flags |= extent_op->flags_to_set;
1596                 btrfs_set_extent_flags(leaf, ei, flags);
1597         }
1598
1599         if (extent_op->update_key) {
1600                 struct btrfs_tree_block_info *bi;
1601                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1602                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1603                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1604         }
1605 }
1606
1607 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1608                                  struct btrfs_delayed_ref_head *head,
1609                                  struct btrfs_delayed_extent_op *extent_op)
1610 {
1611         struct btrfs_fs_info *fs_info = trans->fs_info;
1612         struct btrfs_root *root;
1613         struct btrfs_key key;
1614         struct btrfs_path *path;
1615         struct btrfs_extent_item *ei;
1616         struct extent_buffer *leaf;
1617         u32 item_size;
1618         int ret;
1619         int metadata = 1;
1620
1621         if (TRANS_ABORTED(trans))
1622                 return 0;
1623
1624         if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1625                 metadata = 0;
1626
1627         path = btrfs_alloc_path();
1628         if (!path)
1629                 return -ENOMEM;
1630
1631         key.objectid = head->bytenr;
1632
1633         if (metadata) {
1634                 key.type = BTRFS_METADATA_ITEM_KEY;
1635                 key.offset = extent_op->level;
1636         } else {
1637                 key.type = BTRFS_EXTENT_ITEM_KEY;
1638                 key.offset = head->num_bytes;
1639         }
1640
1641         root = btrfs_extent_root(fs_info, key.objectid);
1642 again:
1643         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1644         if (ret < 0) {
1645                 goto out;
1646         } else if (ret > 0) {
1647                 if (metadata) {
1648                         if (path->slots[0] > 0) {
1649                                 path->slots[0]--;
1650                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
1651                                                       path->slots[0]);
1652                                 if (key.objectid == head->bytenr &&
1653                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
1654                                     key.offset == head->num_bytes)
1655                                         ret = 0;
1656                         }
1657                         if (ret > 0) {
1658                                 btrfs_release_path(path);
1659                                 metadata = 0;
1660
1661                                 key.objectid = head->bytenr;
1662                                 key.offset = head->num_bytes;
1663                                 key.type = BTRFS_EXTENT_ITEM_KEY;
1664                                 goto again;
1665                         }
1666                 } else {
1667                         ret = -EUCLEAN;
1668                         btrfs_err(fs_info,
1669                   "missing extent item for extent %llu num_bytes %llu level %d",
1670                                   head->bytenr, head->num_bytes, extent_op->level);
1671                         goto out;
1672                 }
1673         }
1674
1675         leaf = path->nodes[0];
1676         item_size = btrfs_item_size(leaf, path->slots[0]);
1677
1678         if (unlikely(item_size < sizeof(*ei))) {
1679                 ret = -EUCLEAN;
1680                 btrfs_err(fs_info,
1681                           "unexpected extent item size, has %u expect >= %zu",
1682                           item_size, sizeof(*ei));
1683                 btrfs_abort_transaction(trans, ret);
1684                 goto out;
1685         }
1686
1687         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688         __run_delayed_extent_op(extent_op, leaf, ei);
1689
1690         btrfs_mark_buffer_dirty(trans, leaf);
1691 out:
1692         btrfs_free_path(path);
1693         return ret;
1694 }
1695
1696 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1697                                 struct btrfs_delayed_ref_head *href,
1698                                 struct btrfs_delayed_ref_node *node,
1699                                 struct btrfs_delayed_extent_op *extent_op,
1700                                 bool insert_reserved)
1701 {
1702         int ret = 0;
1703         struct btrfs_fs_info *fs_info = trans->fs_info;
1704         u64 parent = 0;
1705         u64 ref_root = 0;
1706
1707         trace_run_delayed_tree_ref(trans->fs_info, node);
1708
1709         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1710                 parent = node->parent;
1711         ref_root = node->ref_root;
1712
1713         if (unlikely(node->ref_mod != 1)) {
1714                 btrfs_err(trans->fs_info,
1715         "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu",
1716                           node->bytenr, node->ref_mod, node->action, ref_root,
1717                           parent);
1718                 return -EUCLEAN;
1719         }
1720         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1721                 struct btrfs_squota_delta delta = {
1722                         .root = href->owning_root,
1723                         .num_bytes = fs_info->nodesize,
1724                         .is_data = false,
1725                         .is_inc = true,
1726                         .generation = trans->transid,
1727                 };
1728
1729                 BUG_ON(!extent_op || !extent_op->update_flags);
1730                 ret = alloc_reserved_tree_block(trans, node, extent_op);
1731                 if (!ret)
1732                         btrfs_record_squota_delta(fs_info, &delta);
1733         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1734                 ret = __btrfs_inc_extent_ref(trans, node, extent_op);
1735         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1736                 ret = __btrfs_free_extent(trans, href, node, extent_op);
1737         } else {
1738                 BUG();
1739         }
1740         return ret;
1741 }
1742
1743 /* helper function to actually process a single delayed ref entry */
1744 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1745                                struct btrfs_delayed_ref_head *href,
1746                                struct btrfs_delayed_ref_node *node,
1747                                struct btrfs_delayed_extent_op *extent_op,
1748                                bool insert_reserved)
1749 {
1750         int ret = 0;
1751
1752         if (TRANS_ABORTED(trans)) {
1753                 if (insert_reserved) {
1754                         btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1755                         free_head_ref_squota_rsv(trans->fs_info, href);
1756                 }
1757                 return 0;
1758         }
1759
1760         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1761             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1762                 ret = run_delayed_tree_ref(trans, href, node, extent_op,
1763                                            insert_reserved);
1764         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1765                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1766                 ret = run_delayed_data_ref(trans, href, node, extent_op,
1767                                            insert_reserved);
1768         else if (node->type == BTRFS_EXTENT_OWNER_REF_KEY)
1769                 ret = 0;
1770         else
1771                 BUG();
1772         if (ret && insert_reserved)
1773                 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1774         if (ret < 0)
1775                 btrfs_err(trans->fs_info,
1776 "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d",
1777                           node->bytenr, node->num_bytes, node->type,
1778                           node->action, node->ref_mod, ret);
1779         return ret;
1780 }
1781
1782 static inline struct btrfs_delayed_ref_node *
1783 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1784 {
1785         struct btrfs_delayed_ref_node *ref;
1786
1787         if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1788                 return NULL;
1789
1790         /*
1791          * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1792          * This is to prevent a ref count from going down to zero, which deletes
1793          * the extent item from the extent tree, when there still are references
1794          * to add, which would fail because they would not find the extent item.
1795          */
1796         if (!list_empty(&head->ref_add_list))
1797                 return list_first_entry(&head->ref_add_list,
1798                                 struct btrfs_delayed_ref_node, add_list);
1799
1800         ref = rb_entry(rb_first_cached(&head->ref_tree),
1801                        struct btrfs_delayed_ref_node, ref_node);
1802         ASSERT(list_empty(&ref->add_list));
1803         return ref;
1804 }
1805
1806 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1807                                       struct btrfs_delayed_ref_head *head)
1808 {
1809         spin_lock(&delayed_refs->lock);
1810         head->processing = false;
1811         delayed_refs->num_heads_ready++;
1812         spin_unlock(&delayed_refs->lock);
1813         btrfs_delayed_ref_unlock(head);
1814 }
1815
1816 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1817                                 struct btrfs_delayed_ref_head *head)
1818 {
1819         struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1820
1821         if (!extent_op)
1822                 return NULL;
1823
1824         if (head->must_insert_reserved) {
1825                 head->extent_op = NULL;
1826                 btrfs_free_delayed_extent_op(extent_op);
1827                 return NULL;
1828         }
1829         return extent_op;
1830 }
1831
1832 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1833                                      struct btrfs_delayed_ref_head *head)
1834 {
1835         struct btrfs_delayed_extent_op *extent_op;
1836         int ret;
1837
1838         extent_op = cleanup_extent_op(head);
1839         if (!extent_op)
1840                 return 0;
1841         head->extent_op = NULL;
1842         spin_unlock(&head->lock);
1843         ret = run_delayed_extent_op(trans, head, extent_op);
1844         btrfs_free_delayed_extent_op(extent_op);
1845         return ret ? ret : 1;
1846 }
1847
1848 u64 btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1849                                   struct btrfs_delayed_ref_root *delayed_refs,
1850                                   struct btrfs_delayed_ref_head *head)
1851 {
1852         u64 ret = 0;
1853
1854         /*
1855          * We had csum deletions accounted for in our delayed refs rsv, we need
1856          * to drop the csum leaves for this update from our delayed_refs_rsv.
1857          */
1858         if (head->total_ref_mod < 0 && head->is_data) {
1859                 int nr_csums;
1860
1861                 spin_lock(&delayed_refs->lock);
1862                 delayed_refs->pending_csums -= head->num_bytes;
1863                 spin_unlock(&delayed_refs->lock);
1864                 nr_csums = btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1865
1866                 btrfs_delayed_refs_rsv_release(fs_info, 0, nr_csums);
1867
1868                 ret = btrfs_calc_delayed_ref_csum_bytes(fs_info, nr_csums);
1869         }
1870         /* must_insert_reserved can be set only if we didn't run the head ref. */
1871         if (head->must_insert_reserved)
1872                 free_head_ref_squota_rsv(fs_info, head);
1873
1874         return ret;
1875 }
1876
1877 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1878                             struct btrfs_delayed_ref_head *head,
1879                             u64 *bytes_released)
1880 {
1881
1882         struct btrfs_fs_info *fs_info = trans->fs_info;
1883         struct btrfs_delayed_ref_root *delayed_refs;
1884         int ret;
1885
1886         delayed_refs = &trans->transaction->delayed_refs;
1887
1888         ret = run_and_cleanup_extent_op(trans, head);
1889         if (ret < 0) {
1890                 unselect_delayed_ref_head(delayed_refs, head);
1891                 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1892                 return ret;
1893         } else if (ret) {
1894                 return ret;
1895         }
1896
1897         /*
1898          * Need to drop our head ref lock and re-acquire the delayed ref lock
1899          * and then re-check to make sure nobody got added.
1900          */
1901         spin_unlock(&head->lock);
1902         spin_lock(&delayed_refs->lock);
1903         spin_lock(&head->lock);
1904         if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1905                 spin_unlock(&head->lock);
1906                 spin_unlock(&delayed_refs->lock);
1907                 return 1;
1908         }
1909         btrfs_delete_ref_head(delayed_refs, head);
1910         spin_unlock(&head->lock);
1911         spin_unlock(&delayed_refs->lock);
1912
1913         if (head->must_insert_reserved) {
1914                 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1915                 if (head->is_data) {
1916                         struct btrfs_root *csum_root;
1917
1918                         csum_root = btrfs_csum_root(fs_info, head->bytenr);
1919                         ret = btrfs_del_csums(trans, csum_root, head->bytenr,
1920                                               head->num_bytes);
1921                 }
1922         }
1923
1924         *bytes_released += btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1925
1926         trace_run_delayed_ref_head(fs_info, head, 0);
1927         btrfs_delayed_ref_unlock(head);
1928         btrfs_put_delayed_ref_head(head);
1929         return ret;
1930 }
1931
1932 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1933                                         struct btrfs_trans_handle *trans)
1934 {
1935         struct btrfs_delayed_ref_root *delayed_refs =
1936                 &trans->transaction->delayed_refs;
1937         struct btrfs_delayed_ref_head *head = NULL;
1938         int ret;
1939
1940         spin_lock(&delayed_refs->lock);
1941         head = btrfs_select_ref_head(delayed_refs);
1942         if (!head) {
1943                 spin_unlock(&delayed_refs->lock);
1944                 return head;
1945         }
1946
1947         /*
1948          * Grab the lock that says we are going to process all the refs for
1949          * this head
1950          */
1951         ret = btrfs_delayed_ref_lock(delayed_refs, head);
1952         spin_unlock(&delayed_refs->lock);
1953
1954         /*
1955          * We may have dropped the spin lock to get the head mutex lock, and
1956          * that might have given someone else time to free the head.  If that's
1957          * true, it has been removed from our list and we can move on.
1958          */
1959         if (ret == -EAGAIN)
1960                 head = ERR_PTR(-EAGAIN);
1961
1962         return head;
1963 }
1964
1965 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1966                                            struct btrfs_delayed_ref_head *locked_ref,
1967                                            u64 *bytes_released)
1968 {
1969         struct btrfs_fs_info *fs_info = trans->fs_info;
1970         struct btrfs_delayed_ref_root *delayed_refs;
1971         struct btrfs_delayed_extent_op *extent_op;
1972         struct btrfs_delayed_ref_node *ref;
1973         bool must_insert_reserved;
1974         int ret;
1975
1976         delayed_refs = &trans->transaction->delayed_refs;
1977
1978         lockdep_assert_held(&locked_ref->mutex);
1979         lockdep_assert_held(&locked_ref->lock);
1980
1981         while ((ref = select_delayed_ref(locked_ref))) {
1982                 if (ref->seq &&
1983                     btrfs_check_delayed_seq(fs_info, ref->seq)) {
1984                         spin_unlock(&locked_ref->lock);
1985                         unselect_delayed_ref_head(delayed_refs, locked_ref);
1986                         return -EAGAIN;
1987                 }
1988
1989                 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1990                 RB_CLEAR_NODE(&ref->ref_node);
1991                 if (!list_empty(&ref->add_list))
1992                         list_del(&ref->add_list);
1993                 /*
1994                  * When we play the delayed ref, also correct the ref_mod on
1995                  * head
1996                  */
1997                 switch (ref->action) {
1998                 case BTRFS_ADD_DELAYED_REF:
1999                 case BTRFS_ADD_DELAYED_EXTENT:
2000                         locked_ref->ref_mod -= ref->ref_mod;
2001                         break;
2002                 case BTRFS_DROP_DELAYED_REF:
2003                         locked_ref->ref_mod += ref->ref_mod;
2004                         break;
2005                 default:
2006                         WARN_ON(1);
2007                 }
2008                 atomic_dec(&delayed_refs->num_entries);
2009
2010                 /*
2011                  * Record the must_insert_reserved flag before we drop the
2012                  * spin lock.
2013                  */
2014                 must_insert_reserved = locked_ref->must_insert_reserved;
2015                 /*
2016                  * Unsetting this on the head ref relinquishes ownership of
2017                  * the rsv_bytes, so it is critical that every possible code
2018                  * path from here forward frees all reserves including qgroup
2019                  * reserve.
2020                  */
2021                 locked_ref->must_insert_reserved = false;
2022
2023                 extent_op = locked_ref->extent_op;
2024                 locked_ref->extent_op = NULL;
2025                 spin_unlock(&locked_ref->lock);
2026
2027                 ret = run_one_delayed_ref(trans, locked_ref, ref, extent_op,
2028                                           must_insert_reserved);
2029                 btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
2030                 *bytes_released += btrfs_calc_delayed_ref_bytes(fs_info, 1);
2031
2032                 btrfs_free_delayed_extent_op(extent_op);
2033                 if (ret) {
2034                         unselect_delayed_ref_head(delayed_refs, locked_ref);
2035                         btrfs_put_delayed_ref(ref);
2036                         return ret;
2037                 }
2038
2039                 btrfs_put_delayed_ref(ref);
2040                 cond_resched();
2041
2042                 spin_lock(&locked_ref->lock);
2043                 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2044         }
2045
2046         return 0;
2047 }
2048
2049 /*
2050  * Returns 0 on success or if called with an already aborted transaction.
2051  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2052  */
2053 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2054                                              u64 min_bytes)
2055 {
2056         struct btrfs_fs_info *fs_info = trans->fs_info;
2057         struct btrfs_delayed_ref_root *delayed_refs;
2058         struct btrfs_delayed_ref_head *locked_ref = NULL;
2059         int ret;
2060         unsigned long count = 0;
2061         unsigned long max_count = 0;
2062         u64 bytes_processed = 0;
2063
2064         delayed_refs = &trans->transaction->delayed_refs;
2065         if (min_bytes == 0) {
2066                 max_count = delayed_refs->num_heads_ready;
2067                 min_bytes = U64_MAX;
2068         }
2069
2070         do {
2071                 if (!locked_ref) {
2072                         locked_ref = btrfs_obtain_ref_head(trans);
2073                         if (IS_ERR_OR_NULL(locked_ref)) {
2074                                 if (PTR_ERR(locked_ref) == -EAGAIN) {
2075                                         continue;
2076                                 } else {
2077                                         break;
2078                                 }
2079                         }
2080                         count++;
2081                 }
2082                 /*
2083                  * We need to try and merge add/drops of the same ref since we
2084                  * can run into issues with relocate dropping the implicit ref
2085                  * and then it being added back again before the drop can
2086                  * finish.  If we merged anything we need to re-loop so we can
2087                  * get a good ref.
2088                  * Or we can get node references of the same type that weren't
2089                  * merged when created due to bumps in the tree mod seq, and
2090                  * we need to merge them to prevent adding an inline extent
2091                  * backref before dropping it (triggering a BUG_ON at
2092                  * insert_inline_extent_backref()).
2093                  */
2094                 spin_lock(&locked_ref->lock);
2095                 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2096
2097                 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, &bytes_processed);
2098                 if (ret < 0 && ret != -EAGAIN) {
2099                         /*
2100                          * Error, btrfs_run_delayed_refs_for_head already
2101                          * unlocked everything so just bail out
2102                          */
2103                         return ret;
2104                 } else if (!ret) {
2105                         /*
2106                          * Success, perform the usual cleanup of a processed
2107                          * head
2108                          */
2109                         ret = cleanup_ref_head(trans, locked_ref, &bytes_processed);
2110                         if (ret > 0 ) {
2111                                 /* We dropped our lock, we need to loop. */
2112                                 ret = 0;
2113                                 continue;
2114                         } else if (ret) {
2115                                 return ret;
2116                         }
2117                 }
2118
2119                 /*
2120                  * Either success case or btrfs_run_delayed_refs_for_head
2121                  * returned -EAGAIN, meaning we need to select another head
2122                  */
2123
2124                 locked_ref = NULL;
2125                 cond_resched();
2126         } while ((min_bytes != U64_MAX && bytes_processed < min_bytes) ||
2127                  (max_count > 0 && count < max_count) ||
2128                  locked_ref);
2129
2130         return 0;
2131 }
2132
2133 #ifdef SCRAMBLE_DELAYED_REFS
2134 /*
2135  * Normally delayed refs get processed in ascending bytenr order. This
2136  * correlates in most cases to the order added. To expose dependencies on this
2137  * order, we start to process the tree in the middle instead of the beginning
2138  */
2139 static u64 find_middle(struct rb_root *root)
2140 {
2141         struct rb_node *n = root->rb_node;
2142         struct btrfs_delayed_ref_node *entry;
2143         int alt = 1;
2144         u64 middle;
2145         u64 first = 0, last = 0;
2146
2147         n = rb_first(root);
2148         if (n) {
2149                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2150                 first = entry->bytenr;
2151         }
2152         n = rb_last(root);
2153         if (n) {
2154                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2155                 last = entry->bytenr;
2156         }
2157         n = root->rb_node;
2158
2159         while (n) {
2160                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2161                 WARN_ON(!entry->in_tree);
2162
2163                 middle = entry->bytenr;
2164
2165                 if (alt)
2166                         n = n->rb_left;
2167                 else
2168                         n = n->rb_right;
2169
2170                 alt = 1 - alt;
2171         }
2172         return middle;
2173 }
2174 #endif
2175
2176 /*
2177  * Start processing the delayed reference count updates and extent insertions
2178  * we have queued up so far.
2179  *
2180  * @trans:      Transaction handle.
2181  * @min_bytes:  How many bytes of delayed references to process. After this
2182  *              many bytes we stop processing delayed references if there are
2183  *              any more. If 0 it means to run all existing delayed references,
2184  *              but not new ones added after running all existing ones.
2185  *              Use (u64)-1 (U64_MAX) to run all existing delayed references
2186  *              plus any new ones that are added.
2187  *
2188  * Returns 0 on success or if called with an aborted transaction
2189  * Returns <0 on error and aborts the transaction
2190  */
2191 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, u64 min_bytes)
2192 {
2193         struct btrfs_fs_info *fs_info = trans->fs_info;
2194         struct btrfs_delayed_ref_root *delayed_refs;
2195         int ret;
2196
2197         /* We'll clean this up in btrfs_cleanup_transaction */
2198         if (TRANS_ABORTED(trans))
2199                 return 0;
2200
2201         if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2202                 return 0;
2203
2204         delayed_refs = &trans->transaction->delayed_refs;
2205 again:
2206 #ifdef SCRAMBLE_DELAYED_REFS
2207         delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2208 #endif
2209         ret = __btrfs_run_delayed_refs(trans, min_bytes);
2210         if (ret < 0) {
2211                 btrfs_abort_transaction(trans, ret);
2212                 return ret;
2213         }
2214
2215         if (min_bytes == U64_MAX) {
2216                 btrfs_create_pending_block_groups(trans);
2217
2218                 spin_lock(&delayed_refs->lock);
2219                 if (RB_EMPTY_ROOT(&delayed_refs->href_root.rb_root)) {
2220                         spin_unlock(&delayed_refs->lock);
2221                         return 0;
2222                 }
2223                 spin_unlock(&delayed_refs->lock);
2224
2225                 cond_resched();
2226                 goto again;
2227         }
2228
2229         return 0;
2230 }
2231
2232 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2233                                 struct extent_buffer *eb, u64 flags)
2234 {
2235         struct btrfs_delayed_extent_op *extent_op;
2236         int level = btrfs_header_level(eb);
2237         int ret;
2238
2239         extent_op = btrfs_alloc_delayed_extent_op();
2240         if (!extent_op)
2241                 return -ENOMEM;
2242
2243         extent_op->flags_to_set = flags;
2244         extent_op->update_flags = true;
2245         extent_op->update_key = false;
2246         extent_op->level = level;
2247
2248         ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2249         if (ret)
2250                 btrfs_free_delayed_extent_op(extent_op);
2251         return ret;
2252 }
2253
2254 static noinline int check_delayed_ref(struct btrfs_root *root,
2255                                       struct btrfs_path *path,
2256                                       u64 objectid, u64 offset, u64 bytenr)
2257 {
2258         struct btrfs_delayed_ref_head *head;
2259         struct btrfs_delayed_ref_node *ref;
2260         struct btrfs_delayed_ref_root *delayed_refs;
2261         struct btrfs_transaction *cur_trans;
2262         struct rb_node *node;
2263         int ret = 0;
2264
2265         spin_lock(&root->fs_info->trans_lock);
2266         cur_trans = root->fs_info->running_transaction;
2267         if (cur_trans)
2268                 refcount_inc(&cur_trans->use_count);
2269         spin_unlock(&root->fs_info->trans_lock);
2270         if (!cur_trans)
2271                 return 0;
2272
2273         delayed_refs = &cur_trans->delayed_refs;
2274         spin_lock(&delayed_refs->lock);
2275         head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2276         if (!head) {
2277                 spin_unlock(&delayed_refs->lock);
2278                 btrfs_put_transaction(cur_trans);
2279                 return 0;
2280         }
2281
2282         if (!mutex_trylock(&head->mutex)) {
2283                 if (path->nowait) {
2284                         spin_unlock(&delayed_refs->lock);
2285                         btrfs_put_transaction(cur_trans);
2286                         return -EAGAIN;
2287                 }
2288
2289                 refcount_inc(&head->refs);
2290                 spin_unlock(&delayed_refs->lock);
2291
2292                 btrfs_release_path(path);
2293
2294                 /*
2295                  * Mutex was contended, block until it's released and let
2296                  * caller try again
2297                  */
2298                 mutex_lock(&head->mutex);
2299                 mutex_unlock(&head->mutex);
2300                 btrfs_put_delayed_ref_head(head);
2301                 btrfs_put_transaction(cur_trans);
2302                 return -EAGAIN;
2303         }
2304         spin_unlock(&delayed_refs->lock);
2305
2306         spin_lock(&head->lock);
2307         /*
2308          * XXX: We should replace this with a proper search function in the
2309          * future.
2310          */
2311         for (node = rb_first_cached(&head->ref_tree); node;
2312              node = rb_next(node)) {
2313                 u64 ref_owner;
2314                 u64 ref_offset;
2315
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                 ref_owner = btrfs_delayed_ref_owner(ref);
2324                 ref_offset = btrfs_delayed_ref_offset(ref);
2325
2326                 /*
2327                  * If our ref doesn't match the one we're currently looking at
2328                  * then we have a cross reference.
2329                  */
2330                 if (ref->ref_root != btrfs_root_id(root) ||
2331                     ref_owner != objectid || 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                                         bool strict)
2346 {
2347         struct btrfs_fs_info *fs_info = root->fs_info;
2348         struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2349         struct extent_buffer *leaf;
2350         struct btrfs_extent_data_ref *ref;
2351         struct btrfs_extent_inline_ref *iref;
2352         struct btrfs_extent_item *ei;
2353         struct btrfs_key key;
2354         u32 item_size;
2355         u32 expected_size;
2356         int type;
2357         int ret;
2358
2359         key.objectid = bytenr;
2360         key.offset = (u64)-1;
2361         key.type = BTRFS_EXTENT_ITEM_KEY;
2362
2363         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2364         if (ret < 0)
2365                 goto out;
2366         if (ret == 0) {
2367                 /*
2368                  * Key with offset -1 found, there would have to exist an extent
2369                  * item with such offset, but this is out of the valid range.
2370                  */
2371                 ret = -EUCLEAN;
2372                 goto out;
2373         }
2374
2375         ret = -ENOENT;
2376         if (path->slots[0] == 0)
2377                 goto out;
2378
2379         path->slots[0]--;
2380         leaf = path->nodes[0];
2381         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2382
2383         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2384                 goto out;
2385
2386         ret = 1;
2387         item_size = btrfs_item_size(leaf, path->slots[0]);
2388         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2389         expected_size = sizeof(*ei) + btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY);
2390
2391         /* No inline refs; we need to bail before checking for owner ref. */
2392         if (item_size == sizeof(*ei))
2393                 goto out;
2394
2395         /* Check for an owner ref; skip over it to the real inline refs. */
2396         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2397         type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2398         if (btrfs_fs_incompat(fs_info, SIMPLE_QUOTA) && type == BTRFS_EXTENT_OWNER_REF_KEY) {
2399                 expected_size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY);
2400                 iref = (struct btrfs_extent_inline_ref *)(iref + 1);
2401         }
2402
2403         /* If extent item has more than 1 inline ref then it's shared */
2404         if (item_size != expected_size)
2405                 goto out;
2406
2407         /*
2408          * If extent created before last snapshot => it's shared unless the
2409          * snapshot has been deleted. Use the heuristic if strict is false.
2410          */
2411         if (!strict &&
2412             (btrfs_extent_generation(leaf, ei) <=
2413              btrfs_root_last_snapshot(&root->root_item)))
2414                 goto out;
2415
2416         /* If this extent has SHARED_DATA_REF then it's shared */
2417         type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2418         if (type != BTRFS_EXTENT_DATA_REF_KEY)
2419                 goto out;
2420
2421         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2422         if (btrfs_extent_refs(leaf, ei) !=
2423             btrfs_extent_data_ref_count(leaf, ref) ||
2424             btrfs_extent_data_ref_root(leaf, ref) != btrfs_root_id(root) ||
2425             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2426             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2427                 goto out;
2428
2429         ret = 0;
2430 out:
2431         return ret;
2432 }
2433
2434 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2435                           u64 bytenr, bool strict, struct btrfs_path *path)
2436 {
2437         int ret;
2438
2439         do {
2440                 ret = check_committed_ref(root, path, objectid,
2441                                           offset, bytenr, strict);
2442                 if (ret && ret != -ENOENT)
2443                         goto out;
2444
2445                 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2446         } while (ret == -EAGAIN);
2447
2448 out:
2449         btrfs_release_path(path);
2450         if (btrfs_is_data_reloc_root(root))
2451                 WARN_ON(ret > 0);
2452         return ret;
2453 }
2454
2455 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2456                            struct btrfs_root *root,
2457                            struct extent_buffer *buf,
2458                            int full_backref, int inc)
2459 {
2460         struct btrfs_fs_info *fs_info = root->fs_info;
2461         u64 parent;
2462         u64 ref_root;
2463         u32 nritems;
2464         struct btrfs_key key;
2465         struct btrfs_file_extent_item *fi;
2466         bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2467         int i;
2468         int action;
2469         int level;
2470         int ret = 0;
2471
2472         if (btrfs_is_testing(fs_info))
2473                 return 0;
2474
2475         ref_root = btrfs_header_owner(buf);
2476         nritems = btrfs_header_nritems(buf);
2477         level = btrfs_header_level(buf);
2478
2479         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2480                 return 0;
2481
2482         if (full_backref)
2483                 parent = buf->start;
2484         else
2485                 parent = 0;
2486         if (inc)
2487                 action = BTRFS_ADD_DELAYED_REF;
2488         else
2489                 action = BTRFS_DROP_DELAYED_REF;
2490
2491         for (i = 0; i < nritems; i++) {
2492                 struct btrfs_ref ref = {
2493                         .action = action,
2494                         .parent = parent,
2495                         .ref_root = ref_root,
2496                 };
2497
2498                 if (level == 0) {
2499                         btrfs_item_key_to_cpu(buf, &key, i);
2500                         if (key.type != BTRFS_EXTENT_DATA_KEY)
2501                                 continue;
2502                         fi = btrfs_item_ptr(buf, i,
2503                                             struct btrfs_file_extent_item);
2504                         if (btrfs_file_extent_type(buf, fi) ==
2505                             BTRFS_FILE_EXTENT_INLINE)
2506                                 continue;
2507                         ref.bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2508                         if (ref.bytenr == 0)
2509                                 continue;
2510
2511                         ref.num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2512                         ref.owning_root = ref_root;
2513
2514                         key.offset -= btrfs_file_extent_offset(buf, fi);
2515                         btrfs_init_data_ref(&ref, key.objectid, key.offset,
2516                                             btrfs_root_id(root), for_reloc);
2517                         if (inc)
2518                                 ret = btrfs_inc_extent_ref(trans, &ref);
2519                         else
2520                                 ret = btrfs_free_extent(trans, &ref);
2521                         if (ret)
2522                                 goto fail;
2523                 } else {
2524                         /* We don't know the owning_root, leave as 0. */
2525                         ref.bytenr = btrfs_node_blockptr(buf, i);
2526                         ref.num_bytes = fs_info->nodesize;
2527
2528                         btrfs_init_tree_ref(&ref, level - 1,
2529                                             btrfs_root_id(root), for_reloc);
2530                         if (inc)
2531                                 ret = btrfs_inc_extent_ref(trans, &ref);
2532                         else
2533                                 ret = btrfs_free_extent(trans, &ref);
2534                         if (ret)
2535                                 goto fail;
2536                 }
2537         }
2538         return 0;
2539 fail:
2540         return ret;
2541 }
2542
2543 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2544                   struct extent_buffer *buf, int full_backref)
2545 {
2546         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2547 }
2548
2549 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2550                   struct extent_buffer *buf, int full_backref)
2551 {
2552         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2553 }
2554
2555 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2556 {
2557         struct btrfs_fs_info *fs_info = root->fs_info;
2558         u64 flags;
2559         u64 ret;
2560
2561         if (data)
2562                 flags = BTRFS_BLOCK_GROUP_DATA;
2563         else if (root == fs_info->chunk_root)
2564                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2565         else
2566                 flags = BTRFS_BLOCK_GROUP_METADATA;
2567
2568         ret = btrfs_get_alloc_profile(fs_info, flags);
2569         return ret;
2570 }
2571
2572 static u64 first_logical_byte(struct btrfs_fs_info *fs_info)
2573 {
2574         struct rb_node *leftmost;
2575         u64 bytenr = 0;
2576
2577         read_lock(&fs_info->block_group_cache_lock);
2578         /* Get the block group with the lowest logical start address. */
2579         leftmost = rb_first_cached(&fs_info->block_group_cache_tree);
2580         if (leftmost) {
2581                 struct btrfs_block_group *bg;
2582
2583                 bg = rb_entry(leftmost, struct btrfs_block_group, cache_node);
2584                 bytenr = bg->start;
2585         }
2586         read_unlock(&fs_info->block_group_cache_lock);
2587
2588         return bytenr;
2589 }
2590
2591 static int pin_down_extent(struct btrfs_trans_handle *trans,
2592                            struct btrfs_block_group *cache,
2593                            u64 bytenr, u64 num_bytes, int reserved)
2594 {
2595         struct btrfs_fs_info *fs_info = cache->fs_info;
2596
2597         spin_lock(&cache->space_info->lock);
2598         spin_lock(&cache->lock);
2599         cache->pinned += num_bytes;
2600         btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2601                                              num_bytes);
2602         if (reserved) {
2603                 cache->reserved -= num_bytes;
2604                 cache->space_info->bytes_reserved -= num_bytes;
2605         }
2606         spin_unlock(&cache->lock);
2607         spin_unlock(&cache->space_info->lock);
2608
2609         set_extent_bit(&trans->transaction->pinned_extents, bytenr,
2610                        bytenr + num_bytes - 1, EXTENT_DIRTY, NULL);
2611         return 0;
2612 }
2613
2614 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2615                      u64 bytenr, u64 num_bytes, int reserved)
2616 {
2617         struct btrfs_block_group *cache;
2618
2619         cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2620         BUG_ON(!cache); /* Logic error */
2621
2622         pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2623
2624         btrfs_put_block_group(cache);
2625         return 0;
2626 }
2627
2628 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2629                                     const struct extent_buffer *eb)
2630 {
2631         struct btrfs_block_group *cache;
2632         int ret;
2633
2634         cache = btrfs_lookup_block_group(trans->fs_info, eb->start);
2635         if (!cache)
2636                 return -EINVAL;
2637
2638         /*
2639          * Fully cache the free space first so that our pin removes the free space
2640          * from the cache.
2641          */
2642         ret = btrfs_cache_block_group(cache, true);
2643         if (ret)
2644                 goto out;
2645
2646         pin_down_extent(trans, cache, eb->start, eb->len, 0);
2647
2648         /* remove us from the free space cache (if we're there at all) */
2649         ret = btrfs_remove_free_space(cache, eb->start, eb->len);
2650 out:
2651         btrfs_put_block_group(cache);
2652         return ret;
2653 }
2654
2655 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2656                                    u64 start, u64 num_bytes)
2657 {
2658         int ret;
2659         struct btrfs_block_group *block_group;
2660
2661         block_group = btrfs_lookup_block_group(fs_info, start);
2662         if (!block_group)
2663                 return -EINVAL;
2664
2665         ret = btrfs_cache_block_group(block_group, true);
2666         if (ret)
2667                 goto out;
2668
2669         ret = btrfs_remove_free_space(block_group, start, num_bytes);
2670 out:
2671         btrfs_put_block_group(block_group);
2672         return ret;
2673 }
2674
2675 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2676 {
2677         struct btrfs_fs_info *fs_info = eb->fs_info;
2678         struct btrfs_file_extent_item *item;
2679         struct btrfs_key key;
2680         int found_type;
2681         int i;
2682         int ret = 0;
2683
2684         if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2685                 return 0;
2686
2687         for (i = 0; i < btrfs_header_nritems(eb); i++) {
2688                 btrfs_item_key_to_cpu(eb, &key, i);
2689                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2690                         continue;
2691                 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2692                 found_type = btrfs_file_extent_type(eb, item);
2693                 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2694                         continue;
2695                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2696                         continue;
2697                 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2698                 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2699                 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2700                 if (ret)
2701                         break;
2702         }
2703
2704         return ret;
2705 }
2706
2707 static void
2708 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2709 {
2710         atomic_inc(&bg->reservations);
2711 }
2712
2713 /*
2714  * Returns the free cluster for the given space info and sets empty_cluster to
2715  * what it should be based on the mount options.
2716  */
2717 static struct btrfs_free_cluster *
2718 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2719                    struct btrfs_space_info *space_info, u64 *empty_cluster)
2720 {
2721         struct btrfs_free_cluster *ret = NULL;
2722
2723         *empty_cluster = 0;
2724         if (btrfs_mixed_space_info(space_info))
2725                 return ret;
2726
2727         if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2728                 ret = &fs_info->meta_alloc_cluster;
2729                 if (btrfs_test_opt(fs_info, SSD))
2730                         *empty_cluster = SZ_2M;
2731                 else
2732                         *empty_cluster = SZ_64K;
2733         } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2734                    btrfs_test_opt(fs_info, SSD_SPREAD)) {
2735                 *empty_cluster = SZ_2M;
2736                 ret = &fs_info->data_alloc_cluster;
2737         }
2738
2739         return ret;
2740 }
2741
2742 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2743                               u64 start, u64 end,
2744                               const bool return_free_space)
2745 {
2746         struct btrfs_block_group *cache = NULL;
2747         struct btrfs_space_info *space_info;
2748         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2749         struct btrfs_free_cluster *cluster = NULL;
2750         u64 len;
2751         u64 total_unpinned = 0;
2752         u64 empty_cluster = 0;
2753         bool readonly;
2754         int ret = 0;
2755
2756         while (start <= end) {
2757                 readonly = false;
2758                 if (!cache ||
2759                     start >= cache->start + cache->length) {
2760                         if (cache)
2761                                 btrfs_put_block_group(cache);
2762                         total_unpinned = 0;
2763                         cache = btrfs_lookup_block_group(fs_info, start);
2764                         if (cache == NULL) {
2765                                 /* Logic error, something removed the block group. */
2766                                 ret = -EUCLEAN;
2767                                 goto out;
2768                         }
2769
2770                         cluster = fetch_cluster_info(fs_info,
2771                                                      cache->space_info,
2772                                                      &empty_cluster);
2773                         empty_cluster <<= 1;
2774                 }
2775
2776                 len = cache->start + cache->length - start;
2777                 len = min(len, end + 1 - start);
2778
2779                 if (return_free_space)
2780                         btrfs_add_free_space(cache, start, len);
2781
2782                 start += len;
2783                 total_unpinned += len;
2784                 space_info = cache->space_info;
2785
2786                 /*
2787                  * If this space cluster has been marked as fragmented and we've
2788                  * unpinned enough in this block group to potentially allow a
2789                  * cluster to be created inside of it go ahead and clear the
2790                  * fragmented check.
2791                  */
2792                 if (cluster && cluster->fragmented &&
2793                     total_unpinned > empty_cluster) {
2794                         spin_lock(&cluster->lock);
2795                         cluster->fragmented = 0;
2796                         spin_unlock(&cluster->lock);
2797                 }
2798
2799                 spin_lock(&space_info->lock);
2800                 spin_lock(&cache->lock);
2801                 cache->pinned -= len;
2802                 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2803                 space_info->max_extent_size = 0;
2804                 if (cache->ro) {
2805                         space_info->bytes_readonly += len;
2806                         readonly = true;
2807                 } else if (btrfs_is_zoned(fs_info)) {
2808                         /* Need reset before reusing in a zoned block group */
2809                         space_info->bytes_zone_unusable += len;
2810                         readonly = true;
2811                 }
2812                 spin_unlock(&cache->lock);
2813                 if (!readonly && return_free_space &&
2814                     global_rsv->space_info == space_info) {
2815                         spin_lock(&global_rsv->lock);
2816                         if (!global_rsv->full) {
2817                                 u64 to_add = min(len, global_rsv->size -
2818                                                       global_rsv->reserved);
2819
2820                                 global_rsv->reserved += to_add;
2821                                 btrfs_space_info_update_bytes_may_use(fs_info,
2822                                                 space_info, to_add);
2823                                 if (global_rsv->reserved >= global_rsv->size)
2824                                         global_rsv->full = 1;
2825                                 len -= to_add;
2826                         }
2827                         spin_unlock(&global_rsv->lock);
2828                 }
2829                 /* Add to any tickets we may have */
2830                 if (!readonly && return_free_space && len)
2831                         btrfs_try_granting_tickets(fs_info, space_info);
2832                 spin_unlock(&space_info->lock);
2833         }
2834
2835         if (cache)
2836                 btrfs_put_block_group(cache);
2837 out:
2838         return ret;
2839 }
2840
2841 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2842 {
2843         struct btrfs_fs_info *fs_info = trans->fs_info;
2844         struct btrfs_block_group *block_group, *tmp;
2845         struct list_head *deleted_bgs;
2846         struct extent_io_tree *unpin;
2847         u64 start;
2848         u64 end;
2849         int ret;
2850
2851         unpin = &trans->transaction->pinned_extents;
2852
2853         while (!TRANS_ABORTED(trans)) {
2854                 struct extent_state *cached_state = NULL;
2855
2856                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2857                 if (!find_first_extent_bit(unpin, 0, &start, &end,
2858                                            EXTENT_DIRTY, &cached_state)) {
2859                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2860                         break;
2861                 }
2862
2863                 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2864                         ret = btrfs_discard_extent(fs_info, start,
2865                                                    end + 1 - start, NULL);
2866
2867                 clear_extent_dirty(unpin, start, end, &cached_state);
2868                 ret = unpin_extent_range(fs_info, start, end, true);
2869                 BUG_ON(ret);
2870                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2871                 free_extent_state(cached_state);
2872                 cond_resched();
2873         }
2874
2875         if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2876                 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2877                 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2878         }
2879
2880         /*
2881          * Transaction is finished.  We don't need the lock anymore.  We
2882          * do need to clean up the block groups in case of a transaction
2883          * abort.
2884          */
2885         deleted_bgs = &trans->transaction->deleted_bgs;
2886         list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2887                 u64 trimmed = 0;
2888
2889                 ret = -EROFS;
2890                 if (!TRANS_ABORTED(trans))
2891                         ret = btrfs_discard_extent(fs_info,
2892                                                    block_group->start,
2893                                                    block_group->length,
2894                                                    &trimmed);
2895
2896                 list_del_init(&block_group->bg_list);
2897                 btrfs_unfreeze_block_group(block_group);
2898                 btrfs_put_block_group(block_group);
2899
2900                 if (ret) {
2901                         const char *errstr = btrfs_decode_error(ret);
2902                         btrfs_warn(fs_info,
2903                            "discard failed while removing blockgroup: errno=%d %s",
2904                                    ret, errstr);
2905                 }
2906         }
2907
2908         return 0;
2909 }
2910
2911 /*
2912  * Parse an extent item's inline extents looking for a simple quotas owner ref.
2913  *
2914  * @fs_info:    the btrfs_fs_info for this mount
2915  * @leaf:       a leaf in the extent tree containing the extent item
2916  * @slot:       the slot in the leaf where the extent item is found
2917  *
2918  * Returns the objectid of the root that originally allocated the extent item
2919  * if the inline owner ref is expected and present, otherwise 0.
2920  *
2921  * If an extent item has an owner ref item, it will be the first inline ref
2922  * item. Therefore the logic is to check whether there are any inline ref
2923  * items, then check the type of the first one.
2924  */
2925 u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info,
2926                                 struct extent_buffer *leaf, int slot)
2927 {
2928         struct btrfs_extent_item *ei;
2929         struct btrfs_extent_inline_ref *iref;
2930         struct btrfs_extent_owner_ref *oref;
2931         unsigned long ptr;
2932         unsigned long end;
2933         int type;
2934
2935         if (!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA))
2936                 return 0;
2937
2938         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
2939         ptr = (unsigned long)(ei + 1);
2940         end = (unsigned long)ei + btrfs_item_size(leaf, slot);
2941
2942         /* No inline ref items of any kind, can't check type. */
2943         if (ptr == end)
2944                 return 0;
2945
2946         iref = (struct btrfs_extent_inline_ref *)ptr;
2947         type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
2948
2949         /* We found an owner ref, get the root out of it. */
2950         if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
2951                 oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
2952                 return btrfs_extent_owner_ref_root_id(leaf, oref);
2953         }
2954
2955         /* We have inline refs, but not an owner ref. */
2956         return 0;
2957 }
2958
2959 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
2960                                      u64 bytenr, struct btrfs_squota_delta *delta)
2961 {
2962         int ret;
2963         u64 num_bytes = delta->num_bytes;
2964
2965         if (delta->is_data) {
2966                 struct btrfs_root *csum_root;
2967
2968                 csum_root = btrfs_csum_root(trans->fs_info, bytenr);
2969                 ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes);
2970                 if (ret) {
2971                         btrfs_abort_transaction(trans, ret);
2972                         return ret;
2973                 }
2974
2975                 ret = btrfs_delete_raid_extent(trans, bytenr, num_bytes);
2976                 if (ret) {
2977                         btrfs_abort_transaction(trans, ret);
2978                         return ret;
2979                 }
2980         }
2981
2982         ret = btrfs_record_squota_delta(trans->fs_info, delta);
2983         if (ret) {
2984                 btrfs_abort_transaction(trans, ret);
2985                 return ret;
2986         }
2987
2988         ret = add_to_free_space_tree(trans, bytenr, num_bytes);
2989         if (ret) {
2990                 btrfs_abort_transaction(trans, ret);
2991                 return ret;
2992         }
2993
2994         ret = btrfs_update_block_group(trans, bytenr, num_bytes, false);
2995         if (ret)
2996                 btrfs_abort_transaction(trans, ret);
2997
2998         return ret;
2999 }
3000
3001 #define abort_and_dump(trans, path, fmt, args...)       \
3002 ({                                                      \
3003         btrfs_abort_transaction(trans, -EUCLEAN);       \
3004         btrfs_print_leaf(path->nodes[0]);               \
3005         btrfs_crit(trans->fs_info, fmt, ##args);        \
3006 })
3007
3008 /*
3009  * Drop one or more refs of @node.
3010  *
3011  * 1. Locate the extent refs.
3012  *    It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
3013  *    Locate it, then reduce the refs number or remove the ref line completely.
3014  *
3015  * 2. Update the refs count in EXTENT/METADATA_ITEM
3016  *
3017  * Inline backref case:
3018  *
3019  * in extent tree we have:
3020  *
3021  *      item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
3022  *              refs 2 gen 6 flags DATA
3023  *              extent data backref root FS_TREE objectid 258 offset 0 count 1
3024  *              extent data backref root FS_TREE objectid 257 offset 0 count 1
3025  *
3026  * This function gets called with:
3027  *
3028  *    node->bytenr = 13631488
3029  *    node->num_bytes = 1048576
3030  *    root_objectid = FS_TREE
3031  *    owner_objectid = 257
3032  *    owner_offset = 0
3033  *    refs_to_drop = 1
3034  *
3035  * Then we should get some like:
3036  *
3037  *      item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
3038  *              refs 1 gen 6 flags DATA
3039  *              extent data backref root FS_TREE objectid 258 offset 0 count 1
3040  *
3041  * Keyed backref case:
3042  *
3043  * in extent tree we have:
3044  *
3045  *      item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
3046  *              refs 754 gen 6 flags DATA
3047  *      [...]
3048  *      item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
3049  *              extent data backref root FS_TREE objectid 866 offset 0 count 1
3050  *
3051  * This function get called with:
3052  *
3053  *    node->bytenr = 13631488
3054  *    node->num_bytes = 1048576
3055  *    root_objectid = FS_TREE
3056  *    owner_objectid = 866
3057  *    owner_offset = 0
3058  *    refs_to_drop = 1
3059  *
3060  * Then we should get some like:
3061  *
3062  *      item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
3063  *              refs 753 gen 6 flags DATA
3064  *
3065  * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
3066  */
3067 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3068                                struct btrfs_delayed_ref_head *href,
3069                                struct btrfs_delayed_ref_node *node,
3070                                struct btrfs_delayed_extent_op *extent_op)
3071 {
3072         struct btrfs_fs_info *info = trans->fs_info;
3073         struct btrfs_key key;
3074         struct btrfs_path *path;
3075         struct btrfs_root *extent_root;
3076         struct extent_buffer *leaf;
3077         struct btrfs_extent_item *ei;
3078         struct btrfs_extent_inline_ref *iref;
3079         int ret;
3080         int is_data;
3081         int extent_slot = 0;
3082         int found_extent = 0;
3083         int num_to_del = 1;
3084         int refs_to_drop = node->ref_mod;
3085         u32 item_size;
3086         u64 refs;
3087         u64 bytenr = node->bytenr;
3088         u64 num_bytes = node->num_bytes;
3089         u64 owner_objectid = btrfs_delayed_ref_owner(node);
3090         u64 owner_offset = btrfs_delayed_ref_offset(node);
3091         bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
3092         u64 delayed_ref_root = href->owning_root;
3093
3094         extent_root = btrfs_extent_root(info, bytenr);
3095         ASSERT(extent_root);
3096
3097         path = btrfs_alloc_path();
3098         if (!path)
3099                 return -ENOMEM;
3100
3101         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3102
3103         if (!is_data && refs_to_drop != 1) {
3104                 btrfs_crit(info,
3105 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
3106                            node->bytenr, refs_to_drop);
3107                 ret = -EINVAL;
3108                 btrfs_abort_transaction(trans, ret);
3109                 goto out;
3110         }
3111
3112         if (is_data)
3113                 skinny_metadata = false;
3114
3115         ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
3116                                     node->parent, node->ref_root, owner_objectid,
3117                                     owner_offset);
3118         if (ret == 0) {
3119                 /*
3120                  * Either the inline backref or the SHARED_DATA_REF/
3121                  * SHARED_BLOCK_REF is found
3122                  *
3123                  * Here is a quick path to locate EXTENT/METADATA_ITEM.
3124                  * It's possible the EXTENT/METADATA_ITEM is near current slot.
3125                  */
3126                 extent_slot = path->slots[0];
3127                 while (extent_slot >= 0) {
3128                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3129                                               extent_slot);
3130                         if (key.objectid != bytenr)
3131                                 break;
3132                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3133                             key.offset == num_bytes) {
3134                                 found_extent = 1;
3135                                 break;
3136                         }
3137                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
3138                             key.offset == owner_objectid) {
3139                                 found_extent = 1;
3140                                 break;
3141                         }
3142
3143                         /* Quick path didn't find the EXTEMT/METADATA_ITEM */
3144                         if (path->slots[0] - extent_slot > 5)
3145                                 break;
3146                         extent_slot--;
3147                 }
3148
3149                 if (!found_extent) {
3150                         if (iref) {
3151                                 abort_and_dump(trans, path,
3152 "invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref",
3153                                            path->slots[0]);
3154                                 ret = -EUCLEAN;
3155                                 goto out;
3156                         }
3157                         /* Must be SHARED_* item, remove the backref first */
3158                         ret = remove_extent_backref(trans, extent_root, path,
3159                                                     NULL, refs_to_drop, is_data);
3160                         if (ret) {
3161                                 btrfs_abort_transaction(trans, ret);
3162                                 goto out;
3163                         }
3164                         btrfs_release_path(path);
3165
3166                         /* Slow path to locate EXTENT/METADATA_ITEM */
3167                         key.objectid = bytenr;
3168                         key.type = BTRFS_EXTENT_ITEM_KEY;
3169                         key.offset = num_bytes;
3170
3171                         if (!is_data && skinny_metadata) {
3172                                 key.type = BTRFS_METADATA_ITEM_KEY;
3173                                 key.offset = owner_objectid;
3174                         }
3175
3176                         ret = btrfs_search_slot(trans, extent_root,
3177                                                 &key, path, -1, 1);
3178                         if (ret > 0 && skinny_metadata && path->slots[0]) {
3179                                 /*
3180                                  * Couldn't find our skinny metadata item,
3181                                  * see if we have ye olde extent item.
3182                                  */
3183                                 path->slots[0]--;
3184                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
3185                                                       path->slots[0]);
3186                                 if (key.objectid == bytenr &&
3187                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
3188                                     key.offset == num_bytes)
3189                                         ret = 0;
3190                         }
3191
3192                         if (ret > 0 && skinny_metadata) {
3193                                 skinny_metadata = false;
3194                                 key.objectid = bytenr;
3195                                 key.type = BTRFS_EXTENT_ITEM_KEY;
3196                                 key.offset = num_bytes;
3197                                 btrfs_release_path(path);
3198                                 ret = btrfs_search_slot(trans, extent_root,
3199                                                         &key, path, -1, 1);
3200                         }
3201
3202                         if (ret) {
3203                                 if (ret > 0)
3204                                         btrfs_print_leaf(path->nodes[0]);
3205                                 btrfs_err(info,
3206                         "umm, got %d back from search, was looking for %llu, slot %d",
3207                                           ret, bytenr, path->slots[0]);
3208                         }
3209                         if (ret < 0) {
3210                                 btrfs_abort_transaction(trans, ret);
3211                                 goto out;
3212                         }
3213                         extent_slot = path->slots[0];
3214                 }
3215         } else if (WARN_ON(ret == -ENOENT)) {
3216                 abort_and_dump(trans, path,
3217 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d",
3218                                bytenr, node->parent, node->ref_root, owner_objectid,
3219                                owner_offset, path->slots[0]);
3220                 goto out;
3221         } else {
3222                 btrfs_abort_transaction(trans, ret);
3223                 goto out;
3224         }
3225
3226         leaf = path->nodes[0];
3227         item_size = btrfs_item_size(leaf, extent_slot);
3228         if (unlikely(item_size < sizeof(*ei))) {
3229                 ret = -EUCLEAN;
3230                 btrfs_err(trans->fs_info,
3231                           "unexpected extent item size, has %u expect >= %zu",
3232                           item_size, sizeof(*ei));
3233                 btrfs_abort_transaction(trans, ret);
3234                 goto out;
3235         }
3236         ei = btrfs_item_ptr(leaf, extent_slot,
3237                             struct btrfs_extent_item);
3238         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3239             key.type == BTRFS_EXTENT_ITEM_KEY) {
3240                 struct btrfs_tree_block_info *bi;
3241
3242                 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3243                         abort_and_dump(trans, path,
3244 "invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu",
3245                                        key.objectid, key.type, key.offset,
3246                                        path->slots[0], owner_objectid, item_size,
3247                                        sizeof(*ei) + sizeof(*bi));
3248                         ret = -EUCLEAN;
3249                         goto out;
3250                 }
3251                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3252                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3253         }
3254
3255         refs = btrfs_extent_refs(leaf, ei);
3256         if (refs < refs_to_drop) {
3257                 abort_and_dump(trans, path,
3258                 "trying to drop %d refs but we only have %llu for bytenr %llu slot %u",
3259                                refs_to_drop, refs, bytenr, path->slots[0]);
3260                 ret = -EUCLEAN;
3261                 goto out;
3262         }
3263         refs -= refs_to_drop;
3264
3265         if (refs > 0) {
3266                 if (extent_op)
3267                         __run_delayed_extent_op(extent_op, leaf, ei);
3268                 /*
3269                  * In the case of inline back ref, reference count will
3270                  * be updated by remove_extent_backref
3271                  */
3272                 if (iref) {
3273                         if (!found_extent) {
3274                                 abort_and_dump(trans, path,
3275 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u",
3276                                                path->slots[0]);
3277                                 ret = -EUCLEAN;
3278                                 goto out;
3279                         }
3280                 } else {
3281                         btrfs_set_extent_refs(leaf, ei, refs);
3282                         btrfs_mark_buffer_dirty(trans, leaf);
3283                 }
3284                 if (found_extent) {
3285                         ret = remove_extent_backref(trans, extent_root, path,
3286                                                     iref, refs_to_drop, is_data);
3287                         if (ret) {
3288                                 btrfs_abort_transaction(trans, ret);
3289                                 goto out;
3290                         }
3291                 }
3292         } else {
3293                 struct btrfs_squota_delta delta = {
3294                         .root = delayed_ref_root,
3295                         .num_bytes = num_bytes,
3296                         .is_data = is_data,
3297                         .is_inc = false,
3298                         .generation = btrfs_extent_generation(leaf, ei),
3299                 };
3300
3301                 /* In this branch refs == 1 */
3302                 if (found_extent) {
3303                         if (is_data && refs_to_drop !=
3304                             extent_data_ref_count(path, iref)) {
3305                                 abort_and_dump(trans, path,
3306                 "invalid refs_to_drop, current refs %u refs_to_drop %u slot %u",
3307                                                extent_data_ref_count(path, iref),
3308                                                refs_to_drop, path->slots[0]);
3309                                 ret = -EUCLEAN;
3310                                 goto out;
3311                         }
3312                         if (iref) {
3313                                 if (path->slots[0] != extent_slot) {
3314                                         abort_and_dump(trans, path,
3315 "invalid iref, extent item key (%llu %u %llu) slot %u doesn't have wanted iref",
3316                                                        key.objectid, key.type,
3317                                                        key.offset, path->slots[0]);
3318                                         ret = -EUCLEAN;
3319                                         goto out;
3320                                 }
3321                         } else {
3322                                 /*
3323                                  * No inline ref, we must be at SHARED_* item,
3324                                  * And it's single ref, it must be:
3325                                  * |    extent_slot       ||extent_slot + 1|
3326                                  * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3327                                  */
3328                                 if (path->slots[0] != extent_slot + 1) {
3329                                         abort_and_dump(trans, path,
3330         "invalid SHARED_* item slot %u, previous item is not EXTENT/METADATA_ITEM",
3331                                                        path->slots[0]);
3332                                         ret = -EUCLEAN;
3333                                         goto out;
3334                                 }
3335                                 path->slots[0] = extent_slot;
3336                                 num_to_del = 2;
3337                         }
3338                 }
3339                 /*
3340                  * We can't infer the data owner from the delayed ref, so we need
3341                  * to try to get it from the owning ref item.
3342                  *
3343                  * If it is not present, then that extent was not written under
3344                  * simple quotas mode, so we don't need to account for its deletion.
3345                  */
3346                 if (is_data)
3347                         delta.root = btrfs_get_extent_owner_root(trans->fs_info,
3348                                                                  leaf, extent_slot);
3349
3350                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3351                                       num_to_del);
3352                 if (ret) {
3353                         btrfs_abort_transaction(trans, ret);
3354                         goto out;
3355                 }
3356                 btrfs_release_path(path);
3357
3358                 ret = do_free_extent_accounting(trans, bytenr, &delta);
3359         }
3360         btrfs_release_path(path);
3361
3362 out:
3363         btrfs_free_path(path);
3364         return ret;
3365 }
3366
3367 /*
3368  * when we free an block, it is possible (and likely) that we free the last
3369  * delayed ref for that extent as well.  This searches the delayed ref tree for
3370  * a given extent, and if there are no other delayed refs to be processed, it
3371  * removes it from the tree.
3372  */
3373 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3374                                       u64 bytenr)
3375 {
3376         struct btrfs_delayed_ref_head *head;
3377         struct btrfs_delayed_ref_root *delayed_refs;
3378         int ret = 0;
3379
3380         delayed_refs = &trans->transaction->delayed_refs;
3381         spin_lock(&delayed_refs->lock);
3382         head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3383         if (!head)
3384                 goto out_delayed_unlock;
3385
3386         spin_lock(&head->lock);
3387         if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3388                 goto out;
3389
3390         if (cleanup_extent_op(head) != NULL)
3391                 goto out;
3392
3393         /*
3394          * waiting for the lock here would deadlock.  If someone else has it
3395          * locked they are already in the process of dropping it anyway
3396          */
3397         if (!mutex_trylock(&head->mutex))
3398                 goto out;
3399
3400         btrfs_delete_ref_head(delayed_refs, head);
3401         head->processing = false;
3402
3403         spin_unlock(&head->lock);
3404         spin_unlock(&delayed_refs->lock);
3405
3406         BUG_ON(head->extent_op);
3407         if (head->must_insert_reserved)
3408                 ret = 1;
3409
3410         btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3411         mutex_unlock(&head->mutex);
3412         btrfs_put_delayed_ref_head(head);
3413         return ret;
3414 out:
3415         spin_unlock(&head->lock);
3416
3417 out_delayed_unlock:
3418         spin_unlock(&delayed_refs->lock);
3419         return 0;
3420 }
3421
3422 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3423                            u64 root_id,
3424                            struct extent_buffer *buf,
3425                            u64 parent, int last_ref)
3426 {
3427         struct btrfs_fs_info *fs_info = trans->fs_info;
3428         struct btrfs_block_group *bg;
3429         int ret;
3430
3431         if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3432                 struct btrfs_ref generic_ref = {
3433                         .action = BTRFS_DROP_DELAYED_REF,
3434                         .bytenr = buf->start,
3435                         .num_bytes = buf->len,
3436                         .parent = parent,
3437                         .owning_root = btrfs_header_owner(buf),
3438                         .ref_root = root_id,
3439                 };
3440
3441                 /*
3442                  * Assert that the extent buffer is not cleared due to
3443                  * EXTENT_BUFFER_ZONED_ZEROOUT. Please refer
3444                  * btrfs_clear_buffer_dirty() and btree_csum_one_bio() for
3445                  * detail.
3446                  */
3447                 ASSERT(btrfs_header_bytenr(buf) != 0);
3448
3449                 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), 0, false);
3450                 btrfs_ref_tree_mod(fs_info, &generic_ref);
3451                 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3452                 BUG_ON(ret); /* -ENOMEM */
3453         }
3454
3455         if (!last_ref)
3456                 return;
3457
3458         if (btrfs_header_generation(buf) != trans->transid)
3459                 goto out;
3460
3461         if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3462                 ret = check_ref_cleanup(trans, buf->start);
3463                 if (!ret)
3464                         goto out;
3465         }
3466
3467         bg = btrfs_lookup_block_group(fs_info, buf->start);
3468
3469         if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3470                 pin_down_extent(trans, bg, buf->start, buf->len, 1);
3471                 btrfs_put_block_group(bg);
3472                 goto out;
3473         }
3474
3475         /*
3476          * If there are tree mod log users we may have recorded mod log
3477          * operations for this node.  If we re-allocate this node we
3478          * could replay operations on this node that happened when it
3479          * existed in a completely different root.  For example if it
3480          * was part of root A, then was reallocated to root B, and we
3481          * are doing a btrfs_old_search_slot(root b), we could replay
3482          * operations that happened when the block was part of root A,
3483          * giving us an inconsistent view of the btree.
3484          *
3485          * We are safe from races here because at this point no other
3486          * node or root points to this extent buffer, so if after this
3487          * check a new tree mod log user joins we will not have an
3488          * existing log of operations on this node that we have to
3489          * contend with.
3490          */
3491
3492         if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags)
3493                      || btrfs_is_zoned(fs_info)) {
3494                 pin_down_extent(trans, bg, buf->start, buf->len, 1);
3495                 btrfs_put_block_group(bg);
3496                 goto out;
3497         }
3498
3499         WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3500
3501         btrfs_add_free_space(bg, buf->start, buf->len);
3502         btrfs_free_reserved_bytes(bg, buf->len, 0);
3503         btrfs_put_block_group(bg);
3504         trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3505
3506 out:
3507
3508         /*
3509          * Deleting the buffer, clear the corrupt flag since it doesn't
3510          * matter anymore.
3511          */
3512         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3513 }
3514
3515 /* Can return -ENOMEM */
3516 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3517 {
3518         struct btrfs_fs_info *fs_info = trans->fs_info;
3519         int ret;
3520
3521         if (btrfs_is_testing(fs_info))
3522                 return 0;
3523
3524         /*
3525          * tree log blocks never actually go into the extent allocation
3526          * tree, just update pinning info and exit early.
3527          */
3528         if (ref->ref_root == BTRFS_TREE_LOG_OBJECTID) {
3529                 btrfs_pin_extent(trans, ref->bytenr, ref->num_bytes, 1);
3530                 ret = 0;
3531         } else if (ref->type == BTRFS_REF_METADATA) {
3532                 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3533         } else {
3534                 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3535         }
3536
3537         if (ref->ref_root != BTRFS_TREE_LOG_OBJECTID)
3538                 btrfs_ref_tree_mod(fs_info, ref);
3539
3540         return ret;
3541 }
3542
3543 enum btrfs_loop_type {
3544         /*
3545          * Start caching block groups but do not wait for progress or for them
3546          * to be done.
3547          */
3548         LOOP_CACHING_NOWAIT,
3549
3550         /*
3551          * Wait for the block group free_space >= the space we're waiting for if
3552          * the block group isn't cached.
3553          */
3554         LOOP_CACHING_WAIT,
3555
3556         /*
3557          * Allow allocations to happen from block groups that do not yet have a
3558          * size classification.
3559          */
3560         LOOP_UNSET_SIZE_CLASS,
3561
3562         /*
3563          * Allocate a chunk and then retry the allocation.
3564          */
3565         LOOP_ALLOC_CHUNK,
3566
3567         /*
3568          * Ignore the size class restrictions for this allocation.
3569          */
3570         LOOP_WRONG_SIZE_CLASS,
3571
3572         /*
3573          * Ignore the empty size, only try to allocate the number of bytes
3574          * needed for this allocation.
3575          */
3576         LOOP_NO_EMPTY_SIZE,
3577 };
3578
3579 static inline void
3580 btrfs_lock_block_group(struct btrfs_block_group *cache,
3581                        int delalloc)
3582 {
3583         if (delalloc)
3584                 down_read(&cache->data_rwsem);
3585 }
3586
3587 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3588                        int delalloc)
3589 {
3590         btrfs_get_block_group(cache);
3591         if (delalloc)
3592                 down_read(&cache->data_rwsem);
3593 }
3594
3595 static struct btrfs_block_group *btrfs_lock_cluster(
3596                    struct btrfs_block_group *block_group,
3597                    struct btrfs_free_cluster *cluster,
3598                    int delalloc)
3599         __acquires(&cluster->refill_lock)
3600 {
3601         struct btrfs_block_group *used_bg = NULL;
3602
3603         spin_lock(&cluster->refill_lock);
3604         while (1) {
3605                 used_bg = cluster->block_group;
3606                 if (!used_bg)
3607                         return NULL;
3608
3609                 if (used_bg == block_group)
3610                         return used_bg;
3611
3612                 btrfs_get_block_group(used_bg);
3613
3614                 if (!delalloc)
3615                         return used_bg;
3616
3617                 if (down_read_trylock(&used_bg->data_rwsem))
3618                         return used_bg;
3619
3620                 spin_unlock(&cluster->refill_lock);
3621
3622                 /* We should only have one-level nested. */
3623                 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3624
3625                 spin_lock(&cluster->refill_lock);
3626                 if (used_bg == cluster->block_group)
3627                         return used_bg;
3628
3629                 up_read(&used_bg->data_rwsem);
3630                 btrfs_put_block_group(used_bg);
3631         }
3632 }
3633
3634 static inline void
3635 btrfs_release_block_group(struct btrfs_block_group *cache,
3636                          int delalloc)
3637 {
3638         if (delalloc)
3639                 up_read(&cache->data_rwsem);
3640         btrfs_put_block_group(cache);
3641 }
3642
3643 /*
3644  * Helper function for find_free_extent().
3645  *
3646  * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3647  * Return >0 to inform caller that we find nothing
3648  * Return 0 means we have found a location and set ffe_ctl->found_offset.
3649  */
3650 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3651                                       struct find_free_extent_ctl *ffe_ctl,
3652                                       struct btrfs_block_group **cluster_bg_ret)
3653 {
3654         struct btrfs_block_group *cluster_bg;
3655         struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3656         u64 aligned_cluster;
3657         u64 offset;
3658         int ret;
3659
3660         cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3661         if (!cluster_bg)
3662                 goto refill_cluster;
3663         if (cluster_bg != bg && (cluster_bg->ro ||
3664             !block_group_bits(cluster_bg, ffe_ctl->flags)))
3665                 goto release_cluster;
3666
3667         offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3668                         ffe_ctl->num_bytes, cluster_bg->start,
3669                         &ffe_ctl->max_extent_size);
3670         if (offset) {
3671                 /* We have a block, we're done */
3672                 spin_unlock(&last_ptr->refill_lock);
3673                 trace_btrfs_reserve_extent_cluster(cluster_bg, ffe_ctl);
3674                 *cluster_bg_ret = cluster_bg;
3675                 ffe_ctl->found_offset = offset;
3676                 return 0;
3677         }
3678         WARN_ON(last_ptr->block_group != cluster_bg);
3679
3680 release_cluster:
3681         /*
3682          * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3683          * lets just skip it and let the allocator find whatever block it can
3684          * find. If we reach this point, we will have tried the cluster
3685          * allocator plenty of times and not have found anything, so we are
3686          * likely way too fragmented for the clustering stuff to find anything.
3687          *
3688          * However, if the cluster is taken from the current block group,
3689          * release the cluster first, so that we stand a better chance of
3690          * succeeding in the unclustered allocation.
3691          */
3692         if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3693                 spin_unlock(&last_ptr->refill_lock);
3694                 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3695                 return -ENOENT;
3696         }
3697
3698         /* This cluster didn't work out, free it and start over */
3699         btrfs_return_cluster_to_free_space(NULL, last_ptr);
3700
3701         if (cluster_bg != bg)
3702                 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3703
3704 refill_cluster:
3705         if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3706                 spin_unlock(&last_ptr->refill_lock);
3707                 return -ENOENT;
3708         }
3709
3710         aligned_cluster = max_t(u64,
3711                         ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3712                         bg->full_stripe_len);
3713         ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3714                         ffe_ctl->num_bytes, aligned_cluster);
3715         if (ret == 0) {
3716                 /* Now pull our allocation out of this cluster */
3717                 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3718                                 ffe_ctl->num_bytes, ffe_ctl->search_start,
3719                                 &ffe_ctl->max_extent_size);
3720                 if (offset) {
3721                         /* We found one, proceed */
3722                         spin_unlock(&last_ptr->refill_lock);
3723                         ffe_ctl->found_offset = offset;
3724                         trace_btrfs_reserve_extent_cluster(bg, ffe_ctl);
3725                         return 0;
3726                 }
3727         }
3728         /*
3729          * At this point we either didn't find a cluster or we weren't able to
3730          * allocate a block from our cluster.  Free the cluster we've been
3731          * trying to use, and go to the next block group.
3732          */
3733         btrfs_return_cluster_to_free_space(NULL, last_ptr);
3734         spin_unlock(&last_ptr->refill_lock);
3735         return 1;
3736 }
3737
3738 /*
3739  * Return >0 to inform caller that we find nothing
3740  * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3741  */
3742 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3743                                         struct find_free_extent_ctl *ffe_ctl)
3744 {
3745         struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3746         u64 offset;
3747
3748         /*
3749          * We are doing an unclustered allocation, set the fragmented flag so
3750          * we don't bother trying to setup a cluster again until we get more
3751          * space.
3752          */
3753         if (unlikely(last_ptr)) {
3754                 spin_lock(&last_ptr->lock);
3755                 last_ptr->fragmented = 1;
3756                 spin_unlock(&last_ptr->lock);
3757         }
3758         if (ffe_ctl->cached) {
3759                 struct btrfs_free_space_ctl *free_space_ctl;
3760
3761                 free_space_ctl = bg->free_space_ctl;
3762                 spin_lock(&free_space_ctl->tree_lock);
3763                 if (free_space_ctl->free_space <
3764                     ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3765                     ffe_ctl->empty_size) {
3766                         ffe_ctl->total_free_space = max_t(u64,
3767                                         ffe_ctl->total_free_space,
3768                                         free_space_ctl->free_space);
3769                         spin_unlock(&free_space_ctl->tree_lock);
3770                         return 1;
3771                 }
3772                 spin_unlock(&free_space_ctl->tree_lock);
3773         }
3774
3775         offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3776                         ffe_ctl->num_bytes, ffe_ctl->empty_size,
3777                         &ffe_ctl->max_extent_size);
3778         if (!offset)
3779                 return 1;
3780         ffe_ctl->found_offset = offset;
3781         return 0;
3782 }
3783
3784 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3785                                    struct find_free_extent_ctl *ffe_ctl,
3786                                    struct btrfs_block_group **bg_ret)
3787 {
3788         int ret;
3789
3790         /* We want to try and use the cluster allocator, so lets look there */
3791         if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3792                 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3793                 if (ret >= 0)
3794                         return ret;
3795                 /* ret == -ENOENT case falls through */
3796         }
3797
3798         return find_free_extent_unclustered(block_group, ffe_ctl);
3799 }
3800
3801 /*
3802  * Tree-log block group locking
3803  * ============================
3804  *
3805  * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3806  * indicates the starting address of a block group, which is reserved only
3807  * for tree-log metadata.
3808  *
3809  * Lock nesting
3810  * ============
3811  *
3812  * space_info::lock
3813  *   block_group::lock
3814  *     fs_info::treelog_bg_lock
3815  */
3816
3817 /*
3818  * Simple allocator for sequential-only block group. It only allows sequential
3819  * allocation. No need to play with trees. This function also reserves the
3820  * bytes as in btrfs_add_reserved_bytes.
3821  */
3822 static int do_allocation_zoned(struct btrfs_block_group *block_group,
3823                                struct find_free_extent_ctl *ffe_ctl,
3824                                struct btrfs_block_group **bg_ret)
3825 {
3826         struct btrfs_fs_info *fs_info = block_group->fs_info;
3827         struct btrfs_space_info *space_info = block_group->space_info;
3828         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3829         u64 start = block_group->start;
3830         u64 num_bytes = ffe_ctl->num_bytes;
3831         u64 avail;
3832         u64 bytenr = block_group->start;
3833         u64 log_bytenr;
3834         u64 data_reloc_bytenr;
3835         int ret = 0;
3836         bool skip = false;
3837
3838         ASSERT(btrfs_is_zoned(block_group->fs_info));
3839
3840         /*
3841          * Do not allow non-tree-log blocks in the dedicated tree-log block
3842          * group, and vice versa.
3843          */
3844         spin_lock(&fs_info->treelog_bg_lock);
3845         log_bytenr = fs_info->treelog_bg;
3846         if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3847                            (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3848                 skip = true;
3849         spin_unlock(&fs_info->treelog_bg_lock);
3850         if (skip)
3851                 return 1;
3852
3853         /*
3854          * Do not allow non-relocation blocks in the dedicated relocation block
3855          * group, and vice versa.
3856          */
3857         spin_lock(&fs_info->relocation_bg_lock);
3858         data_reloc_bytenr = fs_info->data_reloc_bg;
3859         if (data_reloc_bytenr &&
3860             ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3861              (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3862                 skip = true;
3863         spin_unlock(&fs_info->relocation_bg_lock);
3864         if (skip)
3865                 return 1;
3866
3867         /* Check RO and no space case before trying to activate it */
3868         spin_lock(&block_group->lock);
3869         if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) {
3870                 ret = 1;
3871                 /*
3872                  * May need to clear fs_info->{treelog,data_reloc}_bg.
3873                  * Return the error after taking the locks.
3874                  */
3875         }
3876         spin_unlock(&block_group->lock);
3877
3878         /* Metadata block group is activated at write time. */
3879         if (!ret && (block_group->flags & BTRFS_BLOCK_GROUP_DATA) &&
3880             !btrfs_zone_activate(block_group)) {
3881                 ret = 1;
3882                 /*
3883                  * May need to clear fs_info->{treelog,data_reloc}_bg.
3884                  * Return the error after taking the locks.
3885                  */
3886         }
3887
3888         spin_lock(&space_info->lock);
3889         spin_lock(&block_group->lock);
3890         spin_lock(&fs_info->treelog_bg_lock);
3891         spin_lock(&fs_info->relocation_bg_lock);
3892
3893         if (ret)
3894                 goto out;
3895
3896         ASSERT(!ffe_ctl->for_treelog ||
3897                block_group->start == fs_info->treelog_bg ||
3898                fs_info->treelog_bg == 0);
3899         ASSERT(!ffe_ctl->for_data_reloc ||
3900                block_group->start == fs_info->data_reloc_bg ||
3901                fs_info->data_reloc_bg == 0);
3902
3903         if (block_group->ro ||
3904             (!ffe_ctl->for_data_reloc &&
3905              test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))) {
3906                 ret = 1;
3907                 goto out;
3908         }
3909
3910         /*
3911          * Do not allow currently using block group to be tree-log dedicated
3912          * block group.
3913          */
3914         if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3915             (block_group->used || block_group->reserved)) {
3916                 ret = 1;
3917                 goto out;
3918         }
3919
3920         /*
3921          * Do not allow currently used block group to be the data relocation
3922          * dedicated block group.
3923          */
3924         if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3925             (block_group->used || block_group->reserved)) {
3926                 ret = 1;
3927                 goto out;
3928         }
3929
3930         WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3931         avail = block_group->zone_capacity - block_group->alloc_offset;
3932         if (avail < num_bytes) {
3933                 if (ffe_ctl->max_extent_size < avail) {
3934                         /*
3935                          * With sequential allocator, free space is always
3936                          * contiguous
3937                          */
3938                         ffe_ctl->max_extent_size = avail;
3939                         ffe_ctl->total_free_space = avail;
3940                 }
3941                 ret = 1;
3942                 goto out;
3943         }
3944
3945         if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3946                 fs_info->treelog_bg = block_group->start;
3947
3948         if (ffe_ctl->for_data_reloc) {
3949                 if (!fs_info->data_reloc_bg)
3950                         fs_info->data_reloc_bg = block_group->start;
3951                 /*
3952                  * Do not allow allocations from this block group, unless it is
3953                  * for data relocation. Compared to increasing the ->ro, setting
3954                  * the ->zoned_data_reloc_ongoing flag still allows nocow
3955                  * writers to come in. See btrfs_inc_nocow_writers().
3956                  *
3957                  * We need to disable an allocation to avoid an allocation of
3958                  * regular (non-relocation data) extent. With mix of relocation
3959                  * extents and regular extents, we can dispatch WRITE commands
3960                  * (for relocation extents) and ZONE APPEND commands (for
3961                  * regular extents) at the same time to the same zone, which
3962                  * easily break the write pointer.
3963                  *
3964                  * Also, this flag avoids this block group to be zone finished.
3965                  */
3966                 set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags);
3967         }
3968
3969         ffe_ctl->found_offset = start + block_group->alloc_offset;
3970         block_group->alloc_offset += num_bytes;
3971         spin_lock(&ctl->tree_lock);
3972         ctl->free_space -= num_bytes;
3973         spin_unlock(&ctl->tree_lock);
3974
3975         /*
3976          * We do not check if found_offset is aligned to stripesize. The
3977          * address is anyway rewritten when using zone append writing.
3978          */
3979
3980         ffe_ctl->search_start = ffe_ctl->found_offset;
3981
3982 out:
3983         if (ret && ffe_ctl->for_treelog)
3984                 fs_info->treelog_bg = 0;
3985         if (ret && ffe_ctl->for_data_reloc)
3986                 fs_info->data_reloc_bg = 0;
3987         spin_unlock(&fs_info->relocation_bg_lock);
3988         spin_unlock(&fs_info->treelog_bg_lock);
3989         spin_unlock(&block_group->lock);
3990         spin_unlock(&space_info->lock);
3991         return ret;
3992 }
3993
3994 static int do_allocation(struct btrfs_block_group *block_group,
3995                          struct find_free_extent_ctl *ffe_ctl,
3996                          struct btrfs_block_group **bg_ret)
3997 {
3998         switch (ffe_ctl->policy) {
3999         case BTRFS_EXTENT_ALLOC_CLUSTERED:
4000                 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
4001         case BTRFS_EXTENT_ALLOC_ZONED:
4002                 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
4003         default:
4004                 BUG();
4005         }
4006 }
4007
4008 static void release_block_group(struct btrfs_block_group *block_group,
4009                                 struct find_free_extent_ctl *ffe_ctl,
4010                                 int delalloc)
4011 {
4012         switch (ffe_ctl->policy) {
4013         case BTRFS_EXTENT_ALLOC_CLUSTERED:
4014                 ffe_ctl->retry_uncached = false;
4015                 break;
4016         case BTRFS_EXTENT_ALLOC_ZONED:
4017                 /* Nothing to do */
4018                 break;
4019         default:
4020                 BUG();
4021         }
4022
4023         BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
4024                ffe_ctl->index);
4025         btrfs_release_block_group(block_group, delalloc);
4026 }
4027
4028 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
4029                                    struct btrfs_key *ins)
4030 {
4031         struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4032
4033         if (!ffe_ctl->use_cluster && last_ptr) {
4034                 spin_lock(&last_ptr->lock);
4035                 last_ptr->window_start = ins->objectid;
4036                 spin_unlock(&last_ptr->lock);
4037         }
4038 }
4039
4040 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
4041                          struct btrfs_key *ins)
4042 {
4043         switch (ffe_ctl->policy) {
4044         case BTRFS_EXTENT_ALLOC_CLUSTERED:
4045                 found_extent_clustered(ffe_ctl, ins);
4046                 break;
4047         case BTRFS_EXTENT_ALLOC_ZONED:
4048                 /* Nothing to do */
4049                 break;
4050         default:
4051                 BUG();
4052         }
4053 }
4054
4055 static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info,
4056                                     struct find_free_extent_ctl *ffe_ctl)
4057 {
4058         /* Block group's activeness is not a requirement for METADATA block groups. */
4059         if (!(ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA))
4060                 return 0;
4061
4062         /* If we can activate new zone, just allocate a chunk and use it */
4063         if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags))
4064                 return 0;
4065
4066         /*
4067          * We already reached the max active zones. Try to finish one block
4068          * group to make a room for a new block group. This is only possible
4069          * for a data block group because btrfs_zone_finish() may need to wait
4070          * for a running transaction which can cause a deadlock for metadata
4071          * allocation.
4072          */
4073         if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
4074                 int ret = btrfs_zone_finish_one_bg(fs_info);
4075
4076                 if (ret == 1)
4077                         return 0;
4078                 else if (ret < 0)
4079                         return ret;
4080         }
4081
4082         /*
4083          * If we have enough free space left in an already active block group
4084          * and we can't activate any other zone now, do not allow allocating a
4085          * new chunk and let find_free_extent() retry with a smaller size.
4086          */
4087         if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size)
4088                 return -ENOSPC;
4089
4090         /*
4091          * Even min_alloc_size is not left in any block groups. Since we cannot
4092          * activate a new block group, allocating it may not help. Let's tell a
4093          * caller to try again and hope it progress something by writing some
4094          * parts of the region. That is only possible for data block groups,
4095          * where a part of the region can be written.
4096          */
4097         if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)
4098                 return -EAGAIN;
4099
4100         /*
4101          * We cannot activate a new block group and no enough space left in any
4102          * block groups. So, allocating a new block group may not help. But,
4103          * there is nothing to do anyway, so let's go with it.
4104          */
4105         return 0;
4106 }
4107
4108 static int can_allocate_chunk(struct btrfs_fs_info *fs_info,
4109                               struct find_free_extent_ctl *ffe_ctl)
4110 {
4111         switch (ffe_ctl->policy) {
4112         case BTRFS_EXTENT_ALLOC_CLUSTERED:
4113                 return 0;
4114         case BTRFS_EXTENT_ALLOC_ZONED:
4115                 return can_allocate_chunk_zoned(fs_info, ffe_ctl);
4116         default:
4117                 BUG();
4118         }
4119 }
4120
4121 /*
4122  * Return >0 means caller needs to re-search for free extent
4123  * Return 0 means we have the needed free extent.
4124  * Return <0 means we failed to locate any free extent.
4125  */
4126 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
4127                                         struct btrfs_key *ins,
4128                                         struct find_free_extent_ctl *ffe_ctl,
4129                                         bool full_search)
4130 {
4131         struct btrfs_root *root = fs_info->chunk_root;
4132         int ret;
4133
4134         if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
4135             ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
4136                 ffe_ctl->orig_have_caching_bg = true;
4137
4138         if (ins->objectid) {
4139                 found_extent(ffe_ctl, ins);
4140                 return 0;
4141         }
4142
4143         if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
4144                 return 1;
4145
4146         ffe_ctl->index++;
4147         if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
4148                 return 1;
4149
4150         /* See the comments for btrfs_loop_type for an explanation of the phases. */
4151         if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
4152                 ffe_ctl->index = 0;
4153                 /*
4154                  * We want to skip the LOOP_CACHING_WAIT step if we don't have
4155                  * any uncached bgs and we've already done a full search
4156                  * through.
4157                  */
4158                 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT &&
4159                     (!ffe_ctl->orig_have_caching_bg && full_search))
4160                         ffe_ctl->loop++;
4161                 ffe_ctl->loop++;
4162
4163                 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
4164                         struct btrfs_trans_handle *trans;
4165                         int exist = 0;
4166
4167                         /* Check if allocation policy allows to create a new chunk */
4168                         ret = can_allocate_chunk(fs_info, ffe_ctl);
4169                         if (ret)
4170                                 return ret;
4171
4172                         trans = current->journal_info;
4173                         if (trans)
4174                                 exist = 1;
4175                         else
4176                                 trans = btrfs_join_transaction(root);
4177
4178                         if (IS_ERR(trans)) {
4179                                 ret = PTR_ERR(trans);
4180                                 return ret;
4181                         }
4182
4183                         ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
4184                                                 CHUNK_ALLOC_FORCE_FOR_EXTENT);
4185
4186                         /* Do not bail out on ENOSPC since we can do more. */
4187                         if (ret == -ENOSPC) {
4188                                 ret = 0;
4189                                 ffe_ctl->loop++;
4190                         }
4191                         else if (ret < 0)
4192                                 btrfs_abort_transaction(trans, ret);
4193                         else
4194                                 ret = 0;
4195                         if (!exist)
4196                                 btrfs_end_transaction(trans);
4197                         if (ret)
4198                                 return ret;
4199                 }
4200
4201                 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4202                         if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4203                                 return -ENOSPC;
4204
4205                         /*
4206                          * Don't loop again if we already have no empty_size and
4207                          * no empty_cluster.
4208                          */
4209                         if (ffe_ctl->empty_size == 0 &&
4210                             ffe_ctl->empty_cluster == 0)
4211                                 return -ENOSPC;
4212                         ffe_ctl->empty_size = 0;
4213                         ffe_ctl->empty_cluster = 0;
4214                 }
4215                 return 1;
4216         }
4217         return -ENOSPC;
4218 }
4219
4220 static bool find_free_extent_check_size_class(struct find_free_extent_ctl *ffe_ctl,
4221                                               struct btrfs_block_group *bg)
4222 {
4223         if (ffe_ctl->policy == BTRFS_EXTENT_ALLOC_ZONED)
4224                 return true;
4225         if (!btrfs_block_group_should_use_size_class(bg))
4226                 return true;
4227         if (ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS)
4228                 return true;
4229         if (ffe_ctl->loop >= LOOP_UNSET_SIZE_CLASS &&
4230             bg->size_class == BTRFS_BG_SZ_NONE)
4231                 return true;
4232         return ffe_ctl->size_class == bg->size_class;
4233 }
4234
4235 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4236                                         struct find_free_extent_ctl *ffe_ctl,
4237                                         struct btrfs_space_info *space_info,
4238                                         struct btrfs_key *ins)
4239 {
4240         /*
4241          * If our free space is heavily fragmented we may not be able to make
4242          * big contiguous allocations, so instead of doing the expensive search
4243          * for free space, simply return ENOSPC with our max_extent_size so we
4244          * can go ahead and search for a more manageable chunk.
4245          *
4246          * If our max_extent_size is large enough for our allocation simply
4247          * disable clustering since we will likely not be able to find enough
4248          * space to create a cluster and induce latency trying.
4249          */
4250         if (space_info->max_extent_size) {
4251                 spin_lock(&space_info->lock);
4252                 if (space_info->max_extent_size &&
4253                     ffe_ctl->num_bytes > space_info->max_extent_size) {
4254                         ins->offset = space_info->max_extent_size;
4255                         spin_unlock(&space_info->lock);
4256                         return -ENOSPC;
4257                 } else if (space_info->max_extent_size) {
4258                         ffe_ctl->use_cluster = false;
4259                 }
4260                 spin_unlock(&space_info->lock);
4261         }
4262
4263         ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4264                                                &ffe_ctl->empty_cluster);
4265         if (ffe_ctl->last_ptr) {
4266                 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4267
4268                 spin_lock(&last_ptr->lock);
4269                 if (last_ptr->block_group)
4270                         ffe_ctl->hint_byte = last_ptr->window_start;
4271                 if (last_ptr->fragmented) {
4272                         /*
4273                          * We still set window_start so we can keep track of the
4274                          * last place we found an allocation to try and save
4275                          * some time.
4276                          */
4277                         ffe_ctl->hint_byte = last_ptr->window_start;
4278                         ffe_ctl->use_cluster = false;
4279                 }
4280                 spin_unlock(&last_ptr->lock);
4281         }
4282
4283         return 0;
4284 }
4285
4286 static int prepare_allocation_zoned(struct btrfs_fs_info *fs_info,
4287                                     struct find_free_extent_ctl *ffe_ctl)
4288 {
4289         if (ffe_ctl->for_treelog) {
4290                 spin_lock(&fs_info->treelog_bg_lock);
4291                 if (fs_info->treelog_bg)
4292                         ffe_ctl->hint_byte = fs_info->treelog_bg;
4293                 spin_unlock(&fs_info->treelog_bg_lock);
4294         } else if (ffe_ctl->for_data_reloc) {
4295                 spin_lock(&fs_info->relocation_bg_lock);
4296                 if (fs_info->data_reloc_bg)
4297                         ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4298                 spin_unlock(&fs_info->relocation_bg_lock);
4299         } else if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
4300                 struct btrfs_block_group *block_group;
4301
4302                 spin_lock(&fs_info->zone_active_bgs_lock);
4303                 list_for_each_entry(block_group, &fs_info->zone_active_bgs, active_bg_list) {
4304                         /*
4305                          * No lock is OK here because avail is monotinically
4306                          * decreasing, and this is just a hint.
4307                          */
4308                         u64 avail = block_group->zone_capacity - block_group->alloc_offset;
4309
4310                         if (block_group_bits(block_group, ffe_ctl->flags) &&
4311                             avail >= ffe_ctl->num_bytes) {
4312                                 ffe_ctl->hint_byte = block_group->start;
4313                                 break;
4314                         }
4315                 }
4316                 spin_unlock(&fs_info->zone_active_bgs_lock);
4317         }
4318
4319         return 0;
4320 }
4321
4322 static int prepare_allocation(struct btrfs_fs_info *fs_info,
4323                               struct find_free_extent_ctl *ffe_ctl,
4324                               struct btrfs_space_info *space_info,
4325                               struct btrfs_key *ins)
4326 {
4327         switch (ffe_ctl->policy) {
4328         case BTRFS_EXTENT_ALLOC_CLUSTERED:
4329                 return prepare_allocation_clustered(fs_info, ffe_ctl,
4330                                                     space_info, ins);
4331         case BTRFS_EXTENT_ALLOC_ZONED:
4332                 return prepare_allocation_zoned(fs_info, ffe_ctl);
4333         default:
4334                 BUG();
4335         }
4336 }
4337
4338 /*
4339  * walks the btree of allocated extents and find a hole of a given size.
4340  * The key ins is changed to record the hole:
4341  * ins->objectid == start position
4342  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4343  * ins->offset == the size of the hole.
4344  * Any available blocks before search_start are skipped.
4345  *
4346  * If there is no suitable free space, we will record the max size of
4347  * the free space extent currently.
4348  *
4349  * The overall logic and call chain:
4350  *
4351  * find_free_extent()
4352  * |- Iterate through all block groups
4353  * |  |- Get a valid block group
4354  * |  |- Try to do clustered allocation in that block group
4355  * |  |- Try to do unclustered allocation in that block group
4356  * |  |- Check if the result is valid
4357  * |  |  |- If valid, then exit
4358  * |  |- Jump to next block group
4359  * |
4360  * |- Push harder to find free extents
4361  *    |- If not found, re-iterate all block groups
4362  */
4363 static noinline int find_free_extent(struct btrfs_root *root,
4364                                      struct btrfs_key *ins,
4365                                      struct find_free_extent_ctl *ffe_ctl)
4366 {
4367         struct btrfs_fs_info *fs_info = root->fs_info;
4368         int ret = 0;
4369         int cache_block_group_error = 0;
4370         struct btrfs_block_group *block_group = NULL;
4371         struct btrfs_space_info *space_info;
4372         bool full_search = false;
4373
4374         WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4375
4376         ffe_ctl->search_start = 0;
4377         /* For clustered allocation */
4378         ffe_ctl->empty_cluster = 0;
4379         ffe_ctl->last_ptr = NULL;
4380         ffe_ctl->use_cluster = true;
4381         ffe_ctl->have_caching_bg = false;
4382         ffe_ctl->orig_have_caching_bg = false;
4383         ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4384         ffe_ctl->loop = 0;
4385         ffe_ctl->retry_uncached = false;
4386         ffe_ctl->cached = 0;
4387         ffe_ctl->max_extent_size = 0;
4388         ffe_ctl->total_free_space = 0;
4389         ffe_ctl->found_offset = 0;
4390         ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4391         ffe_ctl->size_class = btrfs_calc_block_group_size_class(ffe_ctl->num_bytes);
4392
4393         if (btrfs_is_zoned(fs_info))
4394                 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4395
4396         ins->type = BTRFS_EXTENT_ITEM_KEY;
4397         ins->objectid = 0;
4398         ins->offset = 0;
4399
4400         trace_find_free_extent(root, ffe_ctl);
4401
4402         space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4403         if (!space_info) {
4404                 btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4405                 return -ENOSPC;
4406         }
4407
4408         ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4409         if (ret < 0)
4410                 return ret;
4411
4412         ffe_ctl->search_start = max(ffe_ctl->search_start,
4413                                     first_logical_byte(fs_info));
4414         ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4415         if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4416                 block_group = btrfs_lookup_block_group(fs_info,
4417                                                        ffe_ctl->search_start);
4418                 /*
4419                  * we don't want to use the block group if it doesn't match our
4420                  * allocation bits, or if its not cached.
4421                  *
4422                  * However if we are re-searching with an ideal block group
4423                  * picked out then we don't care that the block group is cached.
4424                  */
4425                 if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4426                     block_group->cached != BTRFS_CACHE_NO) {
4427                         down_read(&space_info->groups_sem);
4428                         if (list_empty(&block_group->list) ||
4429                             block_group->ro) {
4430                                 /*
4431                                  * someone is removing this block group,
4432                                  * we can't jump into the have_block_group
4433                                  * target because our list pointers are not
4434                                  * valid
4435                                  */
4436                                 btrfs_put_block_group(block_group);
4437                                 up_read(&space_info->groups_sem);
4438                         } else {
4439                                 ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4440                                                         block_group->flags);
4441                                 btrfs_lock_block_group(block_group,
4442                                                        ffe_ctl->delalloc);
4443                                 ffe_ctl->hinted = true;
4444                                 goto have_block_group;
4445                         }
4446                 } else if (block_group) {
4447                         btrfs_put_block_group(block_group);
4448                 }
4449         }
4450 search:
4451         trace_find_free_extent_search_loop(root, ffe_ctl);
4452         ffe_ctl->have_caching_bg = false;
4453         if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4454             ffe_ctl->index == 0)
4455                 full_search = true;
4456         down_read(&space_info->groups_sem);
4457         list_for_each_entry(block_group,
4458                             &space_info->block_groups[ffe_ctl->index], list) {
4459                 struct btrfs_block_group *bg_ret;
4460
4461                 ffe_ctl->hinted = false;
4462                 /* If the block group is read-only, we can skip it entirely. */
4463                 if (unlikely(block_group->ro)) {
4464                         if (ffe_ctl->for_treelog)
4465                                 btrfs_clear_treelog_bg(block_group);
4466                         if (ffe_ctl->for_data_reloc)
4467                                 btrfs_clear_data_reloc_bg(block_group);
4468                         continue;
4469                 }
4470
4471                 btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4472                 ffe_ctl->search_start = block_group->start;
4473
4474                 /*
4475                  * this can happen if we end up cycling through all the
4476                  * raid types, but we want to make sure we only allocate
4477                  * for the proper type.
4478                  */
4479                 if (!block_group_bits(block_group, ffe_ctl->flags)) {
4480                         u64 extra = BTRFS_BLOCK_GROUP_DUP |
4481                                 BTRFS_BLOCK_GROUP_RAID1_MASK |
4482                                 BTRFS_BLOCK_GROUP_RAID56_MASK |
4483                                 BTRFS_BLOCK_GROUP_RAID10;
4484
4485                         /*
4486                          * if they asked for extra copies and this block group
4487                          * doesn't provide them, bail.  This does allow us to
4488                          * fill raid0 from raid1.
4489                          */
4490                         if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4491                                 goto loop;
4492
4493                         /*
4494                          * This block group has different flags than we want.
4495                          * It's possible that we have MIXED_GROUP flag but no
4496                          * block group is mixed.  Just skip such block group.
4497                          */
4498                         btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4499                         continue;
4500                 }
4501
4502 have_block_group:
4503                 trace_find_free_extent_have_block_group(root, ffe_ctl, block_group);
4504                 ffe_ctl->cached = btrfs_block_group_done(block_group);
4505                 if (unlikely(!ffe_ctl->cached)) {
4506                         ffe_ctl->have_caching_bg = true;
4507                         ret = btrfs_cache_block_group(block_group, false);
4508
4509                         /*
4510                          * If we get ENOMEM here or something else we want to
4511                          * try other block groups, because it may not be fatal.
4512                          * However if we can't find anything else we need to
4513                          * save our return here so that we return the actual
4514                          * error that caused problems, not ENOSPC.
4515                          */
4516                         if (ret < 0) {
4517                                 if (!cache_block_group_error)
4518                                         cache_block_group_error = ret;
4519                                 ret = 0;
4520                                 goto loop;
4521                         }
4522                         ret = 0;
4523                 }
4524
4525                 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) {
4526                         if (!cache_block_group_error)
4527                                 cache_block_group_error = -EIO;
4528                         goto loop;
4529                 }
4530
4531                 if (!find_free_extent_check_size_class(ffe_ctl, block_group))
4532                         goto loop;
4533
4534                 bg_ret = NULL;
4535                 ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4536                 if (ret > 0)
4537                         goto loop;
4538
4539                 if (bg_ret && bg_ret != block_group) {
4540                         btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4541                         block_group = bg_ret;
4542                 }
4543
4544                 /* Checks */
4545                 ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4546                                                  fs_info->stripesize);
4547
4548                 /* move on to the next group */
4549                 if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4550                     block_group->start + block_group->length) {
4551                         btrfs_add_free_space_unused(block_group,
4552                                             ffe_ctl->found_offset,
4553                                             ffe_ctl->num_bytes);
4554                         goto loop;
4555                 }
4556
4557                 if (ffe_ctl->found_offset < ffe_ctl->search_start)
4558                         btrfs_add_free_space_unused(block_group,
4559                                         ffe_ctl->found_offset,
4560                                         ffe_ctl->search_start - ffe_ctl->found_offset);
4561
4562                 ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4563                                                ffe_ctl->num_bytes,
4564                                                ffe_ctl->delalloc,
4565                                                ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS);
4566                 if (ret == -EAGAIN) {
4567                         btrfs_add_free_space_unused(block_group,
4568                                         ffe_ctl->found_offset,
4569                                         ffe_ctl->num_bytes);
4570                         goto loop;
4571                 }
4572                 btrfs_inc_block_group_reservations(block_group);
4573
4574                 /* we are all good, lets return */
4575                 ins->objectid = ffe_ctl->search_start;
4576                 ins->offset = ffe_ctl->num_bytes;
4577
4578                 trace_btrfs_reserve_extent(block_group, ffe_ctl);
4579                 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4580                 break;
4581 loop:
4582                 if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
4583                     !ffe_ctl->retry_uncached) {
4584                         ffe_ctl->retry_uncached = true;
4585                         btrfs_wait_block_group_cache_progress(block_group,
4586                                                 ffe_ctl->num_bytes +
4587                                                 ffe_ctl->empty_cluster +
4588                                                 ffe_ctl->empty_size);
4589                         goto have_block_group;
4590                 }
4591                 release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4592                 cond_resched();
4593         }
4594         up_read(&space_info->groups_sem);
4595
4596         ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4597         if (ret > 0)
4598                 goto search;
4599
4600         if (ret == -ENOSPC && !cache_block_group_error) {
4601                 /*
4602                  * Use ffe_ctl->total_free_space as fallback if we can't find
4603                  * any contiguous hole.
4604                  */
4605                 if (!ffe_ctl->max_extent_size)
4606                         ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4607                 spin_lock(&space_info->lock);
4608                 space_info->max_extent_size = ffe_ctl->max_extent_size;
4609                 spin_unlock(&space_info->lock);
4610                 ins->offset = ffe_ctl->max_extent_size;
4611         } else if (ret == -ENOSPC) {
4612                 ret = cache_block_group_error;
4613         }
4614         return ret;
4615 }
4616
4617 /*
4618  * Entry point to the extent allocator. Tries to find a hole that is at least
4619  * as big as @num_bytes.
4620  *
4621  * @root           -    The root that will contain this extent
4622  *
4623  * @ram_bytes      -    The amount of space in ram that @num_bytes take. This
4624  *                      is used for accounting purposes. This value differs
4625  *                      from @num_bytes only in the case of compressed extents.
4626  *
4627  * @num_bytes      -    Number of bytes to allocate on-disk.
4628  *
4629  * @min_alloc_size -    Indicates the minimum amount of space that the
4630  *                      allocator should try to satisfy. In some cases
4631  *                      @num_bytes may be larger than what is required and if
4632  *                      the filesystem is fragmented then allocation fails.
4633  *                      However, the presence of @min_alloc_size gives a
4634  *                      chance to try and satisfy the smaller allocation.
4635  *
4636  * @empty_size     -    A hint that you plan on doing more COW. This is the
4637  *                      size in bytes the allocator should try to find free
4638  *                      next to the block it returns.  This is just a hint and
4639  *                      may be ignored by the allocator.
4640  *
4641  * @hint_byte      -    Hint to the allocator to start searching above the byte
4642  *                      address passed. It might be ignored.
4643  *
4644  * @ins            -    This key is modified to record the found hole. It will
4645  *                      have the following values:
4646  *                      ins->objectid == start position
4647  *                      ins->flags = BTRFS_EXTENT_ITEM_KEY
4648  *                      ins->offset == the size of the hole.
4649  *
4650  * @is_data        -    Boolean flag indicating whether an extent is
4651  *                      allocated for data (true) or metadata (false)
4652  *
4653  * @delalloc       -    Boolean flag indicating whether this allocation is for
4654  *                      delalloc or not. If 'true' data_rwsem of block groups
4655  *                      is going to be acquired.
4656  *
4657  *
4658  * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4659  * case -ENOSPC is returned then @ins->offset will contain the size of the
4660  * largest available hole the allocator managed to find.
4661  */
4662 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4663                          u64 num_bytes, u64 min_alloc_size,
4664                          u64 empty_size, u64 hint_byte,
4665                          struct btrfs_key *ins, int is_data, int delalloc)
4666 {
4667         struct btrfs_fs_info *fs_info = root->fs_info;
4668         struct find_free_extent_ctl ffe_ctl = {};
4669         bool final_tried = num_bytes == min_alloc_size;
4670         u64 flags;
4671         int ret;
4672         bool for_treelog = (btrfs_root_id(root) == BTRFS_TREE_LOG_OBJECTID);
4673         bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4674
4675         flags = get_alloc_profile_by_root(root, is_data);
4676 again:
4677         WARN_ON(num_bytes < fs_info->sectorsize);
4678
4679         ffe_ctl.ram_bytes = ram_bytes;
4680         ffe_ctl.num_bytes = num_bytes;
4681         ffe_ctl.min_alloc_size = min_alloc_size;
4682         ffe_ctl.empty_size = empty_size;
4683         ffe_ctl.flags = flags;
4684         ffe_ctl.delalloc = delalloc;
4685         ffe_ctl.hint_byte = hint_byte;
4686         ffe_ctl.for_treelog = for_treelog;
4687         ffe_ctl.for_data_reloc = for_data_reloc;
4688
4689         ret = find_free_extent(root, ins, &ffe_ctl);
4690         if (!ret && !is_data) {
4691                 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4692         } else if (ret == -ENOSPC) {
4693                 if (!final_tried && ins->offset) {
4694                         num_bytes = min(num_bytes >> 1, ins->offset);
4695                         num_bytes = round_down(num_bytes,
4696                                                fs_info->sectorsize);
4697                         num_bytes = max(num_bytes, min_alloc_size);
4698                         ram_bytes = num_bytes;
4699                         if (num_bytes == min_alloc_size)
4700                                 final_tried = true;
4701                         goto again;
4702                 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4703                         struct btrfs_space_info *sinfo;
4704
4705                         sinfo = btrfs_find_space_info(fs_info, flags);
4706                         btrfs_err(fs_info,
4707         "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4708                                   flags, num_bytes, for_treelog, for_data_reloc);
4709                         if (sinfo)
4710                                 btrfs_dump_space_info(fs_info, sinfo,
4711                                                       num_bytes, 1);
4712                 }
4713         }
4714
4715         return ret;
4716 }
4717
4718 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4719                                u64 start, u64 len, int delalloc)
4720 {
4721         struct btrfs_block_group *cache;
4722
4723         cache = btrfs_lookup_block_group(fs_info, start);
4724         if (!cache) {
4725                 btrfs_err(fs_info, "Unable to find block group for %llu",
4726                           start);
4727                 return -ENOSPC;
4728         }
4729
4730         btrfs_add_free_space(cache, start, len);
4731         btrfs_free_reserved_bytes(cache, len, delalloc);
4732         trace_btrfs_reserved_extent_free(fs_info, start, len);
4733
4734         btrfs_put_block_group(cache);
4735         return 0;
4736 }
4737
4738 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans,
4739                               const struct extent_buffer *eb)
4740 {
4741         struct btrfs_block_group *cache;
4742         int ret = 0;
4743
4744         cache = btrfs_lookup_block_group(trans->fs_info, eb->start);
4745         if (!cache) {
4746                 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4747                           eb->start);
4748                 return -ENOSPC;
4749         }
4750
4751         ret = pin_down_extent(trans, cache, eb->start, eb->len, 1);
4752         btrfs_put_block_group(cache);
4753         return ret;
4754 }
4755
4756 static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
4757                                  u64 num_bytes)
4758 {
4759         struct btrfs_fs_info *fs_info = trans->fs_info;
4760         int ret;
4761
4762         ret = remove_from_free_space_tree(trans, bytenr, num_bytes);
4763         if (ret)
4764                 return ret;
4765
4766         ret = btrfs_update_block_group(trans, bytenr, num_bytes, true);
4767         if (ret) {
4768                 ASSERT(!ret);
4769                 btrfs_err(fs_info, "update block group failed for %llu %llu",
4770                           bytenr, num_bytes);
4771                 return ret;
4772         }
4773
4774         trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes);
4775         return 0;
4776 }
4777
4778 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4779                                       u64 parent, u64 root_objectid,
4780                                       u64 flags, u64 owner, u64 offset,
4781                                       struct btrfs_key *ins, int ref_mod, u64 oref_root)
4782 {
4783         struct btrfs_fs_info *fs_info = trans->fs_info;
4784         struct btrfs_root *extent_root;
4785         int ret;
4786         struct btrfs_extent_item *extent_item;
4787         struct btrfs_extent_owner_ref *oref;
4788         struct btrfs_extent_inline_ref *iref;
4789         struct btrfs_path *path;
4790         struct extent_buffer *leaf;
4791         int type;
4792         u32 size;
4793         const bool simple_quota = (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE);
4794
4795         if (parent > 0)
4796                 type = BTRFS_SHARED_DATA_REF_KEY;
4797         else
4798                 type = BTRFS_EXTENT_DATA_REF_KEY;
4799
4800         size = sizeof(*extent_item);
4801         if (simple_quota)
4802                 size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY);
4803         size += btrfs_extent_inline_ref_size(type);
4804
4805         path = btrfs_alloc_path();
4806         if (!path)
4807                 return -ENOMEM;
4808
4809         extent_root = btrfs_extent_root(fs_info, ins->objectid);
4810         ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size);
4811         if (ret) {
4812                 btrfs_free_path(path);
4813                 return ret;
4814         }
4815
4816         leaf = path->nodes[0];
4817         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4818                                      struct btrfs_extent_item);
4819         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4820         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4821         btrfs_set_extent_flags(leaf, extent_item,
4822                                flags | BTRFS_EXTENT_FLAG_DATA);
4823
4824         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4825         if (simple_quota) {
4826                 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_EXTENT_OWNER_REF_KEY);
4827                 oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
4828                 btrfs_set_extent_owner_ref_root_id(leaf, oref, oref_root);
4829                 iref = (struct btrfs_extent_inline_ref *)(oref + 1);
4830         }
4831         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4832
4833         if (parent > 0) {
4834                 struct btrfs_shared_data_ref *ref;
4835                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4836                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4837                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4838         } else {
4839                 struct btrfs_extent_data_ref *ref;
4840                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4841                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4842                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4843                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4844                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4845         }
4846
4847         btrfs_mark_buffer_dirty(trans, path->nodes[0]);
4848         btrfs_free_path(path);
4849
4850         return alloc_reserved_extent(trans, ins->objectid, ins->offset);
4851 }
4852
4853 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4854                                      struct btrfs_delayed_ref_node *node,
4855                                      struct btrfs_delayed_extent_op *extent_op)
4856 {
4857         struct btrfs_fs_info *fs_info = trans->fs_info;
4858         struct btrfs_root *extent_root;
4859         int ret;
4860         struct btrfs_extent_item *extent_item;
4861         struct btrfs_key extent_key;
4862         struct btrfs_tree_block_info *block_info;
4863         struct btrfs_extent_inline_ref *iref;
4864         struct btrfs_path *path;
4865         struct extent_buffer *leaf;
4866         u32 size = sizeof(*extent_item) + sizeof(*iref);
4867         u64 flags = extent_op->flags_to_set;
4868         /* The owner of a tree block is the level. */
4869         int level = btrfs_delayed_ref_owner(node);
4870         bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4871
4872         extent_key.objectid = node->bytenr;
4873         if (skinny_metadata) {
4874                 /* The owner of a tree block is the level. */
4875                 extent_key.offset = level;
4876                 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4877         } else {
4878                 extent_key.offset = node->num_bytes;
4879                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4880                 size += sizeof(*block_info);
4881         }
4882
4883         path = btrfs_alloc_path();
4884         if (!path)
4885                 return -ENOMEM;
4886
4887         extent_root = btrfs_extent_root(fs_info, extent_key.objectid);
4888         ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key,
4889                                       size);
4890         if (ret) {
4891                 btrfs_free_path(path);
4892                 return ret;
4893         }
4894
4895         leaf = path->nodes[0];
4896         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4897                                      struct btrfs_extent_item);
4898         btrfs_set_extent_refs(leaf, extent_item, 1);
4899         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4900         btrfs_set_extent_flags(leaf, extent_item,
4901                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4902
4903         if (skinny_metadata) {
4904                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4905         } else {
4906                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4907                 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4908                 btrfs_set_tree_block_level(leaf, block_info, level);
4909                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4910         }
4911
4912         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4913                 btrfs_set_extent_inline_ref_type(leaf, iref,
4914                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4915                 btrfs_set_extent_inline_ref_offset(leaf, iref, node->parent);
4916         } else {
4917                 btrfs_set_extent_inline_ref_type(leaf, iref,
4918                                                  BTRFS_TREE_BLOCK_REF_KEY);
4919                 btrfs_set_extent_inline_ref_offset(leaf, iref, node->ref_root);
4920         }
4921
4922         btrfs_mark_buffer_dirty(trans, leaf);
4923         btrfs_free_path(path);
4924
4925         return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize);
4926 }
4927
4928 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4929                                      struct btrfs_root *root, u64 owner,
4930                                      u64 offset, u64 ram_bytes,
4931                                      struct btrfs_key *ins)
4932 {
4933         struct btrfs_ref generic_ref = {
4934                 .action = BTRFS_ADD_DELAYED_EXTENT,
4935                 .bytenr = ins->objectid,
4936                 .num_bytes = ins->offset,
4937                 .owning_root = btrfs_root_id(root),
4938                 .ref_root = btrfs_root_id(root),
4939         };
4940
4941         ASSERT(generic_ref.ref_root != BTRFS_TREE_LOG_OBJECTID);
4942
4943         if (btrfs_is_data_reloc_root(root) && is_fstree(root->relocation_src_root))
4944                 generic_ref.owning_root = root->relocation_src_root;
4945
4946         btrfs_init_data_ref(&generic_ref, owner, offset, 0, false);
4947         btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4948
4949         return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4950 }
4951
4952 /*
4953  * this is used by the tree logging recovery code.  It records that
4954  * an extent has been allocated and makes sure to clear the free
4955  * space cache bits as well
4956  */
4957 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4958                                    u64 root_objectid, u64 owner, u64 offset,
4959                                    struct btrfs_key *ins)
4960 {
4961         struct btrfs_fs_info *fs_info = trans->fs_info;
4962         int ret;
4963         struct btrfs_block_group *block_group;
4964         struct btrfs_space_info *space_info;
4965         struct btrfs_squota_delta delta = {
4966                 .root = root_objectid,
4967                 .num_bytes = ins->offset,
4968                 .generation = trans->transid,
4969                 .is_data = true,
4970                 .is_inc = true,
4971         };
4972
4973         /*
4974          * Mixed block groups will exclude before processing the log so we only
4975          * need to do the exclude dance if this fs isn't mixed.
4976          */
4977         if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4978                 ret = __exclude_logged_extent(fs_info, ins->objectid,
4979                                               ins->offset);
4980                 if (ret)
4981                         return ret;
4982         }
4983
4984         block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4985         if (!block_group)
4986                 return -EINVAL;
4987
4988         space_info = block_group->space_info;
4989         spin_lock(&space_info->lock);
4990         spin_lock(&block_group->lock);
4991         space_info->bytes_reserved += ins->offset;
4992         block_group->reserved += ins->offset;
4993         spin_unlock(&block_group->lock);
4994         spin_unlock(&space_info->lock);
4995
4996         ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4997                                          offset, ins, 1, root_objectid);
4998         if (ret)
4999                 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
5000         ret = btrfs_record_squota_delta(fs_info, &delta);
5001         btrfs_put_block_group(block_group);
5002         return ret;
5003 }
5004
5005 #ifdef CONFIG_BTRFS_DEBUG
5006 /*
5007  * Extra safety check in case the extent tree is corrupted and extent allocator
5008  * chooses to use a tree block which is already used and locked.
5009  */
5010 static bool check_eb_lock_owner(const struct extent_buffer *eb)
5011 {
5012         if (eb->lock_owner == current->pid) {
5013                 btrfs_err_rl(eb->fs_info,
5014 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
5015                              eb->start, btrfs_header_owner(eb), current->pid);
5016                 return true;
5017         }
5018         return false;
5019 }
5020 #else
5021 static bool check_eb_lock_owner(struct extent_buffer *eb)
5022 {
5023         return false;
5024 }
5025 #endif
5026
5027 static struct extent_buffer *
5028 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5029                       u64 bytenr, int level, u64 owner,
5030                       enum btrfs_lock_nesting nest)
5031 {
5032         struct btrfs_fs_info *fs_info = root->fs_info;
5033         struct extent_buffer *buf;
5034         u64 lockdep_owner = owner;
5035
5036         buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
5037         if (IS_ERR(buf))
5038                 return buf;
5039
5040         if (check_eb_lock_owner(buf)) {
5041                 free_extent_buffer(buf);
5042                 return ERR_PTR(-EUCLEAN);
5043         }
5044
5045         /*
5046          * The reloc trees are just snapshots, so we need them to appear to be
5047          * just like any other fs tree WRT lockdep.
5048          *
5049          * The exception however is in replace_path() in relocation, where we
5050          * hold the lock on the original fs root and then search for the reloc
5051          * root.  At that point we need to make sure any reloc root buffers are
5052          * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
5053          * lockdep happy.
5054          */
5055         if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID &&
5056             !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
5057                 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
5058
5059         /* btrfs_clear_buffer_dirty() accesses generation field. */
5060         btrfs_set_header_generation(buf, trans->transid);
5061
5062         /*
5063          * This needs to stay, because we could allocate a freed block from an
5064          * old tree into a new tree, so we need to make sure this new block is
5065          * set to the appropriate level and owner.
5066          */
5067         btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level);
5068
5069         btrfs_tree_lock_nested(buf, nest);
5070         btrfs_clear_buffer_dirty(trans, buf);
5071         clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
5072         clear_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &buf->bflags);
5073
5074         set_extent_buffer_uptodate(buf);
5075
5076         memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
5077         btrfs_set_header_level(buf, level);
5078         btrfs_set_header_bytenr(buf, buf->start);
5079         btrfs_set_header_generation(buf, trans->transid);
5080         btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
5081         btrfs_set_header_owner(buf, owner);
5082         write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
5083         write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
5084         if (btrfs_root_id(root) == BTRFS_TREE_LOG_OBJECTID) {
5085                 buf->log_index = root->log_transid % 2;
5086                 /*
5087                  * we allow two log transactions at a time, use different
5088                  * EXTENT bit to differentiate dirty pages.
5089                  */
5090                 if (buf->log_index == 0)
5091                         set_extent_bit(&root->dirty_log_pages, buf->start,
5092                                        buf->start + buf->len - 1,
5093                                        EXTENT_DIRTY, NULL);
5094                 else
5095                         set_extent_bit(&root->dirty_log_pages, buf->start,
5096                                        buf->start + buf->len - 1,
5097                                        EXTENT_NEW, NULL);
5098         } else {
5099                 buf->log_index = -1;
5100                 set_extent_bit(&trans->transaction->dirty_pages, buf->start,
5101                                buf->start + buf->len - 1, EXTENT_DIRTY, NULL);
5102         }
5103         /* this returns a buffer locked for blocking */
5104         return buf;
5105 }
5106
5107 /*
5108  * finds a free extent and does all the dirty work required for allocation
5109  * returns the tree buffer or an ERR_PTR on error.
5110  */
5111 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
5112                                              struct btrfs_root *root,
5113                                              u64 parent, u64 root_objectid,
5114                                              const struct btrfs_disk_key *key,
5115                                              int level, u64 hint,
5116                                              u64 empty_size,
5117                                              u64 reloc_src_root,
5118                                              enum btrfs_lock_nesting nest)
5119 {
5120         struct btrfs_fs_info *fs_info = root->fs_info;
5121         struct btrfs_key ins;
5122         struct btrfs_block_rsv *block_rsv;
5123         struct extent_buffer *buf;
5124         struct btrfs_delayed_extent_op *extent_op;
5125         u64 flags = 0;
5126         int ret;
5127         u32 blocksize = fs_info->nodesize;
5128         bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
5129         u64 owning_root;
5130
5131 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5132         if (btrfs_is_testing(fs_info)) {
5133                 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
5134                                             level, root_objectid, nest);
5135                 if (!IS_ERR(buf))
5136                         root->alloc_bytenr += blocksize;
5137                 return buf;
5138         }
5139 #endif
5140
5141         block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
5142         if (IS_ERR(block_rsv))
5143                 return ERR_CAST(block_rsv);
5144
5145         ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
5146                                    empty_size, hint, &ins, 0, 0);
5147         if (ret)
5148                 goto out_unuse;
5149
5150         buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
5151                                     root_objectid, nest);
5152         if (IS_ERR(buf)) {
5153                 ret = PTR_ERR(buf);
5154                 goto out_free_reserved;
5155         }
5156         owning_root = btrfs_header_owner(buf);
5157
5158         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5159                 if (parent == 0)
5160                         parent = ins.objectid;
5161                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5162                 owning_root = reloc_src_root;
5163         } else
5164                 BUG_ON(parent > 0);
5165
5166         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5167                 struct btrfs_ref generic_ref = {
5168                         .action = BTRFS_ADD_DELAYED_EXTENT,
5169                         .bytenr = ins.objectid,
5170                         .num_bytes = ins.offset,
5171                         .parent = parent,
5172                         .owning_root = owning_root,
5173                         .ref_root = root_objectid,
5174                 };
5175                 extent_op = btrfs_alloc_delayed_extent_op();
5176                 if (!extent_op) {
5177                         ret = -ENOMEM;
5178                         goto out_free_buf;
5179                 }
5180                 if (key)
5181                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5182                 else
5183                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5184                 extent_op->flags_to_set = flags;
5185                 extent_op->update_key = skinny_metadata ? false : true;
5186                 extent_op->update_flags = true;
5187                 extent_op->level = level;
5188
5189                 btrfs_init_tree_ref(&generic_ref, level, btrfs_root_id(root), false);
5190                 btrfs_ref_tree_mod(fs_info, &generic_ref);
5191                 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
5192                 if (ret)
5193                         goto out_free_delayed;
5194         }
5195         return buf;
5196
5197 out_free_delayed:
5198         btrfs_free_delayed_extent_op(extent_op);
5199 out_free_buf:
5200         btrfs_tree_unlock(buf);
5201         free_extent_buffer(buf);
5202 out_free_reserved:
5203         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
5204 out_unuse:
5205         btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
5206         return ERR_PTR(ret);
5207 }
5208
5209 struct walk_control {
5210         u64 refs[BTRFS_MAX_LEVEL];
5211         u64 flags[BTRFS_MAX_LEVEL];
5212         struct btrfs_key update_progress;
5213         struct btrfs_key drop_progress;
5214         int drop_level;
5215         int stage;
5216         int level;
5217         int shared_level;
5218         int update_ref;
5219         int keep_locks;
5220         int reada_slot;
5221         int reada_count;
5222         int restarted;
5223 };
5224
5225 #define DROP_REFERENCE  1
5226 #define UPDATE_BACKREF  2
5227
5228 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5229                                      struct btrfs_root *root,
5230                                      struct walk_control *wc,
5231                                      struct btrfs_path *path)
5232 {
5233         struct btrfs_fs_info *fs_info = root->fs_info;
5234         u64 bytenr;
5235         u64 generation;
5236         u64 refs;
5237         u64 flags;
5238         u32 nritems;
5239         struct btrfs_key key;
5240         struct extent_buffer *eb;
5241         int ret;
5242         int slot;
5243         int nread = 0;
5244
5245         if (path->slots[wc->level] < wc->reada_slot) {
5246                 wc->reada_count = wc->reada_count * 2 / 3;
5247                 wc->reada_count = max(wc->reada_count, 2);
5248         } else {
5249                 wc->reada_count = wc->reada_count * 3 / 2;
5250                 wc->reada_count = min_t(int, wc->reada_count,
5251                                         BTRFS_NODEPTRS_PER_BLOCK(fs_info));
5252         }
5253
5254         eb = path->nodes[wc->level];
5255         nritems = btrfs_header_nritems(eb);
5256
5257         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5258                 if (nread >= wc->reada_count)
5259                         break;
5260
5261                 cond_resched();
5262                 bytenr = btrfs_node_blockptr(eb, slot);
5263                 generation = btrfs_node_ptr_generation(eb, slot);
5264
5265                 if (slot == path->slots[wc->level])
5266                         goto reada;
5267
5268                 if (wc->stage == UPDATE_BACKREF &&
5269                     generation <= root->root_key.offset)
5270                         continue;
5271
5272                 /* We don't lock the tree block, it's OK to be racy here */
5273                 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5274                                                wc->level - 1, 1, &refs,
5275                                                &flags, NULL);
5276                 /* We don't care about errors in readahead. */
5277                 if (ret < 0)
5278                         continue;
5279                 BUG_ON(refs == 0);
5280
5281                 if (wc->stage == DROP_REFERENCE) {
5282                         if (refs == 1)
5283                                 goto reada;
5284
5285                         if (wc->level == 1 &&
5286                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5287                                 continue;
5288                         if (!wc->update_ref ||
5289                             generation <= root->root_key.offset)
5290                                 continue;
5291                         btrfs_node_key_to_cpu(eb, &key, slot);
5292                         ret = btrfs_comp_cpu_keys(&key,
5293                                                   &wc->update_progress);
5294                         if (ret < 0)
5295                                 continue;
5296                 } else {
5297                         if (wc->level == 1 &&
5298                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5299                                 continue;
5300                 }
5301 reada:
5302                 btrfs_readahead_node_child(eb, slot);
5303                 nread++;
5304         }
5305         wc->reada_slot = slot;
5306 }
5307
5308 /*
5309  * helper to process tree block while walking down the tree.
5310  *
5311  * when wc->stage == UPDATE_BACKREF, this function updates
5312  * back refs for pointers in the block.
5313  *
5314  * NOTE: return value 1 means we should stop walking down.
5315  */
5316 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5317                                    struct btrfs_root *root,
5318                                    struct btrfs_path *path,
5319                                    struct walk_control *wc, int lookup_info)
5320 {
5321         struct btrfs_fs_info *fs_info = root->fs_info;
5322         int level = wc->level;
5323         struct extent_buffer *eb = path->nodes[level];
5324         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5325         int ret;
5326
5327         if (wc->stage == UPDATE_BACKREF && btrfs_header_owner(eb) != btrfs_root_id(root))
5328                 return 1;
5329
5330         /*
5331          * when reference count of tree block is 1, it won't increase
5332          * again. once full backref flag is set, we never clear it.
5333          */
5334         if (lookup_info &&
5335             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5336              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5337                 BUG_ON(!path->locks[level]);
5338                 ret = btrfs_lookup_extent_info(trans, fs_info,
5339                                                eb->start, level, 1,
5340                                                &wc->refs[level],
5341                                                &wc->flags[level],
5342                                                NULL);
5343                 BUG_ON(ret == -ENOMEM);
5344                 if (ret)
5345                         return ret;
5346                 BUG_ON(wc->refs[level] == 0);
5347         }
5348
5349         if (wc->stage == DROP_REFERENCE) {
5350                 if (wc->refs[level] > 1)
5351                         return 1;
5352
5353                 if (path->locks[level] && !wc->keep_locks) {
5354                         btrfs_tree_unlock_rw(eb, path->locks[level]);
5355                         path->locks[level] = 0;
5356                 }
5357                 return 0;
5358         }
5359
5360         /* wc->stage == UPDATE_BACKREF */
5361         if (!(wc->flags[level] & flag)) {
5362                 BUG_ON(!path->locks[level]);
5363                 ret = btrfs_inc_ref(trans, root, eb, 1);
5364                 BUG_ON(ret); /* -ENOMEM */
5365                 ret = btrfs_dec_ref(trans, root, eb, 0);
5366                 BUG_ON(ret); /* -ENOMEM */
5367                 ret = btrfs_set_disk_extent_flags(trans, eb, flag);
5368                 BUG_ON(ret); /* -ENOMEM */
5369                 wc->flags[level] |= flag;
5370         }
5371
5372         /*
5373          * the block is shared by multiple trees, so it's not good to
5374          * keep the tree lock
5375          */
5376         if (path->locks[level] && level > 0) {
5377                 btrfs_tree_unlock_rw(eb, path->locks[level]);
5378                 path->locks[level] = 0;
5379         }
5380         return 0;
5381 }
5382
5383 /*
5384  * This is used to verify a ref exists for this root to deal with a bug where we
5385  * would have a drop_progress key that hadn't been updated properly.
5386  */
5387 static int check_ref_exists(struct btrfs_trans_handle *trans,
5388                             struct btrfs_root *root, u64 bytenr, u64 parent,
5389                             int level)
5390 {
5391         struct btrfs_path *path;
5392         struct btrfs_extent_inline_ref *iref;
5393         int ret;
5394
5395         path = btrfs_alloc_path();
5396         if (!path)
5397                 return -ENOMEM;
5398
5399         ret = lookup_extent_backref(trans, path, &iref, bytenr,
5400                                     root->fs_info->nodesize, parent,
5401                                     btrfs_root_id(root), level, 0);
5402         btrfs_free_path(path);
5403         if (ret == -ENOENT)
5404                 return 0;
5405         if (ret < 0)
5406                 return ret;
5407         return 1;
5408 }
5409
5410 /*
5411  * helper to process tree block pointer.
5412  *
5413  * when wc->stage == DROP_REFERENCE, this function checks
5414  * reference count of the block pointed to. if the block
5415  * is shared and we need update back refs for the subtree
5416  * rooted at the block, this function changes wc->stage to
5417  * UPDATE_BACKREF. if the block is shared and there is no
5418  * need to update back, this function drops the reference
5419  * to the block.
5420  *
5421  * NOTE: return value 1 means we should stop walking down.
5422  */
5423 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5424                                  struct btrfs_root *root,
5425                                  struct btrfs_path *path,
5426                                  struct walk_control *wc, int *lookup_info)
5427 {
5428         struct btrfs_fs_info *fs_info = root->fs_info;
5429         u64 bytenr;
5430         u64 generation;
5431         u64 owner_root = 0;
5432         struct btrfs_tree_parent_check check = { 0 };
5433         struct btrfs_key key;
5434         struct extent_buffer *next;
5435         int level = wc->level;
5436         int reada = 0;
5437         int ret = 0;
5438         bool need_account = false;
5439
5440         generation = btrfs_node_ptr_generation(path->nodes[level],
5441                                                path->slots[level]);
5442         /*
5443          * if the lower level block was created before the snapshot
5444          * was created, we know there is no need to update back refs
5445          * for the subtree
5446          */
5447         if (wc->stage == UPDATE_BACKREF &&
5448             generation <= root->root_key.offset) {
5449                 *lookup_info = 1;
5450                 return 1;
5451         }
5452
5453         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5454
5455         check.level = level - 1;
5456         check.transid = generation;
5457         check.owner_root = btrfs_root_id(root);
5458         check.has_first_key = true;
5459         btrfs_node_key_to_cpu(path->nodes[level], &check.first_key,
5460                               path->slots[level]);
5461
5462         next = find_extent_buffer(fs_info, bytenr);
5463         if (!next) {
5464                 next = btrfs_find_create_tree_block(fs_info, bytenr,
5465                                 btrfs_root_id(root), level - 1);
5466                 if (IS_ERR(next))
5467                         return PTR_ERR(next);
5468                 reada = 1;
5469         }
5470         btrfs_tree_lock(next);
5471
5472         ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5473                                        &wc->refs[level - 1],
5474                                        &wc->flags[level - 1],
5475                                        &owner_root);
5476         if (ret < 0)
5477                 goto out_unlock;
5478
5479         if (unlikely(wc->refs[level - 1] == 0)) {
5480                 btrfs_err(fs_info, "Missing references.");
5481                 ret = -EIO;
5482                 goto out_unlock;
5483         }
5484         *lookup_info = 0;
5485
5486         if (wc->stage == DROP_REFERENCE) {
5487                 if (wc->refs[level - 1] > 1) {
5488                         need_account = true;
5489                         if (level == 1 &&
5490                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5491                                 goto skip;
5492
5493                         if (!wc->update_ref ||
5494                             generation <= root->root_key.offset)
5495                                 goto skip;
5496
5497                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5498                                               path->slots[level]);
5499                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5500                         if (ret < 0)
5501                                 goto skip;
5502
5503                         wc->stage = UPDATE_BACKREF;
5504                         wc->shared_level = level - 1;
5505                 }
5506         } else {
5507                 if (level == 1 &&
5508                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5509                         goto skip;
5510         }
5511
5512         if (!btrfs_buffer_uptodate(next, generation, 0)) {
5513                 btrfs_tree_unlock(next);
5514                 free_extent_buffer(next);
5515                 next = NULL;
5516                 *lookup_info = 1;
5517         }
5518
5519         if (!next) {
5520                 if (reada && level == 1)
5521                         reada_walk_down(trans, root, wc, path);
5522                 next = read_tree_block(fs_info, bytenr, &check);
5523                 if (IS_ERR(next)) {
5524                         return PTR_ERR(next);
5525                 } else if (!extent_buffer_uptodate(next)) {
5526                         free_extent_buffer(next);
5527                         return -EIO;
5528                 }
5529                 btrfs_tree_lock(next);
5530         }
5531
5532         level--;
5533         ASSERT(level == btrfs_header_level(next));
5534         if (level != btrfs_header_level(next)) {
5535                 btrfs_err(root->fs_info, "mismatched level");
5536                 ret = -EIO;
5537                 goto out_unlock;
5538         }
5539         path->nodes[level] = next;
5540         path->slots[level] = 0;
5541         path->locks[level] = BTRFS_WRITE_LOCK;
5542         wc->level = level;
5543         if (wc->level == 1)
5544                 wc->reada_slot = 0;
5545         return 0;
5546 skip:
5547         wc->refs[level - 1] = 0;
5548         wc->flags[level - 1] = 0;
5549         if (wc->stage == DROP_REFERENCE) {
5550                 struct btrfs_ref ref = {
5551                         .action = BTRFS_DROP_DELAYED_REF,
5552                         .bytenr = bytenr,
5553                         .num_bytes = fs_info->nodesize,
5554                         .owning_root = owner_root,
5555                         .ref_root = btrfs_root_id(root),
5556                 };
5557                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5558                         ref.parent = path->nodes[level]->start;
5559                 } else {
5560                         ASSERT(btrfs_root_id(root) ==
5561                                btrfs_header_owner(path->nodes[level]));
5562                         if (btrfs_root_id(root) !=
5563                             btrfs_header_owner(path->nodes[level])) {
5564                                 btrfs_err(root->fs_info,
5565                                                 "mismatched block owner");
5566                                 ret = -EIO;
5567                                 goto out_unlock;
5568                         }
5569                 }
5570
5571                 /*
5572                  * If we had a drop_progress we need to verify the refs are set
5573                  * as expected.  If we find our ref then we know that from here
5574                  * on out everything should be correct, and we can clear the
5575                  * ->restarted flag.
5576                  */
5577                 if (wc->restarted) {
5578                         ret = check_ref_exists(trans, root, bytenr, ref.parent,
5579                                                level - 1);
5580                         if (ret < 0)
5581                                 goto out_unlock;
5582                         if (ret == 0)
5583                                 goto no_delete;
5584                         ret = 0;
5585                         wc->restarted = 0;
5586                 }
5587
5588                 /*
5589                  * Reloc tree doesn't contribute to qgroup numbers, and we have
5590                  * already accounted them at merge time (replace_path),
5591                  * thus we could skip expensive subtree trace here.
5592                  */
5593                 if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID && need_account) {
5594                         ret = btrfs_qgroup_trace_subtree(trans, next,
5595                                                          generation, level - 1);
5596                         if (ret) {
5597                                 btrfs_err_rl(fs_info,
5598                                              "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5599                                              ret);
5600                         }
5601                 }
5602
5603                 /*
5604                  * We need to update the next key in our walk control so we can
5605                  * update the drop_progress key accordingly.  We don't care if
5606                  * find_next_key doesn't find a key because that means we're at
5607                  * the end and are going to clean up now.
5608                  */
5609                 wc->drop_level = level;
5610                 find_next_key(path, level, &wc->drop_progress);
5611
5612                 btrfs_init_tree_ref(&ref, level - 1, 0, false);
5613                 ret = btrfs_free_extent(trans, &ref);
5614                 if (ret)
5615                         goto out_unlock;
5616         }
5617 no_delete:
5618         *lookup_info = 1;
5619         ret = 1;
5620
5621 out_unlock:
5622         btrfs_tree_unlock(next);
5623         free_extent_buffer(next);
5624
5625         return ret;
5626 }
5627
5628 /*
5629  * helper to process tree block while walking up the tree.
5630  *
5631  * when wc->stage == DROP_REFERENCE, this function drops
5632  * reference count on the block.
5633  *
5634  * when wc->stage == UPDATE_BACKREF, this function changes
5635  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5636  * to UPDATE_BACKREF previously while processing the block.
5637  *
5638  * NOTE: return value 1 means we should stop walking up.
5639  */
5640 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5641                                  struct btrfs_root *root,
5642                                  struct btrfs_path *path,
5643                                  struct walk_control *wc)
5644 {
5645         struct btrfs_fs_info *fs_info = root->fs_info;
5646         int ret;
5647         int level = wc->level;
5648         struct extent_buffer *eb = path->nodes[level];
5649         u64 parent = 0;
5650
5651         if (wc->stage == UPDATE_BACKREF) {
5652                 BUG_ON(wc->shared_level < level);
5653                 if (level < wc->shared_level)
5654                         goto out;
5655
5656                 ret = find_next_key(path, level + 1, &wc->update_progress);
5657                 if (ret > 0)
5658                         wc->update_ref = 0;
5659
5660                 wc->stage = DROP_REFERENCE;
5661                 wc->shared_level = -1;
5662                 path->slots[level] = 0;
5663
5664                 /*
5665                  * check reference count again if the block isn't locked.
5666                  * we should start walking down the tree again if reference
5667                  * count is one.
5668                  */
5669                 if (!path->locks[level]) {
5670                         BUG_ON(level == 0);
5671                         btrfs_tree_lock(eb);
5672                         path->locks[level] = BTRFS_WRITE_LOCK;
5673
5674                         ret = btrfs_lookup_extent_info(trans, fs_info,
5675                                                        eb->start, level, 1,
5676                                                        &wc->refs[level],
5677                                                        &wc->flags[level],
5678                                                        NULL);
5679                         if (ret < 0) {
5680                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
5681                                 path->locks[level] = 0;
5682                                 return ret;
5683                         }
5684                         BUG_ON(wc->refs[level] == 0);
5685                         if (wc->refs[level] == 1) {
5686                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
5687                                 path->locks[level] = 0;
5688                                 return 1;
5689                         }
5690                 }
5691         }
5692
5693         /* wc->stage == DROP_REFERENCE */
5694         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5695
5696         if (wc->refs[level] == 1) {
5697                 if (level == 0) {
5698                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5699                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5700                         else
5701                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5702                         BUG_ON(ret); /* -ENOMEM */
5703                         if (is_fstree(btrfs_root_id(root))) {
5704                                 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5705                                 if (ret) {
5706                                         btrfs_err_rl(fs_info,
5707         "error %d accounting leaf items, quota is out of sync, rescan required",
5708                                              ret);
5709                                 }
5710                         }
5711                 }
5712                 /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */
5713                 if (!path->locks[level]) {
5714                         btrfs_tree_lock(eb);
5715                         path->locks[level] = BTRFS_WRITE_LOCK;
5716                 }
5717                 btrfs_clear_buffer_dirty(trans, eb);
5718         }
5719
5720         if (eb == root->node) {
5721                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5722                         parent = eb->start;
5723                 else if (btrfs_root_id(root) != btrfs_header_owner(eb))
5724                         goto owner_mismatch;
5725         } else {
5726                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5727                         parent = path->nodes[level + 1]->start;
5728                 else if (btrfs_root_id(root) !=
5729                          btrfs_header_owner(path->nodes[level + 1]))
5730                         goto owner_mismatch;
5731         }
5732
5733         btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent,
5734                               wc->refs[level] == 1);
5735 out:
5736         wc->refs[level] = 0;
5737         wc->flags[level] = 0;
5738         return 0;
5739
5740 owner_mismatch:
5741         btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5742                      btrfs_header_owner(eb), btrfs_root_id(root));
5743         return -EUCLEAN;
5744 }
5745
5746 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5747                                    struct btrfs_root *root,
5748                                    struct btrfs_path *path,
5749                                    struct walk_control *wc)
5750 {
5751         int level = wc->level;
5752         int lookup_info = 1;
5753         int ret = 0;
5754
5755         while (level >= 0) {
5756                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5757                 if (ret)
5758                         break;
5759
5760                 if (level == 0)
5761                         break;
5762
5763                 if (path->slots[level] >=
5764                     btrfs_header_nritems(path->nodes[level]))
5765                         break;
5766
5767                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5768                 if (ret > 0) {
5769                         path->slots[level]++;
5770                         continue;
5771                 } else if (ret < 0)
5772                         break;
5773                 level = wc->level;
5774         }
5775         return (ret == 1) ? 0 : ret;
5776 }
5777
5778 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5779                                  struct btrfs_root *root,
5780                                  struct btrfs_path *path,
5781                                  struct walk_control *wc, int max_level)
5782 {
5783         int level = wc->level;
5784         int ret;
5785
5786         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5787         while (level < max_level && path->nodes[level]) {
5788                 wc->level = level;
5789                 if (path->slots[level] + 1 <
5790                     btrfs_header_nritems(path->nodes[level])) {
5791                         path->slots[level]++;
5792                         return 0;
5793                 } else {
5794                         ret = walk_up_proc(trans, root, path, wc);
5795                         if (ret > 0)
5796                                 return 0;
5797                         if (ret < 0)
5798                                 return ret;
5799
5800                         if (path->locks[level]) {
5801                                 btrfs_tree_unlock_rw(path->nodes[level],
5802                                                      path->locks[level]);
5803                                 path->locks[level] = 0;
5804                         }
5805                         free_extent_buffer(path->nodes[level]);
5806                         path->nodes[level] = NULL;
5807                         level++;
5808                 }
5809         }
5810         return 1;
5811 }
5812
5813 /*
5814  * drop a subvolume tree.
5815  *
5816  * this function traverses the tree freeing any blocks that only
5817  * referenced by the tree.
5818  *
5819  * when a shared tree block is found. this function decreases its
5820  * reference count by one. if update_ref is true, this function
5821  * also make sure backrefs for the shared block and all lower level
5822  * blocks are properly updated.
5823  *
5824  * If called with for_reloc == 0, may exit early with -EAGAIN
5825  */
5826 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5827 {
5828         const bool is_reloc_root = (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID);
5829         struct btrfs_fs_info *fs_info = root->fs_info;
5830         struct btrfs_path *path;
5831         struct btrfs_trans_handle *trans;
5832         struct btrfs_root *tree_root = fs_info->tree_root;
5833         struct btrfs_root_item *root_item = &root->root_item;
5834         struct walk_control *wc;
5835         struct btrfs_key key;
5836         int err = 0;
5837         int ret;
5838         int level;
5839         bool root_dropped = false;
5840         bool unfinished_drop = false;
5841
5842         btrfs_debug(fs_info, "Drop subvolume %llu", btrfs_root_id(root));
5843
5844         path = btrfs_alloc_path();
5845         if (!path) {
5846                 err = -ENOMEM;
5847                 goto out;
5848         }
5849
5850         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5851         if (!wc) {
5852                 btrfs_free_path(path);
5853                 err = -ENOMEM;
5854                 goto out;
5855         }
5856
5857         /*
5858          * Use join to avoid potential EINTR from transaction start. See
5859          * wait_reserve_ticket and the whole reservation callchain.
5860          */
5861         if (for_reloc)
5862                 trans = btrfs_join_transaction(tree_root);
5863         else
5864                 trans = btrfs_start_transaction(tree_root, 0);
5865         if (IS_ERR(trans)) {
5866                 err = PTR_ERR(trans);
5867                 goto out_free;
5868         }
5869
5870         err = btrfs_run_delayed_items(trans);
5871         if (err)
5872                 goto out_end_trans;
5873
5874         /*
5875          * This will help us catch people modifying the fs tree while we're
5876          * dropping it.  It is unsafe to mess with the fs tree while it's being
5877          * dropped as we unlock the root node and parent nodes as we walk down
5878          * the tree, assuming nothing will change.  If something does change
5879          * then we'll have stale information and drop references to blocks we've
5880          * already dropped.
5881          */
5882         set_bit(BTRFS_ROOT_DELETING, &root->state);
5883         unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
5884
5885         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5886                 level = btrfs_header_level(root->node);
5887                 path->nodes[level] = btrfs_lock_root_node(root);
5888                 path->slots[level] = 0;
5889                 path->locks[level] = BTRFS_WRITE_LOCK;
5890                 memset(&wc->update_progress, 0,
5891                        sizeof(wc->update_progress));
5892         } else {
5893                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5894                 memcpy(&wc->update_progress, &key,
5895                        sizeof(wc->update_progress));
5896
5897                 level = btrfs_root_drop_level(root_item);
5898                 BUG_ON(level == 0);
5899                 path->lowest_level = level;
5900                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5901                 path->lowest_level = 0;
5902                 if (ret < 0) {
5903                         err = ret;
5904                         goto out_end_trans;
5905                 }
5906                 WARN_ON(ret > 0);
5907
5908                 /*
5909                  * unlock our path, this is safe because only this
5910                  * function is allowed to delete this snapshot
5911                  */
5912                 btrfs_unlock_up_safe(path, 0);
5913
5914                 level = btrfs_header_level(root->node);
5915                 while (1) {
5916                         btrfs_tree_lock(path->nodes[level]);
5917                         path->locks[level] = BTRFS_WRITE_LOCK;
5918
5919                         ret = btrfs_lookup_extent_info(trans, fs_info,
5920                                                 path->nodes[level]->start,
5921                                                 level, 1, &wc->refs[level],
5922                                                 &wc->flags[level], NULL);
5923                         if (ret < 0) {
5924                                 err = ret;
5925                                 goto out_end_trans;
5926                         }
5927                         BUG_ON(wc->refs[level] == 0);
5928
5929                         if (level == btrfs_root_drop_level(root_item))
5930                                 break;
5931
5932                         btrfs_tree_unlock(path->nodes[level]);
5933                         path->locks[level] = 0;
5934                         WARN_ON(wc->refs[level] != 1);
5935                         level--;
5936                 }
5937         }
5938
5939         wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5940         wc->level = level;
5941         wc->shared_level = -1;
5942         wc->stage = DROP_REFERENCE;
5943         wc->update_ref = update_ref;
5944         wc->keep_locks = 0;
5945         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5946
5947         while (1) {
5948
5949                 ret = walk_down_tree(trans, root, path, wc);
5950                 if (ret < 0) {
5951                         btrfs_abort_transaction(trans, ret);
5952                         err = ret;
5953                         break;
5954                 }
5955
5956                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5957                 if (ret < 0) {
5958                         btrfs_abort_transaction(trans, ret);
5959                         err = ret;
5960                         break;
5961                 }
5962
5963                 if (ret > 0) {
5964                         BUG_ON(wc->stage != DROP_REFERENCE);
5965                         break;
5966                 }
5967
5968                 if (wc->stage == DROP_REFERENCE) {
5969                         wc->drop_level = wc->level;
5970                         btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5971                                               &wc->drop_progress,
5972                                               path->slots[wc->drop_level]);
5973                 }
5974                 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5975                                       &wc->drop_progress);
5976                 btrfs_set_root_drop_level(root_item, wc->drop_level);
5977
5978                 BUG_ON(wc->level == 0);
5979                 if (btrfs_should_end_transaction(trans) ||
5980                     (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5981                         ret = btrfs_update_root(trans, tree_root,
5982                                                 &root->root_key,
5983                                                 root_item);
5984                         if (ret) {
5985                                 btrfs_abort_transaction(trans, ret);
5986                                 err = ret;
5987                                 goto out_end_trans;
5988                         }
5989
5990                         if (!is_reloc_root)
5991                                 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
5992
5993                         btrfs_end_transaction_throttle(trans);
5994                         if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5995                                 btrfs_debug(fs_info,
5996                                             "drop snapshot early exit");
5997                                 err = -EAGAIN;
5998                                 goto out_free;
5999                         }
6000
6001                        /*
6002                         * Use join to avoid potential EINTR from transaction
6003                         * start. See wait_reserve_ticket and the whole
6004                         * reservation callchain.
6005                         */
6006                         if (for_reloc)
6007                                 trans = btrfs_join_transaction(tree_root);
6008                         else
6009                                 trans = btrfs_start_transaction(tree_root, 0);
6010                         if (IS_ERR(trans)) {
6011                                 err = PTR_ERR(trans);
6012                                 goto out_free;
6013                         }
6014                 }
6015         }
6016         btrfs_release_path(path);
6017         if (err)
6018                 goto out_end_trans;
6019
6020         ret = btrfs_del_root(trans, &root->root_key);
6021         if (ret) {
6022                 btrfs_abort_transaction(trans, ret);
6023                 err = ret;
6024                 goto out_end_trans;
6025         }
6026
6027         if (!is_reloc_root) {
6028                 ret = btrfs_find_root(tree_root, &root->root_key, path,
6029                                       NULL, NULL);
6030                 if (ret < 0) {
6031                         btrfs_abort_transaction(trans, ret);
6032                         err = ret;
6033                         goto out_end_trans;
6034                 } else if (ret > 0) {
6035                         /* if we fail to delete the orphan item this time
6036                          * around, it'll get picked up the next time.
6037                          *
6038                          * The most common failure here is just -ENOENT.
6039                          */
6040                         btrfs_del_orphan_item(trans, tree_root, btrfs_root_id(root));
6041                 }
6042         }
6043
6044         /*
6045          * This subvolume is going to be completely dropped, and won't be
6046          * recorded as dirty roots, thus pertrans meta rsv will not be freed at
6047          * commit transaction time.  So free it here manually.
6048          */
6049         btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
6050         btrfs_qgroup_free_meta_all_pertrans(root);
6051
6052         if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
6053                 btrfs_add_dropped_root(trans, root);
6054         else
6055                 btrfs_put_root(root);
6056         root_dropped = true;
6057 out_end_trans:
6058         if (!is_reloc_root)
6059                 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
6060
6061         btrfs_end_transaction_throttle(trans);
6062 out_free:
6063         kfree(wc);
6064         btrfs_free_path(path);
6065 out:
6066         /*
6067          * We were an unfinished drop root, check to see if there are any
6068          * pending, and if not clear and wake up any waiters.
6069          */
6070         if (!err && unfinished_drop)
6071                 btrfs_maybe_wake_unfinished_drop(fs_info);
6072
6073         /*
6074          * So if we need to stop dropping the snapshot for whatever reason we
6075          * need to make sure to add it back to the dead root list so that we
6076          * keep trying to do the work later.  This also cleans up roots if we
6077          * don't have it in the radix (like when we recover after a power fail
6078          * or unmount) so we don't leak memory.
6079          */
6080         if (!for_reloc && !root_dropped)
6081                 btrfs_add_dead_root(root);
6082         return err;
6083 }
6084
6085 /*
6086  * drop subtree rooted at tree block 'node'.
6087  *
6088  * NOTE: this function will unlock and release tree block 'node'
6089  * only used by relocation code
6090  */
6091 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6092                         struct btrfs_root *root,
6093                         struct extent_buffer *node,
6094                         struct extent_buffer *parent)
6095 {
6096         struct btrfs_fs_info *fs_info = root->fs_info;
6097         struct btrfs_path *path;
6098         struct walk_control *wc;
6099         int level;
6100         int parent_level;
6101         int ret = 0;
6102
6103         BUG_ON(btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID);
6104
6105         path = btrfs_alloc_path();
6106         if (!path)
6107                 return -ENOMEM;
6108
6109         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6110         if (!wc) {
6111                 btrfs_free_path(path);
6112                 return -ENOMEM;
6113         }
6114
6115         btrfs_assert_tree_write_locked(parent);
6116         parent_level = btrfs_header_level(parent);
6117         atomic_inc(&parent->refs);
6118         path->nodes[parent_level] = parent;
6119         path->slots[parent_level] = btrfs_header_nritems(parent);
6120
6121         btrfs_assert_tree_write_locked(node);
6122         level = btrfs_header_level(node);
6123         path->nodes[level] = node;
6124         path->slots[level] = 0;
6125         path->locks[level] = BTRFS_WRITE_LOCK;
6126
6127         wc->refs[parent_level] = 1;
6128         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6129         wc->level = level;
6130         wc->shared_level = -1;
6131         wc->stage = DROP_REFERENCE;
6132         wc->update_ref = 0;
6133         wc->keep_locks = 1;
6134         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
6135
6136         while (1) {
6137                 ret = walk_down_tree(trans, root, path, wc);
6138                 if (ret < 0)
6139                         break;
6140
6141                 ret = walk_up_tree(trans, root, path, wc, parent_level);
6142                 if (ret) {
6143                         if (ret > 0)
6144                                 ret = 0;
6145                         break;
6146                 }
6147         }
6148
6149         kfree(wc);
6150         btrfs_free_path(path);
6151         return ret;
6152 }
6153
6154 /*
6155  * Unpin the extent range in an error context and don't add the space back.
6156  * Errors are not propagated further.
6157  */
6158 void btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, u64 start, u64 end)
6159 {
6160         unpin_extent_range(fs_info, start, end, false);
6161 }
6162
6163 /*
6164  * It used to be that old block groups would be left around forever.
6165  * Iterating over them would be enough to trim unused space.  Since we
6166  * now automatically remove them, we also need to iterate over unallocated
6167  * space.
6168  *
6169  * We don't want a transaction for this since the discard may take a
6170  * substantial amount of time.  We don't require that a transaction be
6171  * running, but we do need to take a running transaction into account
6172  * to ensure that we're not discarding chunks that were released or
6173  * allocated in the current transaction.
6174  *
6175  * Holding the chunks lock will prevent other threads from allocating
6176  * or releasing chunks, but it won't prevent a running transaction
6177  * from committing and releasing the memory that the pending chunks
6178  * list head uses.  For that, we need to take a reference to the
6179  * transaction and hold the commit root sem.  We only need to hold
6180  * it while performing the free space search since we have already
6181  * held back allocations.
6182  */
6183 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
6184 {
6185         u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0;
6186         int ret;
6187
6188         *trimmed = 0;
6189
6190         /* Discard not supported = nothing to do. */
6191         if (!bdev_max_discard_sectors(device->bdev))
6192                 return 0;
6193
6194         /* Not writable = nothing to do. */
6195         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
6196                 return 0;
6197
6198         /* No free space = nothing to do. */
6199         if (device->total_bytes <= device->bytes_used)
6200                 return 0;
6201
6202         ret = 0;
6203
6204         while (1) {
6205                 struct btrfs_fs_info *fs_info = device->fs_info;
6206                 u64 bytes;
6207
6208                 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
6209                 if (ret)
6210                         break;
6211
6212                 find_first_clear_extent_bit(&device->alloc_state, start,
6213                                             &start, &end,
6214                                             CHUNK_TRIMMED | CHUNK_ALLOCATED);
6215
6216                 /* Check if there are any CHUNK_* bits left */
6217                 if (start > device->total_bytes) {
6218                         WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6219                         btrfs_warn_in_rcu(fs_info,
6220 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
6221                                           start, end - start + 1,
6222                                           btrfs_dev_name(device),
6223                                           device->total_bytes);
6224                         mutex_unlock(&fs_info->chunk_mutex);
6225                         ret = 0;
6226                         break;
6227                 }
6228
6229                 /* Ensure we skip the reserved space on each device. */
6230                 start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
6231
6232                 /*
6233                  * If find_first_clear_extent_bit find a range that spans the
6234                  * end of the device it will set end to -1, in this case it's up
6235                  * to the caller to trim the value to the size of the device.
6236                  */
6237                 end = min(end, device->total_bytes - 1);
6238
6239                 len = end - start + 1;
6240
6241                 /* We didn't find any extents */
6242                 if (!len) {
6243                         mutex_unlock(&fs_info->chunk_mutex);
6244                         ret = 0;
6245                         break;
6246                 }
6247
6248                 ret = btrfs_issue_discard(device->bdev, start, len,
6249                                           &bytes);
6250                 if (!ret)
6251                         set_extent_bit(&device->alloc_state, start,
6252                                        start + bytes - 1, CHUNK_TRIMMED, NULL);
6253                 mutex_unlock(&fs_info->chunk_mutex);
6254
6255                 if (ret)
6256                         break;
6257
6258                 start += len;
6259                 *trimmed += bytes;
6260
6261                 if (fatal_signal_pending(current)) {
6262                         ret = -ERESTARTSYS;
6263                         break;
6264                 }
6265
6266                 cond_resched();
6267         }
6268
6269         return ret;
6270 }
6271
6272 /*
6273  * Trim the whole filesystem by:
6274  * 1) trimming the free space in each block group
6275  * 2) trimming the unallocated space on each device
6276  *
6277  * This will also continue trimming even if a block group or device encounters
6278  * an error.  The return value will be the last error, or 0 if nothing bad
6279  * happens.
6280  */
6281 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6282 {
6283         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6284         struct btrfs_block_group *cache = NULL;
6285         struct btrfs_device *device;
6286         u64 group_trimmed;
6287         u64 range_end = U64_MAX;
6288         u64 start;
6289         u64 end;
6290         u64 trimmed = 0;
6291         u64 bg_failed = 0;
6292         u64 dev_failed = 0;
6293         int bg_ret = 0;
6294         int dev_ret = 0;
6295         int ret = 0;
6296
6297         if (range->start == U64_MAX)
6298                 return -EINVAL;
6299
6300         /*
6301          * Check range overflow if range->len is set.
6302          * The default range->len is U64_MAX.
6303          */
6304         if (range->len != U64_MAX &&
6305             check_add_overflow(range->start, range->len, &range_end))
6306                 return -EINVAL;
6307
6308         cache = btrfs_lookup_first_block_group(fs_info, range->start);
6309         for (; cache; cache = btrfs_next_block_group(cache)) {
6310                 if (cache->start >= range_end) {
6311                         btrfs_put_block_group(cache);
6312                         break;
6313                 }
6314
6315                 start = max(range->start, cache->start);
6316                 end = min(range_end, cache->start + cache->length);
6317
6318                 if (end - start >= range->minlen) {
6319                         if (!btrfs_block_group_done(cache)) {
6320                                 ret = btrfs_cache_block_group(cache, true);
6321                                 if (ret) {
6322                                         bg_failed++;
6323                                         bg_ret = ret;
6324                                         continue;
6325                                 }
6326                         }
6327                         ret = btrfs_trim_block_group(cache,
6328                                                      &group_trimmed,
6329                                                      start,
6330                                                      end,
6331                                                      range->minlen);
6332
6333                         trimmed += group_trimmed;
6334                         if (ret) {
6335                                 bg_failed++;
6336                                 bg_ret = ret;
6337                                 continue;
6338                         }
6339                 }
6340         }
6341
6342         if (bg_failed)
6343                 btrfs_warn(fs_info,
6344                         "failed to trim %llu block group(s), last error %d",
6345                         bg_failed, bg_ret);
6346
6347         mutex_lock(&fs_devices->device_list_mutex);
6348         list_for_each_entry(device, &fs_devices->devices, dev_list) {
6349                 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6350                         continue;
6351
6352                 ret = btrfs_trim_free_extents(device, &group_trimmed);
6353                 if (ret) {
6354                         dev_failed++;
6355                         dev_ret = ret;
6356                         break;
6357                 }
6358
6359                 trimmed += group_trimmed;
6360         }
6361         mutex_unlock(&fs_devices->device_list_mutex);
6362
6363         if (dev_failed)
6364                 btrfs_warn(fs_info,
6365                         "failed to trim %llu device(s), last error %d",
6366                         dev_failed, dev_ret);
6367         range->len = trimmed;
6368         if (bg_ret)
6369                 return bg_ret;
6370         return dev_ret;
6371 }