fs/btrfs/ctree.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007,2008 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/sched.h>
   7 #include <linux/slab.h>
   8 #include <linux/rbtree.h>
   9 #include <linux/mm.h>
  10 #include "ctree.h"
  11 #include "disk-io.h"
  12 #include "transaction.h"
  13 #include "print-tree.h"
  14 #include "locking.h"
  15 #include "volumes.h"
  16 #include "qgroup.h"
  17 #include "tree-mod-log.h"
  18
  19 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
  20                       *root, struct btrfs_path *path, int level);
  21 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  22                       const struct btrfs_key *ins_key, struct btrfs_path *path,
  23                       int data_size, int extend);
  24 static int push_node_left(struct btrfs_trans_handle *trans,
  25                           struct extent_buffer *dst,
  26                           struct extent_buffer *src, int empty);
  27 static int balance_node_right(struct btrfs_trans_handle *trans,
  28                               struct extent_buffer *dst_buf,
  29                               struct extent_buffer *src_buf);
  30 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
  31                     int level, int slot);
  32
  33 static const struct btrfs_csums {
  34         u16             size;
  35         const char      name[10];
  36         const char      driver[12];
  37 } btrfs_csums[] = {
  38         [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" },
  39         [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" },
  40         [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" },
  41         [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b",
  42                                      .driver = "blake2b-256" },
  43 };
  44
  45 int btrfs_super_csum_size(const struct btrfs_super_block *s)
  46 {
  47         u16 t = btrfs_super_csum_type(s);
  48         /*
  49          * csum type is validated at mount time
  50          */
  51         return btrfs_csums[t].size;
  52 }
  53
  54 const char *btrfs_super_csum_name(u16 csum_type)
  55 {
  56         /* csum type is validated at mount time */
  57         return btrfs_csums[csum_type].name;
  58 }
  59
  60 /*
  61  * Return driver name if defined, otherwise the name that's also a valid driver
  62  * name
  63  */
  64 const char *btrfs_super_csum_driver(u16 csum_type)
  65 {
  66         /* csum type is validated at mount time */
  67         return btrfs_csums[csum_type].driver[0] ?
  68                 btrfs_csums[csum_type].driver :
  69                 btrfs_csums[csum_type].name;
  70 }
  71
  72 size_t __attribute_const__ btrfs_get_num_csums(void)
  73 {
  74         return ARRAY_SIZE(btrfs_csums);
  75 }
  76
  77 struct btrfs_path *btrfs_alloc_path(void)
  78 {
  79         return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
  80 }
  81
  82 /* this also releases the path */
  83 void btrfs_free_path(struct btrfs_path *p)
  84 {
  85         if (!p)
  86                 return;
  87         btrfs_release_path(p);
  88         kmem_cache_free(btrfs_path_cachep, p);
  89 }
  90
  91 /*
  92  * path release drops references on the extent buffers in the path
  93  * and it drops any locks held by this path
  94  *
  95  * It is safe to call this on paths that no locks or extent buffers held.
  96  */
  97 noinline void btrfs_release_path(struct btrfs_path *p)
  98 {
  99         int i;
 100
 101         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
 102                 p->slots[i] = 0;
 103                 if (!p->nodes[i])
 104                         continue;
 105                 if (p->locks[i]) {
 106                         btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
 107                         p->locks[i] = 0;
 108                 }
 109                 free_extent_buffer(p->nodes[i]);
 110                 p->nodes[i] = NULL;
 111         }
 112 }
 113
 114 /*
 115  * safely gets a reference on the root node of a tree.  A lock
 116  * is not taken, so a concurrent writer may put a different node
 117  * at the root of the tree.  See btrfs_lock_root_node for the
 118  * looping required.
 119  *
 120  * The extent buffer returned by this has a reference taken, so
 121  * it won't disappear.  It may stop being the root of the tree
 122  * at any time because there are no locks held.
 123  */
 124 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
 125 {
 126         struct extent_buffer *eb;
 127
 128         while (1) {
 129                 rcu_read_lock();
 130                 eb = rcu_dereference(root->node);
 131
 132                 /*
 133                  * RCU really hurts here, we could free up the root node because
 134                  * it was COWed but we may not get the new root node yet so do
 135                  * the inc_not_zero dance and if it doesn't work then
 136                  * synchronize_rcu and try again.
 137                  */
 138                 if (atomic_inc_not_zero(&eb->refs)) {
 139                         rcu_read_unlock();
 140                         break;
 141                 }
 142                 rcu_read_unlock();
 143                 synchronize_rcu();
 144         }
 145         return eb;
 146 }
 147
 148 /*
 149  * Cowonly root (not-shareable trees, everything not subvolume or reloc roots),
 150  * just get put onto a simple dirty list.  Transaction walks this list to make
 151  * sure they get properly updated on disk.
 152  */
 153 static void add_root_to_dirty_list(struct btrfs_root *root)
 154 {
 155         struct btrfs_fs_info *fs_info = root->fs_info;
 156
 157         if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
 158             !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
 159                 return;
 160
 161         spin_lock(&fs_info->trans_lock);
 162         if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
 163                 /* Want the extent tree to be the last on the list */
 164                 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
 165                         list_move_tail(&root->dirty_list,
 166                                        &fs_info->dirty_cowonly_roots);
 167                 else
 168                         list_move(&root->dirty_list,
 169                                   &fs_info->dirty_cowonly_roots);
 170         }
 171         spin_unlock(&fs_info->trans_lock);
 172 }
 173
 174 /*
 175  * used by snapshot creation to make a copy of a root for a tree with
 176  * a given objectid.  The buffer with the new root node is returned in
 177  * cow_ret, and this func returns zero on success or a negative error code.
 178  */
 179 int btrfs_copy_root(struct btrfs_trans_handle *trans,
 180                       struct btrfs_root *root,
 181                       struct extent_buffer *buf,
 182                       struct extent_buffer **cow_ret, u64 new_root_objectid)
 183 {
 184         struct btrfs_fs_info *fs_info = root->fs_info;
 185         struct extent_buffer *cow;
 186         int ret = 0;
 187         int level;
 188         struct btrfs_disk_key disk_key;
 189
 190         WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
 191                 trans->transid != fs_info->running_transaction->transid);
 192         WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
 193                 trans->transid != root->last_trans);
 194
 195         level = btrfs_header_level(buf);
 196         if (level == 0)
 197                 btrfs_item_key(buf, &disk_key, 0);
 198         else
 199                 btrfs_node_key(buf, &disk_key, 0);
 200
 201         cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
 202                                      &disk_key, level, buf->start, 0,
 203                                      BTRFS_NESTING_NEW_ROOT);
 204         if (IS_ERR(cow))
 205                 return PTR_ERR(cow);
 206
 207         copy_extent_buffer_full(cow, buf);
 208         btrfs_set_header_bytenr(cow, cow->start);
 209         btrfs_set_header_generation(cow, trans->transid);
 210         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
 211         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
 212                                      BTRFS_HEADER_FLAG_RELOC);
 213         if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
 214                 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
 215         else
 216                 btrfs_set_header_owner(cow, new_root_objectid);
 217
 218         write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
 219
 220         WARN_ON(btrfs_header_generation(buf) > trans->transid);
 221         if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
 222                 ret = btrfs_inc_ref(trans, root, cow, 1);
 223         else
 224                 ret = btrfs_inc_ref(trans, root, cow, 0);
 225         if (ret) {
 226                 btrfs_tree_unlock(cow);
 227                 free_extent_buffer(cow);
 228                 btrfs_abort_transaction(trans, ret);
 229                 return ret;
 230         }
 231
 232         btrfs_mark_buffer_dirty(cow);
 233         *cow_ret = cow;
 234         return 0;
 235 }
 236
 237 /*
 238  * check if the tree block can be shared by multiple trees
 239  */
 240 int btrfs_block_can_be_shared(struct btrfs_root *root,
 241                               struct extent_buffer *buf)
 242 {
 243         /*
 244          * Tree blocks not in shareable trees and tree roots are never shared.
 245          * If a block was allocated after the last snapshot and the block was
 246          * not allocated by tree relocation, we know the block is not shared.
 247          */
 248         if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
 249             buf != root->node && buf != root->commit_root &&
 250             (btrfs_header_generation(buf) <=
 251              btrfs_root_last_snapshot(&root->root_item) ||
 252              btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
 253                 return 1;
 254
 255         return 0;
 256 }
 257
 258 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
 259                                        struct btrfs_root *root,
 260                                        struct extent_buffer *buf,
 261                                        struct extent_buffer *cow,
 262                                        int *last_ref)
 263 {
 264         struct btrfs_fs_info *fs_info = root->fs_info;
 265         u64 refs;
 266         u64 owner;
 267         u64 flags;
 268         u64 new_flags = 0;
 269         int ret;
 270
 271         /*
 272          * Backrefs update rules:
 273          *
 274          * Always use full backrefs for extent pointers in tree block
 275          * allocated by tree relocation.
 276          *
 277          * If a shared tree block is no longer referenced by its owner
 278          * tree (btrfs_header_owner(buf) == root->root_key.objectid),
 279          * use full backrefs for extent pointers in tree block.
 280          *
 281          * If a tree block is been relocating
 282          * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
 283          * use full backrefs for extent pointers in tree block.
 284          * The reason for this is some operations (such as drop tree)
 285          * are only allowed for blocks use full backrefs.
 286          */
 287
 288         if (btrfs_block_can_be_shared(root, buf)) {
 289                 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
 290                                                btrfs_header_level(buf), 1,
 291                                                &refs, &flags);
 292                 if (ret)
 293                         return ret;
 294                 if (refs == 0) {
 295                         ret = -EROFS;
 296                         btrfs_handle_fs_error(fs_info, ret, NULL);
 297                         return ret;
 298                 }
 299         } else {
 300                 refs = 1;
 301                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
 302                     btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
 303                         flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 304                 else
 305                         flags = 0;
 306         }
 307
 308         owner = btrfs_header_owner(buf);
 309         BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
 310                !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
 311
 312         if (refs > 1) {
 313                 if ((owner == root->root_key.objectid ||
 314                      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
 315                     !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
 316                         ret = btrfs_inc_ref(trans, root, buf, 1);
 317                         if (ret)
 318                                 return ret;
 319
 320                         if (root->root_key.objectid ==
 321                             BTRFS_TREE_RELOC_OBJECTID) {
 322                                 ret = btrfs_dec_ref(trans, root, buf, 0);
 323                                 if (ret)
 324                                         return ret;
 325                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 326                                 if (ret)
 327                                         return ret;
 328                         }
 329                         new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
 330                 } else {
 331
 332                         if (root->root_key.objectid ==
 333                             BTRFS_TREE_RELOC_OBJECTID)
 334                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 335                         else
 336                                 ret = btrfs_inc_ref(trans, root, cow, 0);
 337                         if (ret)
 338                                 return ret;
 339                 }
 340                 if (new_flags != 0) {
 341                         int level = btrfs_header_level(buf);
 342
 343                         ret = btrfs_set_disk_extent_flags(trans, buf,
 344                                                           new_flags, level, 0);
 345                         if (ret)
 346                                 return ret;
 347                 }
 348         } else {
 349                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
 350                         if (root->root_key.objectid ==
 351                             BTRFS_TREE_RELOC_OBJECTID)
 352                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 353                         else
 354                                 ret = btrfs_inc_ref(trans, root, cow, 0);
 355                         if (ret)
 356                                 return ret;
 357                         ret = btrfs_dec_ref(trans, root, buf, 1);
 358                         if (ret)
 359                                 return ret;
 360                 }
 361                 btrfs_clean_tree_block(buf);
 362                 *last_ref = 1;
 363         }
 364         return 0;
 365 }
 366
 367 static struct extent_buffer *alloc_tree_block_no_bg_flush(
 368                                           struct btrfs_trans_handle *trans,
 369                                           struct btrfs_root *root,
 370                                           u64 parent_start,
 371                                           const struct btrfs_disk_key *disk_key,
 372                                           int level,
 373                                           u64 hint,
 374                                           u64 empty_size,
 375                                           enum btrfs_lock_nesting nest)
 376 {
 377         struct btrfs_fs_info *fs_info = root->fs_info;
 378         struct extent_buffer *ret;
 379
 380         /*
 381          * If we are COWing a node/leaf from the extent, chunk, device or free
 382          * space trees, make sure that we do not finish block group creation of
 383          * pending block groups. We do this to avoid a deadlock.
 384          * COWing can result in allocation of a new chunk, and flushing pending
 385          * block groups (btrfs_create_pending_block_groups()) can be triggered
 386          * when finishing allocation of a new chunk. Creation of a pending block
 387          * group modifies the extent, chunk, device and free space trees,
 388          * therefore we could deadlock with ourselves since we are holding a
 389          * lock on an extent buffer that btrfs_create_pending_block_groups() may
 390          * try to COW later.
 391          * For similar reasons, we also need to delay flushing pending block
 392          * groups when splitting a leaf or node, from one of those trees, since
 393          * we are holding a write lock on it and its parent or when inserting a
 394          * new root node for one of those trees.
 395          */
 396         if (root == fs_info->extent_root ||
 397             root == fs_info->chunk_root ||
 398             root == fs_info->dev_root ||
 399             root == fs_info->free_space_root)
 400                 trans->can_flush_pending_bgs = false;
 401
 402         ret = btrfs_alloc_tree_block(trans, root, parent_start,
 403                                      root->root_key.objectid, disk_key, level,
 404                                      hint, empty_size, nest);
 405         trans->can_flush_pending_bgs = true;
 406
 407         return ret;
 408 }
 409
 410 /*
 411  * does the dirty work in cow of a single block.  The parent block (if
 412  * supplied) is updated to point to the new cow copy.  The new buffer is marked
 413  * dirty and returned locked.  If you modify the block it needs to be marked
 414  * dirty again.
 415  *
 416  * search_start -- an allocation hint for the new block
 417  *
 418  * empty_size -- a hint that you plan on doing more cow.  This is the size in
 419  * bytes the allocator should try to find free next to the block it returns.
 420  * This is just a hint and may be ignored by the allocator.
 421  */
 422 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 423                              struct btrfs_root *root,
 424                              struct extent_buffer *buf,
 425                              struct extent_buffer *parent, int parent_slot,
 426                              struct extent_buffer **cow_ret,
 427                              u64 search_start, u64 empty_size,
 428                              enum btrfs_lock_nesting nest)
 429 {
 430         struct btrfs_fs_info *fs_info = root->fs_info;
 431         struct btrfs_disk_key disk_key;
 432         struct extent_buffer *cow;
 433         int level, ret;
 434         int last_ref = 0;
 435         int unlock_orig = 0;
 436         u64 parent_start = 0;
 437
 438         if (*cow_ret == buf)
 439                 unlock_orig = 1;
 440
 441         btrfs_assert_tree_locked(buf);
 442
 443         WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
 444                 trans->transid != fs_info->running_transaction->transid);
 445         WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
 446                 trans->transid != root->last_trans);
 447
 448         level = btrfs_header_level(buf);
 449
 450         if (level == 0)
 451                 btrfs_item_key(buf, &disk_key, 0);
 452         else
 453                 btrfs_node_key(buf, &disk_key, 0);
 454
 455         if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
 456                 parent_start = parent->start;
 457
 458         cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key,
 459                                            level, search_start, empty_size, nest);
 460         if (IS_ERR(cow))
 461                 return PTR_ERR(cow);
 462
 463         /* cow is set to blocking by btrfs_init_new_buffer */
 464
 465         copy_extent_buffer_full(cow, buf);
 466         btrfs_set_header_bytenr(cow, cow->start);
 467         btrfs_set_header_generation(cow, trans->transid);
 468         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
 469         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
 470                                      BTRFS_HEADER_FLAG_RELOC);
 471         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
 472                 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
 473         else
 474                 btrfs_set_header_owner(cow, root->root_key.objectid);
 475
 476         write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
 477
 478         ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
 479         if (ret) {
 480                 btrfs_tree_unlock(cow);
 481                 free_extent_buffer(cow);
 482                 btrfs_abort_transaction(trans, ret);
 483                 return ret;
 484         }
 485
 486         if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
 487                 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
 488                 if (ret) {
 489                         btrfs_tree_unlock(cow);
 490                         free_extent_buffer(cow);
 491                         btrfs_abort_transaction(trans, ret);
 492                         return ret;
 493                 }
 494         }
 495
 496         if (buf == root->node) {
 497                 WARN_ON(parent && parent != buf);
 498                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
 499                     btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
 500                         parent_start = buf->start;
 501
 502                 atomic_inc(&cow->refs);
 503                 ret = btrfs_tree_mod_log_insert_root(root->node, cow, true);
 504                 BUG_ON(ret < 0);
 505                 rcu_assign_pointer(root->node, cow);
 506
 507                 btrfs_free_tree_block(trans, root, buf, parent_start,
 508                                       last_ref);
 509                 free_extent_buffer(buf);
 510                 add_root_to_dirty_list(root);
 511         } else {
 512                 WARN_ON(trans->transid != btrfs_header_generation(parent));
 513                 btrfs_tree_mod_log_insert_key(parent, parent_slot,
 514                                               BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS);
 515                 btrfs_set_node_blockptr(parent, parent_slot,
 516                                         cow->start);
 517                 btrfs_set_node_ptr_generation(parent, parent_slot,
 518                                               trans->transid);
 519                 btrfs_mark_buffer_dirty(parent);
 520                 if (last_ref) {
 521                         ret = btrfs_tree_mod_log_free_eb(buf);
 522                         if (ret) {
 523                                 btrfs_tree_unlock(cow);
 524                                 free_extent_buffer(cow);
 525                                 btrfs_abort_transaction(trans, ret);
 526                                 return ret;
 527                         }
 528                 }
 529                 btrfs_free_tree_block(trans, root, buf, parent_start,
 530                                       last_ref);
 531         }
 532         if (unlock_orig)
 533                 btrfs_tree_unlock(buf);
 534         free_extent_buffer_stale(buf);
 535         btrfs_mark_buffer_dirty(cow);
 536         *cow_ret = cow;
 537         return 0;
 538 }
 539
 540 static inline int should_cow_block(struct btrfs_trans_handle *trans,
 541                                    struct btrfs_root *root,
 542                                    struct extent_buffer *buf)
 543 {
 544         if (btrfs_is_testing(root->fs_info))
 545                 return 0;
 546
 547         /* Ensure we can see the FORCE_COW bit */
 548         smp_mb__before_atomic();
 549
 550         /*
 551          * We do not need to cow a block if
 552          * 1) this block is not created or changed in this transaction;
 553          * 2) this block does not belong to TREE_RELOC tree;
 554          * 3) the root is not forced COW.
 555          *
 556          * What is forced COW:
 557          *    when we create snapshot during committing the transaction,
 558          *    after we've finished copying src root, we must COW the shared
 559          *    block to ensure the metadata consistency.
 560          */
 561         if (btrfs_header_generation(buf) == trans->transid &&
 562             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
 563             !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
 564               btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
 565             !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
 566                 return 0;
 567         return 1;
 568 }
 569
 570 /*
 571  * cows a single block, see __btrfs_cow_block for the real work.
 572  * This version of it has extra checks so that a block isn't COWed more than
 573  * once per transaction, as long as it hasn't been written yet
 574  */
 575 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
 576                     struct btrfs_root *root, struct extent_buffer *buf,
 577                     struct extent_buffer *parent, int parent_slot,
 578                     struct extent_buffer **cow_ret,
 579                     enum btrfs_lock_nesting nest)
 580 {
 581         struct btrfs_fs_info *fs_info = root->fs_info;
 582         u64 search_start;
 583         int ret;
 584
 585         if (test_bit(BTRFS_ROOT_DELETING, &root->state))
 586                 btrfs_err(fs_info,
 587                         "COW'ing blocks on a fs root that's being dropped");
 588
 589         if (trans->transaction != fs_info->running_transaction)
 590                 WARN(1, KERN_CRIT "trans %llu running %llu\n",
 591                        trans->transid,
 592                        fs_info->running_transaction->transid);
 593
 594         if (trans->transid != fs_info->generation)
 595                 WARN(1, KERN_CRIT "trans %llu running %llu\n",
 596                        trans->transid, fs_info->generation);
 597
 598         if (!should_cow_block(trans, root, buf)) {
 599                 *cow_ret = buf;
 600                 return 0;
 601         }
 602
 603         search_start = buf->start & ~((u64)SZ_1G - 1);
 604
 605         /*
 606          * Before CoWing this block for later modification, check if it's
 607          * the subtree root and do the delayed subtree trace if needed.
 608          *
 609          * Also We don't care about the error, as it's handled internally.
 610          */
 611         btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
 612         ret = __btrfs_cow_block(trans, root, buf, parent,
 613                                  parent_slot, cow_ret, search_start, 0, nest);
 614
 615         trace_btrfs_cow_block(root, buf, *cow_ret);
 616
 617         return ret;
 618 }
 619 ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO);
 620
 621 /*
 622  * helper function for defrag to decide if two blocks pointed to by a
 623  * node are actually close by
 624  */
 625 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
 626 {
 627         if (blocknr < other && other - (blocknr + blocksize) < 32768)
 628                 return 1;
 629         if (blocknr > other && blocknr - (other + blocksize) < 32768)
 630                 return 1;
 631         return 0;
 632 }
 633
 634 #ifdef __LITTLE_ENDIAN
 635
 636 /*
 637  * Compare two keys, on little-endian the disk order is same as CPU order and
 638  * we can avoid the conversion.
 639  */
 640 static int comp_keys(const struct btrfs_disk_key *disk_key,
 641                      const struct btrfs_key *k2)
 642 {
 643         const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key;
 644
 645         return btrfs_comp_cpu_keys(k1, k2);
 646 }
 647
 648 #else
 649
 650 /*
 651  * compare two keys in a memcmp fashion
 652  */
 653 static int comp_keys(const struct btrfs_disk_key *disk,
 654                      const struct btrfs_key *k2)
 655 {
 656         struct btrfs_key k1;
 657
 658         btrfs_disk_key_to_cpu(&k1, disk);
 659
 660         return btrfs_comp_cpu_keys(&k1, k2);
 661 }
 662 #endif
 663
 664 /*
 665  * same as comp_keys only with two btrfs_key's
 666  */
 667 int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
 668 {
 669         if (k1->objectid > k2->objectid)
 670                 return 1;
 671         if (k1->objectid < k2->objectid)
 672                 return -1;
 673         if (k1->type > k2->type)
 674                 return 1;
 675         if (k1->type < k2->type)
 676                 return -1;
 677         if (k1->offset > k2->offset)
 678                 return 1;
 679         if (k1->offset < k2->offset)
 680                 return -1;
 681         return 0;
 682 }
 683
 684 /*
 685  * this is used by the defrag code to go through all the
 686  * leaves pointed to by a node and reallocate them so that
 687  * disk order is close to key order
 688  */
 689 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
 690                        struct btrfs_root *root, struct extent_buffer *parent,
 691                        int start_slot, u64 *last_ret,
 692                        struct btrfs_key *progress)
 693 {
 694         struct btrfs_fs_info *fs_info = root->fs_info;
 695         struct extent_buffer *cur;
 696         u64 blocknr;
 697         u64 search_start = *last_ret;
 698         u64 last_block = 0;
 699         u64 other;
 700         u32 parent_nritems;
 701         int end_slot;
 702         int i;
 703         int err = 0;
 704         u32 blocksize;
 705         int progress_passed = 0;
 706         struct btrfs_disk_key disk_key;
 707
 708         WARN_ON(trans->transaction != fs_info->running_transaction);
 709         WARN_ON(trans->transid != fs_info->generation);
 710
 711         parent_nritems = btrfs_header_nritems(parent);
 712         blocksize = fs_info->nodesize;
 713         end_slot = parent_nritems - 1;
 714
 715         if (parent_nritems <= 1)
 716                 return 0;
 717
 718         for (i = start_slot; i <= end_slot; i++) {
 719                 int close = 1;
 720
 721                 btrfs_node_key(parent, &disk_key, i);
 722                 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
 723                         continue;
 724
 725                 progress_passed = 1;
 726                 blocknr = btrfs_node_blockptr(parent, i);
 727                 if (last_block == 0)
 728                         last_block = blocknr;
 729
 730                 if (i > 0) {
 731                         other = btrfs_node_blockptr(parent, i - 1);
 732                         close = close_blocks(blocknr, other, blocksize);
 733                 }
 734                 if (!close && i < end_slot) {
 735                         other = btrfs_node_blockptr(parent, i + 1);
 736                         close = close_blocks(blocknr, other, blocksize);
 737                 }
 738                 if (close) {
 739                         last_block = blocknr;
 740                         continue;
 741                 }
 742
 743                 cur = btrfs_read_node_slot(parent, i);
 744                 if (IS_ERR(cur))
 745                         return PTR_ERR(cur);
 746                 if (search_start == 0)
 747                         search_start = last_block;
 748
 749                 btrfs_tree_lock(cur);
 750                 err = __btrfs_cow_block(trans, root, cur, parent, i,
 751                                         &cur, search_start,
 752                                         min(16 * blocksize,
 753                                             (end_slot - i) * blocksize),
 754                                         BTRFS_NESTING_COW);
 755                 if (err) {
 756                         btrfs_tree_unlock(cur);
 757                         free_extent_buffer(cur);
 758                         break;
 759                 }
 760                 search_start = cur->start;
 761                 last_block = cur->start;
 762                 *last_ret = search_start;
 763                 btrfs_tree_unlock(cur);
 764                 free_extent_buffer(cur);
 765         }
 766         return err;
 767 }
 768
 769 /*
 770  * search for key in the extent_buffer.  The items start at offset p,
 771  * and they are item_size apart.  There are 'max' items in p.
 772  *
 773  * the slot in the array is returned via slot, and it points to
 774  * the place where you would insert key if it is not found in
 775  * the array.
 776  *
 777  * slot may point to max if the key is bigger than all of the keys
 778  */
 779 static noinline int generic_bin_search(struct extent_buffer *eb,
 780                                        unsigned long p, int item_size,
 781                                        const struct btrfs_key *key,
 782                                        int max, int *slot)
 783 {
 784         int low = 0;
 785         int high = max;
 786         int ret;
 787         const int key_size = sizeof(struct btrfs_disk_key);
 788
 789         if (low > high) {
 790                 btrfs_err(eb->fs_info,
 791                  "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
 792                           __func__, low, high, eb->start,
 793                           btrfs_header_owner(eb), btrfs_header_level(eb));
 794                 return -EINVAL;
 795         }
 796
 797         while (low < high) {
 798                 unsigned long oip;
 799                 unsigned long offset;
 800                 struct btrfs_disk_key *tmp;
 801                 struct btrfs_disk_key unaligned;
 802                 int mid;
 803
 804                 mid = (low + high) / 2;
 805                 offset = p + mid * item_size;
 806                 oip = offset_in_page(offset);
 807
 808                 if (oip + key_size <= PAGE_SIZE) {
 809                         const unsigned long idx = get_eb_page_index(offset);
 810                         char *kaddr = page_address(eb->pages[idx]);
 811
 812                         oip = get_eb_offset_in_page(eb, offset);
 813                         tmp = (struct btrfs_disk_key *)(kaddr + oip);
 814                 } else {
 815                         read_extent_buffer(eb, &unaligned, offset, key_size);
 816                         tmp = &unaligned;
 817                 }
 818
 819                 ret = comp_keys(tmp, key);
 820
 821                 if (ret < 0)
 822                         low = mid + 1;
 823                 else if (ret > 0)
 824                         high = mid;
 825                 else {
 826                         *slot = mid;
 827                         return 0;
 828                 }
 829         }
 830         *slot = low;
 831         return 1;
 832 }
 833
 834 /*
 835  * simple bin_search frontend that does the right thing for
 836  * leaves vs nodes
 837  */
 838 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
 839                      int *slot)
 840 {
 841         if (btrfs_header_level(eb) == 0)
 842                 return generic_bin_search(eb,
 843                                           offsetof(struct btrfs_leaf, items),
 844                                           sizeof(struct btrfs_item),
 845                                           key, btrfs_header_nritems(eb),
 846                                           slot);
 847         else
 848                 return generic_bin_search(eb,
 849                                           offsetof(struct btrfs_node, ptrs),
 850                                           sizeof(struct btrfs_key_ptr),
 851                                           key, btrfs_header_nritems(eb),
 852                                           slot);
 853 }
 854
 855 static void root_add_used(struct btrfs_root *root, u32 size)
 856 {
 857         spin_lock(&root->accounting_lock);
 858         btrfs_set_root_used(&root->root_item,
 859                             btrfs_root_used(&root->root_item) + size);
 860         spin_unlock(&root->accounting_lock);
 861 }
 862
 863 static void root_sub_used(struct btrfs_root *root, u32 size)
 864 {
 865         spin_lock(&root->accounting_lock);
 866         btrfs_set_root_used(&root->root_item,
 867                             btrfs_root_used(&root->root_item) - size);
 868         spin_unlock(&root->accounting_lock);
 869 }
 870
 871 /* given a node and slot number, this reads the blocks it points to.  The
 872  * extent buffer is returned with a reference taken (but unlocked).
 873  */
 874 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
 875                                            int slot)
 876 {
 877         int level = btrfs_header_level(parent);
 878         struct extent_buffer *eb;
 879         struct btrfs_key first_key;
 880
 881         if (slot < 0 || slot >= btrfs_header_nritems(parent))
 882                 return ERR_PTR(-ENOENT);
 883
 884         BUG_ON(level == 0);
 885
 886         btrfs_node_key_to_cpu(parent, &first_key, slot);
 887         eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
 888                              btrfs_header_owner(parent),
 889                              btrfs_node_ptr_generation(parent, slot),
 890                              level - 1, &first_key);
 891         if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
 892                 free_extent_buffer(eb);
 893                 eb = ERR_PTR(-EIO);
 894         }
 895
 896         return eb;
 897 }
 898
 899 /*
 900  * node level balancing, used to make sure nodes are in proper order for
 901  * item deletion.  We balance from the top down, so we have to make sure
 902  * that a deletion won't leave an node completely empty later on.
 903  */
 904 static noinline int balance_level(struct btrfs_trans_handle *trans,
 905                          struct btrfs_root *root,
 906                          struct btrfs_path *path, int level)
 907 {
 908         struct btrfs_fs_info *fs_info = root->fs_info;
 909         struct extent_buffer *right = NULL;
 910         struct extent_buffer *mid;
 911         struct extent_buffer *left = NULL;
 912         struct extent_buffer *parent = NULL;
 913         int ret = 0;
 914         int wret;
 915         int pslot;
 916         int orig_slot = path->slots[level];
 917         u64 orig_ptr;
 918
 919         ASSERT(level > 0);
 920
 921         mid = path->nodes[level];
 922
 923         WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK);
 924         WARN_ON(btrfs_header_generation(mid) != trans->transid);
 925
 926         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
 927
 928         if (level < BTRFS_MAX_LEVEL - 1) {
 929                 parent = path->nodes[level + 1];
 930                 pslot = path->slots[level + 1];
 931         }
 932
 933         /*
 934          * deal with the case where there is only one pointer in the root
 935          * by promoting the node below to a root
 936          */
 937         if (!parent) {
 938                 struct extent_buffer *child;
 939
 940                 if (btrfs_header_nritems(mid) != 1)
 941                         return 0;
 942
 943                 /* promote the child to a root */
 944                 child = btrfs_read_node_slot(mid, 0);
 945                 if (IS_ERR(child)) {
 946                         ret = PTR_ERR(child);
 947                         btrfs_handle_fs_error(fs_info, ret, NULL);
 948                         goto enospc;
 949                 }
 950
 951                 btrfs_tree_lock(child);
 952                 ret = btrfs_cow_block(trans, root, child, mid, 0, &child,
 953                                       BTRFS_NESTING_COW);
 954                 if (ret) {
 955                         btrfs_tree_unlock(child);
 956                         free_extent_buffer(child);
 957                         goto enospc;
 958                 }
 959
 960                 ret = btrfs_tree_mod_log_insert_root(root->node, child, true);
 961                 BUG_ON(ret < 0);
 962                 rcu_assign_pointer(root->node, child);
 963
 964                 add_root_to_dirty_list(root);
 965                 btrfs_tree_unlock(child);
 966
 967                 path->locks[level] = 0;
 968                 path->nodes[level] = NULL;
 969                 btrfs_clean_tree_block(mid);
 970                 btrfs_tree_unlock(mid);
 971                 /* once for the path */
 972                 free_extent_buffer(mid);
 973
 974                 root_sub_used(root, mid->len);
 975                 btrfs_free_tree_block(trans, root, mid, 0, 1);
 976                 /* once for the root ptr */
 977                 free_extent_buffer_stale(mid);
 978                 return 0;
 979         }
 980         if (btrfs_header_nritems(mid) >
 981             BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
 982                 return 0;
 983
 984         left = btrfs_read_node_slot(parent, pslot - 1);
 985         if (IS_ERR(left))
 986                 left = NULL;
 987
 988         if (left) {
 989                 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
 990                 wret = btrfs_cow_block(trans, root, left,
 991                                        parent, pslot - 1, &left,
 992                                        BTRFS_NESTING_LEFT_COW);
 993                 if (wret) {
 994                         ret = wret;
 995                         goto enospc;
 996                 }
 997         }
 998
 999         right = btrfs_read_node_slot(parent, pslot + 1);
1000         if (IS_ERR(right))
1001                 right = NULL;
1002
1003         if (right) {
1004                 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
1005                 wret = btrfs_cow_block(trans, root, right,
1006                                        parent, pslot + 1, &right,
1007                                        BTRFS_NESTING_RIGHT_COW);
1008                 if (wret) {
1009                         ret = wret;
1010                         goto enospc;
1011                 }
1012         }
1013
1014         /* first, try to make some room in the middle buffer */
1015         if (left) {
1016                 orig_slot += btrfs_header_nritems(left);
1017                 wret = push_node_left(trans, left, mid, 1);
1018                 if (wret < 0)
1019                         ret = wret;
1020         }
1021
1022         /*
1023          * then try to empty the right most buffer into the middle
1024          */
1025         if (right) {
1026                 wret = push_node_left(trans, mid, right, 1);
1027                 if (wret < 0 && wret != -ENOSPC)
1028                         ret = wret;
1029                 if (btrfs_header_nritems(right) == 0) {
1030                         btrfs_clean_tree_block(right);
1031                         btrfs_tree_unlock(right);
1032                         del_ptr(root, path, level + 1, pslot + 1);
1033                         root_sub_used(root, right->len);
1034                         btrfs_free_tree_block(trans, root, right, 0, 1);
1035                         free_extent_buffer_stale(right);
1036                         right = NULL;
1037                 } else {
1038                         struct btrfs_disk_key right_key;
1039                         btrfs_node_key(right, &right_key, 0);
1040                         ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1,
1041                                         BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS);
1042                         BUG_ON(ret < 0);
1043                         btrfs_set_node_key(parent, &right_key, pslot + 1);
1044                         btrfs_mark_buffer_dirty(parent);
1045                 }
1046         }
1047         if (btrfs_header_nritems(mid) == 1) {
1048                 /*
1049                  * we're not allowed to leave a node with one item in the
1050                  * tree during a delete.  A deletion from lower in the tree
1051                  * could try to delete the only pointer in this node.
1052                  * So, pull some keys from the left.
1053                  * There has to be a left pointer at this point because
1054                  * otherwise we would have pulled some pointers from the
1055                  * right
1056                  */
1057                 if (!left) {
1058                         ret = -EROFS;
1059                         btrfs_handle_fs_error(fs_info, ret, NULL);
1060                         goto enospc;
1061                 }
1062                 wret = balance_node_right(trans, mid, left);
1063                 if (wret < 0) {
1064                         ret = wret;
1065                         goto enospc;
1066                 }
1067                 if (wret == 1) {
1068                         wret = push_node_left(trans, left, mid, 1);
1069                         if (wret < 0)
1070                                 ret = wret;
1071                 }
1072                 BUG_ON(wret == 1);
1073         }
1074         if (btrfs_header_nritems(mid) == 0) {
1075                 btrfs_clean_tree_block(mid);
1076                 btrfs_tree_unlock(mid);
1077                 del_ptr(root, path, level + 1, pslot);
1078                 root_sub_used(root, mid->len);
1079                 btrfs_free_tree_block(trans, root, mid, 0, 1);
1080                 free_extent_buffer_stale(mid);
1081                 mid = NULL;
1082         } else {
1083                 /* update the parent key to reflect our changes */
1084                 struct btrfs_disk_key mid_key;
1085                 btrfs_node_key(mid, &mid_key, 0);
1086                 ret = btrfs_tree_mod_log_insert_key(parent, pslot,
1087                                 BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS);
1088                 BUG_ON(ret < 0);
1089                 btrfs_set_node_key(parent, &mid_key, pslot);
1090                 btrfs_mark_buffer_dirty(parent);
1091         }
1092
1093         /* update the path */
1094         if (left) {
1095                 if (btrfs_header_nritems(left) > orig_slot) {
1096                         atomic_inc(&left->refs);
1097                         /* left was locked after cow */
1098                         path->nodes[level] = left;
1099                         path->slots[level + 1] -= 1;
1100                         path->slots[level] = orig_slot;
1101                         if (mid) {
1102                                 btrfs_tree_unlock(mid);
1103                                 free_extent_buffer(mid);
1104                         }
1105                 } else {
1106                         orig_slot -= btrfs_header_nritems(left);
1107                         path->slots[level] = orig_slot;
1108                 }
1109         }
1110         /* double check we haven't messed things up */
1111         if (orig_ptr !=
1112             btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1113                 BUG();
1114 enospc:
1115         if (right) {
1116                 btrfs_tree_unlock(right);
1117                 free_extent_buffer(right);
1118         }
1119         if (left) {
1120                 if (path->nodes[level] != left)
1121                         btrfs_tree_unlock(left);
1122                 free_extent_buffer(left);
1123         }
1124         return ret;
1125 }
1126
1127 /* Node balancing for insertion.  Here we only split or push nodes around
1128  * when they are completely full.  This is also done top down, so we
1129  * have to be pessimistic.
1130  */
1131 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1132                                           struct btrfs_root *root,
1133                                           struct btrfs_path *path, int level)
1134 {
1135         struct btrfs_fs_info *fs_info = root->fs_info;
1136         struct extent_buffer *right = NULL;
1137         struct extent_buffer *mid;
1138         struct extent_buffer *left = NULL;
1139         struct extent_buffer *parent = NULL;
1140         int ret = 0;
1141         int wret;
1142         int pslot;
1143         int orig_slot = path->slots[level];
1144
1145         if (level == 0)
1146                 return 1;
1147
1148         mid = path->nodes[level];
1149         WARN_ON(btrfs_header_generation(mid) != trans->transid);
1150
1151         if (level < BTRFS_MAX_LEVEL - 1) {
1152                 parent = path->nodes[level + 1];
1153                 pslot = path->slots[level + 1];
1154         }
1155
1156         if (!parent)
1157                 return 1;
1158
1159         left = btrfs_read_node_slot(parent, pslot - 1);
1160         if (IS_ERR(left))
1161                 left = NULL;
1162
1163         /* first, try to make some room in the middle buffer */
1164         if (left) {
1165                 u32 left_nr;
1166
1167                 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
1168
1169                 left_nr = btrfs_header_nritems(left);
1170                 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
1171                         wret = 1;
1172                 } else {
1173                         ret = btrfs_cow_block(trans, root, left, parent,
1174                                               pslot - 1, &left,
1175                                               BTRFS_NESTING_LEFT_COW);
1176                         if (ret)
1177                                 wret = 1;
1178                         else {
1179                                 wret = push_node_left(trans, left, mid, 0);
1180                         }
1181                 }
1182                 if (wret < 0)
1183                         ret = wret;
1184                 if (wret == 0) {
1185                         struct btrfs_disk_key disk_key;
1186                         orig_slot += left_nr;
1187                         btrfs_node_key(mid, &disk_key, 0);
1188                         ret = btrfs_tree_mod_log_insert_key(parent, pslot,
1189                                         BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS);
1190                         BUG_ON(ret < 0);
1191                         btrfs_set_node_key(parent, &disk_key, pslot);
1192                         btrfs_mark_buffer_dirty(parent);
1193                         if (btrfs_header_nritems(left) > orig_slot) {
1194                                 path->nodes[level] = left;
1195                                 path->slots[level + 1] -= 1;
1196                                 path->slots[level] = orig_slot;
1197                                 btrfs_tree_unlock(mid);
1198                                 free_extent_buffer(mid);
1199                         } else {
1200                                 orig_slot -=
1201                                         btrfs_header_nritems(left);
1202                                 path->slots[level] = orig_slot;
1203                                 btrfs_tree_unlock(left);
1204                                 free_extent_buffer(left);
1205                         }
1206                         return 0;
1207                 }
1208                 btrfs_tree_unlock(left);
1209                 free_extent_buffer(left);
1210         }
1211         right = btrfs_read_node_slot(parent, pslot + 1);
1212         if (IS_ERR(right))
1213                 right = NULL;
1214
1215         /*
1216          * then try to empty the right most buffer into the middle
1217          */
1218         if (right) {
1219                 u32 right_nr;
1220
1221                 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
1222
1223                 right_nr = btrfs_header_nritems(right);
1224                 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
1225                         wret = 1;
1226                 } else {
1227                         ret = btrfs_cow_block(trans, root, right,
1228                                               parent, pslot + 1,
1229                                               &right, BTRFS_NESTING_RIGHT_COW);
1230                         if (ret)
1231                                 wret = 1;
1232                         else {
1233                                 wret = balance_node_right(trans, right, mid);
1234                         }
1235                 }
1236                 if (wret < 0)
1237                         ret = wret;
1238                 if (wret == 0) {
1239                         struct btrfs_disk_key disk_key;
1240
1241                         btrfs_node_key(right, &disk_key, 0);
1242                         ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1,
1243                                         BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS);
1244                         BUG_ON(ret < 0);
1245                         btrfs_set_node_key(parent, &disk_key, pslot + 1);
1246                         btrfs_mark_buffer_dirty(parent);
1247
1248                         if (btrfs_header_nritems(mid) <= orig_slot) {
1249                                 path->nodes[level] = right;
1250                                 path->slots[level + 1] += 1;
1251                                 path->slots[level] = orig_slot -
1252                                         btrfs_header_nritems(mid);
1253                                 btrfs_tree_unlock(mid);
1254                                 free_extent_buffer(mid);
1255                         } else {
1256                                 btrfs_tree_unlock(right);
1257                                 free_extent_buffer(right);
1258                         }
1259                         return 0;
1260                 }
1261                 btrfs_tree_unlock(right);
1262                 free_extent_buffer(right);
1263         }
1264         return 1;
1265 }
1266
1267 /*
1268  * readahead one full node of leaves, finding things that are close
1269  * to the block in 'slot', and triggering ra on them.
1270  */
1271 static void reada_for_search(struct btrfs_fs_info *fs_info,
1272                              struct btrfs_path *path,
1273                              int level, int slot, u64 objectid)
1274 {
1275         struct extent_buffer *node;
1276         struct btrfs_disk_key disk_key;
1277         u32 nritems;
1278         u64 search;
1279         u64 target;
1280         u64 nread = 0;
1281         u64 nread_max;
1282         struct extent_buffer *eb;
1283         u32 nr;
1284         u32 blocksize;
1285         u32 nscan = 0;
1286
1287         if (level != 1 && path->reada != READA_FORWARD_ALWAYS)
1288                 return;
1289
1290         if (!path->nodes[level])
1291                 return;
1292
1293         node = path->nodes[level];
1294
1295         /*
1296          * Since the time between visiting leaves is much shorter than the time
1297          * between visiting nodes, limit read ahead of nodes to 1, to avoid too
1298          * much IO at once (possibly random).
1299          */
1300         if (path->reada == READA_FORWARD_ALWAYS) {
1301                 if (level > 1)
1302                         nread_max = node->fs_info->nodesize;
1303                 else
1304                         nread_max = SZ_128K;
1305         } else {
1306                 nread_max = SZ_64K;
1307         }
1308
1309         search = btrfs_node_blockptr(node, slot);
1310         blocksize = fs_info->nodesize;
1311         eb = find_extent_buffer(fs_info, search);
1312         if (eb) {
1313                 free_extent_buffer(eb);
1314                 return;
1315         }
1316
1317         target = search;
1318
1319         nritems = btrfs_header_nritems(node);
1320         nr = slot;
1321
1322         while (1) {
1323                 if (path->reada == READA_BACK) {
1324                         if (nr == 0)
1325                                 break;
1326                         nr--;
1327                 } else if (path->reada == READA_FORWARD ||
1328                            path->reada == READA_FORWARD_ALWAYS) {
1329                         nr++;
1330                         if (nr >= nritems)
1331                                 break;
1332                 }
1333                 if (path->reada == READA_BACK && objectid) {
1334                         btrfs_node_key(node, &disk_key, nr);
1335                         if (btrfs_disk_key_objectid(&disk_key) != objectid)
1336                                 break;
1337                 }
1338                 search = btrfs_node_blockptr(node, nr);
1339                 if (path->reada == READA_FORWARD_ALWAYS ||
1340                     (search <= target && target - search <= 65536) ||
1341                     (search > target && search - target <= 65536)) {
1342                         btrfs_readahead_node_child(node, nr);
1343                         nread += blocksize;
1344                 }
1345                 nscan++;
1346                 if (nread > nread_max || nscan > 32)
1347                         break;
1348         }
1349 }
1350
1351 static noinline void reada_for_balance(struct btrfs_path *path, int level)
1352 {
1353         struct extent_buffer *parent;
1354         int slot;
1355         int nritems;
1356
1357         parent = path->nodes[level + 1];
1358         if (!parent)
1359                 return;
1360
1361         nritems = btrfs_header_nritems(parent);
1362         slot = path->slots[level + 1];
1363
1364         if (slot > 0)
1365                 btrfs_readahead_node_child(parent, slot - 1);
1366         if (slot + 1 < nritems)
1367                 btrfs_readahead_node_child(parent, slot + 1);
1368 }
1369
1370
1371 /*
1372  * when we walk down the tree, it is usually safe to unlock the higher layers
1373  * in the tree.  The exceptions are when our path goes through slot 0, because
1374  * operations on the tree might require changing key pointers higher up in the
1375  * tree.
1376  *
1377  * callers might also have set path->keep_locks, which tells this code to keep
1378  * the lock if the path points to the last slot in the block.  This is part of
1379  * walking through the tree, and selecting the next slot in the higher block.
1380  *
1381  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
1382  * if lowest_unlock is 1, level 0 won't be unlocked
1383  */
1384 static noinline void unlock_up(struct btrfs_path *path, int level,
1385                                int lowest_unlock, int min_write_lock_level,
1386                                int *write_lock_level)
1387 {
1388         int i;
1389         int skip_level = level;
1390         int no_skips = 0;
1391         struct extent_buffer *t;
1392
1393         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1394                 if (!path->nodes[i])
1395                         break;
1396                 if (!path->locks[i])
1397                         break;
1398                 if (!no_skips && path->slots[i] == 0) {
1399                         skip_level = i + 1;
1400                         continue;
1401                 }
1402                 if (!no_skips && path->keep_locks) {
1403                         u32 nritems;
1404                         t = path->nodes[i];
1405                         nritems = btrfs_header_nritems(t);
1406                         if (nritems < 1 || path->slots[i] >= nritems - 1) {
1407                                 skip_level = i + 1;
1408                                 continue;
1409                         }
1410                 }
1411                 if (skip_level < i && i >= lowest_unlock)
1412                         no_skips = 1;
1413
1414                 t = path->nodes[i];
1415                 if (i >= lowest_unlock && i > skip_level) {
1416                         btrfs_tree_unlock_rw(t, path->locks[i]);
1417                         path->locks[i] = 0;
1418                         if (write_lock_level &&
1419                             i > min_write_lock_level &&
1420                             i <= *write_lock_level) {
1421                                 *write_lock_level = i - 1;
1422                         }
1423                 }
1424         }
1425 }
1426
1427 /*
1428  * helper function for btrfs_search_slot.  The goal is to find a block
1429  * in cache without setting the path to blocking.  If we find the block
1430  * we return zero and the path is unchanged.
1431  *
1432  * If we can't find the block, we set the path blocking and do some
1433  * reada.  -EAGAIN is returned and the search must be repeated.
1434  */
1435 static int
1436 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
1437                       struct extent_buffer **eb_ret, int level, int slot,
1438                       const struct btrfs_key *key)
1439 {
1440         struct btrfs_fs_info *fs_info = root->fs_info;
1441         u64 blocknr;
1442         u64 gen;
1443         struct extent_buffer *tmp;
1444         struct btrfs_key first_key;
1445         int ret;
1446         int parent_level;
1447
1448         blocknr = btrfs_node_blockptr(*eb_ret, slot);
1449         gen = btrfs_node_ptr_generation(*eb_ret, slot);
1450         parent_level = btrfs_header_level(*eb_ret);
1451         btrfs_node_key_to_cpu(*eb_ret, &first_key, slot);
1452
1453         tmp = find_extent_buffer(fs_info, blocknr);
1454         if (tmp) {
1455                 if (p->reada == READA_FORWARD_ALWAYS)
1456                         reada_for_search(fs_info, p, level, slot, key->objectid);
1457
1458                 /* first we do an atomic uptodate check */
1459                 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
1460                         /*
1461                          * Do extra check for first_key, eb can be stale due to
1462                          * being cached, read from scrub, or have multiple
1463                          * parents (shared tree blocks).
1464                          */
1465                         if (btrfs_verify_level_key(tmp,
1466                                         parent_level - 1, &first_key, gen)) {
1467                                 free_extent_buffer(tmp);
1468                                 return -EUCLEAN;
1469                         }
1470                         *eb_ret = tmp;
1471                         return 0;
1472                 }
1473
1474                 /* now we're allowed to do a blocking uptodate check */
1475                 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
1476                 if (!ret) {
1477                         *eb_ret = tmp;
1478                         return 0;
1479                 }
1480                 free_extent_buffer(tmp);
1481                 btrfs_release_path(p);
1482                 return -EIO;
1483         }
1484
1485         /*
1486          * reduce lock contention at high levels
1487          * of the btree by dropping locks before
1488          * we read.  Don't release the lock on the current
1489          * level because we need to walk this node to figure
1490          * out which blocks to read.
1491          */
1492         btrfs_unlock_up_safe(p, level + 1);
1493
1494         if (p->reada != READA_NONE)
1495                 reada_for_search(fs_info, p, level, slot, key->objectid);
1496
1497         ret = -EAGAIN;
1498         tmp = read_tree_block(fs_info, blocknr, root->root_key.objectid,
1499                               gen, parent_level - 1, &first_key);
1500         if (!IS_ERR(tmp)) {
1501                 /*
1502                  * If the read above didn't mark this buffer up to date,
1503                  * it will never end up being up to date.  Set ret to EIO now
1504                  * and give up so that our caller doesn't loop forever
1505                  * on our EAGAINs.
1506                  */
1507                 if (!extent_buffer_uptodate(tmp))
1508                         ret = -EIO;
1509                 free_extent_buffer(tmp);
1510         } else {
1511                 ret = PTR_ERR(tmp);
1512         }
1513
1514         btrfs_release_path(p);
1515         return ret;
1516 }
1517
1518 /*
1519  * helper function for btrfs_search_slot.  This does all of the checks
1520  * for node-level blocks and does any balancing required based on
1521  * the ins_len.
1522  *
1523  * If no extra work was required, zero is returned.  If we had to
1524  * drop the path, -EAGAIN is returned and btrfs_search_slot must
1525  * start over
1526  */
1527 static int
1528 setup_nodes_for_search(struct btrfs_trans_handle *trans,
1529                        struct btrfs_root *root, struct btrfs_path *p,
1530                        struct extent_buffer *b, int level, int ins_len,
1531                        int *write_lock_level)
1532 {
1533         struct btrfs_fs_info *fs_info = root->fs_info;
1534         int ret = 0;
1535
1536         if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1537             BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
1538
1539                 if (*write_lock_level < level + 1) {
1540                         *write_lock_level = level + 1;
1541                         btrfs_release_path(p);
1542                         return -EAGAIN;
1543                 }
1544
1545                 reada_for_balance(p, level);
1546                 ret = split_node(trans, root, p, level);
1547
1548                 b = p->nodes[level];
1549         } else if (ins_len < 0 && btrfs_header_nritems(b) <
1550                    BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
1551
1552                 if (*write_lock_level < level + 1) {
1553                         *write_lock_level = level + 1;
1554                         btrfs_release_path(p);
1555                         return -EAGAIN;
1556                 }
1557
1558                 reada_for_balance(p, level);
1559                 ret = balance_level(trans, root, p, level);
1560                 if (ret)
1561                         return ret;
1562
1563                 b = p->nodes[level];
1564                 if (!b) {
1565                         btrfs_release_path(p);
1566                         return -EAGAIN;
1567                 }
1568                 BUG_ON(btrfs_header_nritems(b) == 1);
1569         }
1570         return ret;
1571 }
1572
1573 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
1574                 u64 iobjectid, u64 ioff, u8 key_type,
1575                 struct btrfs_key *found_key)
1576 {
1577         int ret;
1578         struct btrfs_key key;
1579         struct extent_buffer *eb;
1580
1581         ASSERT(path);
1582         ASSERT(found_key);
1583
1584         key.type = key_type;
1585         key.objectid = iobjectid;
1586         key.offset = ioff;
1587
1588         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1589         if (ret < 0)
1590                 return ret;
1591
1592         eb = path->nodes[0];
1593         if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1594                 ret = btrfs_next_leaf(fs_root, path);
1595                 if (ret)
1596                         return ret;
1597                 eb = path->nodes[0];
1598         }
1599
1600         btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1601         if (found_key->type != key.type ||
1602                         found_key->objectid != key.objectid)
1603                 return 1;
1604
1605         return 0;
1606 }
1607
1608 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
1609                                                         struct btrfs_path *p,
1610                                                         int write_lock_level)
1611 {
1612         struct btrfs_fs_info *fs_info = root->fs_info;
1613         struct extent_buffer *b;
1614         int root_lock;
1615         int level = 0;
1616
1617         /* We try very hard to do read locks on the root */
1618         root_lock = BTRFS_READ_LOCK;
1619
1620         if (p->search_commit_root) {
1621                 /*
1622                  * The commit roots are read only so we always do read locks,
1623                  * and we always must hold the commit_root_sem when doing
1624                  * searches on them, the only exception is send where we don't
1625                  * want to block transaction commits for a long time, so
1626                  * we need to clone the commit root in order to avoid races
1627                  * with transaction commits that create a snapshot of one of
1628                  * the roots used by a send operation.
1629                  */
1630                 if (p->need_commit_sem) {
1631                         down_read(&fs_info->commit_root_sem);
1632                         b = btrfs_clone_extent_buffer(root->commit_root);
1633                         up_read(&fs_info->commit_root_sem);
1634                         if (!b)
1635                                 return ERR_PTR(-ENOMEM);
1636
1637                 } else {
1638                         b = root->commit_root;
1639                         atomic_inc(&b->refs);
1640                 }
1641                 level = btrfs_header_level(b);
1642                 /*
1643                  * Ensure that all callers have set skip_locking when
1644                  * p->search_commit_root = 1.
1645                  */
1646                 ASSERT(p->skip_locking == 1);
1647
1648                 goto out;
1649         }
1650
1651         if (p->skip_locking) {
1652                 b = btrfs_root_node(root);
1653                 level = btrfs_header_level(b);
1654                 goto out;
1655         }
1656
1657         /*
1658          * If the level is set to maximum, we can skip trying to get the read
1659          * lock.
1660          */
1661         if (write_lock_level < BTRFS_MAX_LEVEL) {
1662                 /*
1663                  * We don't know the level of the root node until we actually
1664                  * have it read locked
1665                  */
1666                 b = btrfs_read_lock_root_node(root);
1667                 level = btrfs_header_level(b);
1668                 if (level > write_lock_level)
1669                         goto out;
1670
1671                 /* Whoops, must trade for write lock */
1672                 btrfs_tree_read_unlock(b);
1673                 free_extent_buffer(b);
1674         }
1675
1676         b = btrfs_lock_root_node(root);
1677         root_lock = BTRFS_WRITE_LOCK;
1678
1679         /* The level might have changed, check again */
1680         level = btrfs_header_level(b);
1681
1682 out:
1683         p->nodes[level] = b;
1684         if (!p->skip_locking)
1685                 p->locks[level] = root_lock;
1686         /*
1687          * Callers are responsible for dropping b's references.
1688          */
1689         return b;
1690 }
1691
1692
1693 /*
1694  * btrfs_search_slot - look for a key in a tree and perform necessary
1695  * modifications to preserve tree invariants.
1696  *
1697  * @trans:      Handle of transaction, used when modifying the tree
1698  * @p:          Holds all btree nodes along the search path
1699  * @root:       The root node of the tree
1700  * @key:        The key we are looking for
1701  * @ins_len:    Indicates purpose of search:
1702  *              >0  for inserts it's size of item inserted (*)
1703  *              <0  for deletions
1704  *               0  for plain searches, not modifying the tree
1705  *
1706  *              (*) If size of item inserted doesn't include
1707  *              sizeof(struct btrfs_item), then p->search_for_extension must
1708  *              be set.
1709  * @cow:        boolean should CoW operations be performed. Must always be 1
1710  *              when modifying the tree.
1711  *
1712  * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
1713  * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
1714  *
1715  * If @key is found, 0 is returned and you can find the item in the leaf level
1716  * of the path (level 0)
1717  *
1718  * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
1719  * points to the slot where it should be inserted
1720  *
1721  * If an error is encountered while searching the tree a negative error number
1722  * is returned
1723  */
1724 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1725                       const struct btrfs_key *key, struct btrfs_path *p,
1726                       int ins_len, int cow)
1727 {
1728         struct extent_buffer *b;
1729         int slot;
1730         int ret;
1731         int err;
1732         int level;
1733         int lowest_unlock = 1;
1734         /* everything at write_lock_level or lower must be write locked */
1735         int write_lock_level = 0;
1736         u8 lowest_level = 0;
1737         int min_write_lock_level;
1738         int prev_cmp;
1739
1740         lowest_level = p->lowest_level;
1741         WARN_ON(lowest_level && ins_len > 0);
1742         WARN_ON(p->nodes[0] != NULL);
1743         BUG_ON(!cow && ins_len);
1744
1745         if (ins_len < 0) {
1746                 lowest_unlock = 2;
1747
1748                 /* when we are removing items, we might have to go up to level
1749                  * two as we update tree pointers  Make sure we keep write
1750                  * for those levels as well
1751                  */
1752                 write_lock_level = 2;
1753         } else if (ins_len > 0) {
1754                 /*
1755                  * for inserting items, make sure we have a write lock on
1756                  * level 1 so we can update keys
1757                  */
1758                 write_lock_level = 1;
1759         }
1760
1761         if (!cow)
1762                 write_lock_level = -1;
1763
1764         if (cow && (p->keep_locks || p->lowest_level))
1765                 write_lock_level = BTRFS_MAX_LEVEL;
1766
1767         min_write_lock_level = write_lock_level;
1768
1769 again:
1770         prev_cmp = -1;
1771         b = btrfs_search_slot_get_root(root, p, write_lock_level);
1772         if (IS_ERR(b)) {
1773                 ret = PTR_ERR(b);
1774                 goto done;
1775         }
1776
1777         while (b) {
1778                 int dec = 0;
1779
1780                 level = btrfs_header_level(b);
1781
1782                 if (cow) {
1783                         bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
1784
1785                         /*
1786                          * if we don't really need to cow this block
1787                          * then we don't want to set the path blocking,
1788                          * so we test it here
1789                          */
1790                         if (!should_cow_block(trans, root, b))
1791                                 goto cow_done;
1792
1793                         /*
1794                          * must have write locks on this node and the
1795                          * parent
1796                          */
1797                         if (level > write_lock_level ||
1798                             (level + 1 > write_lock_level &&
1799                             level + 1 < BTRFS_MAX_LEVEL &&
1800                             p->nodes[level + 1])) {
1801                                 write_lock_level = level + 1;
1802                                 btrfs_release_path(p);
1803                                 goto again;
1804                         }
1805
1806                         if (last_level)
1807                                 err = btrfs_cow_block(trans, root, b, NULL, 0,
1808                                                       &b,
1809                                                       BTRFS_NESTING_COW);
1810                         else
1811                                 err = btrfs_cow_block(trans, root, b,
1812                                                       p->nodes[level + 1],
1813                                                       p->slots[level + 1], &b,
1814                                                       BTRFS_NESTING_COW);
1815                         if (err) {
1816                                 ret = err;
1817                                 goto done;
1818                         }
1819                 }
1820 cow_done:
1821                 p->nodes[level] = b;
1822                 /*
1823                  * Leave path with blocking locks to avoid massive
1824                  * lock context switch, this is made on purpose.
1825                  */
1826
1827                 /*
1828                  * we have a lock on b and as long as we aren't changing
1829                  * the tree, there is no way to for the items in b to change.
1830                  * It is safe to drop the lock on our parent before we
1831                  * go through the expensive btree search on b.
1832                  *
1833                  * If we're inserting or deleting (ins_len != 0), then we might
1834                  * be changing slot zero, which may require changing the parent.
1835                  * So, we can't drop the lock until after we know which slot
1836                  * we're operating on.
1837                  */
1838                 if (!ins_len && !p->keep_locks) {
1839                         int u = level + 1;
1840
1841                         if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
1842                                 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
1843                                 p->locks[u] = 0;
1844                         }
1845                 }
1846
1847                 /*
1848                  * If btrfs_bin_search returns an exact match (prev_cmp == 0)
1849                  * we can safely assume the target key will always be in slot 0
1850                  * on lower levels due to the invariants BTRFS' btree provides,
1851                  * namely that a btrfs_key_ptr entry always points to the
1852                  * lowest key in the child node, thus we can skip searching
1853                  * lower levels
1854                  */
1855                 if (prev_cmp == 0) {
1856                         slot = 0;
1857                         ret = 0;
1858                 } else {
1859                         ret = btrfs_bin_search(b, key, &slot);
1860                         prev_cmp = ret;
1861                         if (ret < 0)
1862                                 goto done;
1863                 }
1864
1865                 if (level == 0) {
1866                         p->slots[level] = slot;
1867                         /*
1868                          * Item key already exists. In this case, if we are
1869                          * allowed to insert the item (for example, in dir_item
1870                          * case, item key collision is allowed), it will be
1871                          * merged with the original item. Only the item size
1872                          * grows, no new btrfs item will be added. If
1873                          * search_for_extension is not set, ins_len already
1874                          * accounts the size btrfs_item, deduct it here so leaf
1875                          * space check will be correct.
1876                          */
1877                         if (ret == 0 && ins_len > 0 && !p->search_for_extension) {
1878                                 ASSERT(ins_len >= sizeof(struct btrfs_item));
1879                                 ins_len -= sizeof(struct btrfs_item);
1880                         }
1881                         if (ins_len > 0 &&
1882                             btrfs_leaf_free_space(b) < ins_len) {
1883                                 if (write_lock_level < 1) {
1884                                         write_lock_level = 1;
1885                                         btrfs_release_path(p);
1886                                         goto again;
1887                                 }
1888
1889                                 err = split_leaf(trans, root, key,
1890                                                  p, ins_len, ret == 0);
1891
1892                                 BUG_ON(err > 0);
1893                                 if (err) {
1894                                         ret = err;
1895                                         goto done;
1896                                 }
1897                         }
1898                         if (!p->search_for_split)
1899                                 unlock_up(p, level, lowest_unlock,
1900                                           min_write_lock_level, NULL);
1901                         goto done;
1902                 }
1903                 if (ret && slot > 0) {
1904                         dec = 1;
1905                         slot--;
1906                 }
1907                 p->slots[level] = slot;
1908                 err = setup_nodes_for_search(trans, root, p, b, level, ins_len,
1909                                              &write_lock_level);
1910                 if (err == -EAGAIN)
1911                         goto again;
1912                 if (err) {
1913                         ret = err;
1914                         goto done;
1915                 }
1916                 b = p->nodes[level];
1917                 slot = p->slots[level];
1918
1919                 /*
1920                  * Slot 0 is special, if we change the key we have to update
1921                  * the parent pointer which means we must have a write lock on
1922                  * the parent
1923                  */
1924                 if (slot == 0 && ins_len && write_lock_level < level + 1) {
1925                         write_lock_level = level + 1;
1926                         btrfs_release_path(p);
1927                         goto again;
1928                 }
1929
1930                 unlock_up(p, level, lowest_unlock, min_write_lock_level,
1931                           &write_lock_level);
1932
1933                 if (level == lowest_level) {
1934                         if (dec)
1935                                 p->slots[level]++;
1936                         goto done;
1937                 }
1938
1939                 err = read_block_for_search(root, p, &b, level, slot, key);
1940                 if (err == -EAGAIN)
1941                         goto again;
1942                 if (err) {
1943                         ret = err;
1944                         goto done;
1945                 }
1946
1947                 if (!p->skip_locking) {
1948                         level = btrfs_header_level(b);
1949                         if (level <= write_lock_level) {
1950                                 btrfs_tree_lock(b);
1951                                 p->locks[level] = BTRFS_WRITE_LOCK;
1952                         } else {
1953                                 btrfs_tree_read_lock(b);
1954                                 p->locks[level] = BTRFS_READ_LOCK;
1955                         }
1956                         p->nodes[level] = b;
1957                 }
1958         }
1959         ret = 1;
1960 done:
1961         if (ret < 0 && !p->skip_release_on_error)
1962                 btrfs_release_path(p);
1963         return ret;
1964 }
1965 ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO);
1966
1967 /*
1968  * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
1969  * current state of the tree together with the operations recorded in the tree
1970  * modification log to search for the key in a previous version of this tree, as
1971  * denoted by the time_seq parameter.
1972  *
1973  * Naturally, there is no support for insert, delete or cow operations.
1974  *
1975  * The resulting path and return value will be set up as if we called
1976  * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
1977  */
1978 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
1979                           struct btrfs_path *p, u64 time_seq)
1980 {
1981         struct btrfs_fs_info *fs_info = root->fs_info;
1982         struct extent_buffer *b;
1983         int slot;
1984         int ret;
1985         int err;
1986         int level;
1987         int lowest_unlock = 1;
1988         u8 lowest_level = 0;
1989
1990         lowest_level = p->lowest_level;
1991         WARN_ON(p->nodes[0] != NULL);
1992
1993         if (p->search_commit_root) {
1994                 BUG_ON(time_seq);
1995                 return btrfs_search_slot(NULL, root, key, p, 0, 0);
1996         }
1997
1998 again:
1999         b = btrfs_get_old_root(root, time_seq);
2000         if (!b) {
2001                 ret = -EIO;
2002                 goto done;
2003         }
2004         level = btrfs_header_level(b);
2005         p->locks[level] = BTRFS_READ_LOCK;
2006
2007         while (b) {
2008                 int dec = 0;
2009
2010                 level = btrfs_header_level(b);
2011                 p->nodes[level] = b;
2012
2013                 /*
2014                  * we have a lock on b and as long as we aren't changing
2015                  * the tree, there is no way to for the items in b to change.
2016                  * It is safe to drop the lock on our parent before we
2017                  * go through the expensive btree search on b.
2018                  */
2019                 btrfs_unlock_up_safe(p, level + 1);
2020
2021                 ret = btrfs_bin_search(b, key, &slot);
2022                 if (ret < 0)
2023                         goto done;
2024
2025                 if (level == 0) {
2026                         p->slots[level] = slot;
2027                         unlock_up(p, level, lowest_unlock, 0, NULL);
2028                         goto done;
2029                 }
2030
2031                 if (ret && slot > 0) {
2032                         dec = 1;
2033                         slot--;
2034                 }
2035                 p->slots[level] = slot;
2036                 unlock_up(p, level, lowest_unlock, 0, NULL);
2037
2038                 if (level == lowest_level) {
2039                         if (dec)
2040                                 p->slots[level]++;
2041                         goto done;
2042                 }
2043
2044                 err = read_block_for_search(root, p, &b, level, slot, key);
2045                 if (err == -EAGAIN)
2046                         goto again;
2047                 if (err) {
2048                         ret = err;
2049                         goto done;
2050                 }
2051
2052                 level = btrfs_header_level(b);
2053                 btrfs_tree_read_lock(b);
2054                 b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq);
2055                 if (!b) {
2056                         ret = -ENOMEM;
2057                         goto done;
2058                 }
2059                 p->locks[level] = BTRFS_READ_LOCK;
2060                 p->nodes[level] = b;
2061         }
2062         ret = 1;
2063 done:
2064         if (ret < 0)
2065                 btrfs_release_path(p);
2066
2067         return ret;
2068 }
2069
2070 /*
2071  * helper to use instead of search slot if no exact match is needed but
2072  * instead the next or previous item should be returned.
2073  * When find_higher is true, the next higher item is returned, the next lower
2074  * otherwise.
2075  * When return_any and find_higher are both true, and no higher item is found,
2076  * return the next lower instead.
2077  * When return_any is true and find_higher is false, and no lower item is found,
2078  * return the next higher instead.
2079  * It returns 0 if any item is found, 1 if none is found (tree empty), and
2080  * < 0 on error
2081  */
2082 int btrfs_search_slot_for_read(struct btrfs_root *root,
2083                                const struct btrfs_key *key,
2084                                struct btrfs_path *p, int find_higher,
2085                                int return_any)
2086 {
2087         int ret;
2088         struct extent_buffer *leaf;
2089
2090 again:
2091         ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
2092         if (ret <= 0)
2093                 return ret;
2094         /*
2095          * a return value of 1 means the path is at the position where the
2096          * item should be inserted. Normally this is the next bigger item,
2097          * but in case the previous item is the last in a leaf, path points
2098          * to the first free slot in the previous leaf, i.e. at an invalid
2099          * item.
2100          */
2101         leaf = p->nodes[0];
2102
2103         if (find_higher) {
2104                 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
2105                         ret = btrfs_next_leaf(root, p);
2106                         if (ret <= 0)
2107                                 return ret;
2108                         if (!return_any)
2109                                 return 1;
2110                         /*
2111                          * no higher item found, return the next
2112                          * lower instead
2113                          */
2114                         return_any = 0;
2115                         find_higher = 0;
2116                         btrfs_release_path(p);
2117                         goto again;
2118                 }
2119         } else {
2120                 if (p->slots[0] == 0) {
2121                         ret = btrfs_prev_leaf(root, p);
2122                         if (ret < 0)
2123                                 return ret;
2124                         if (!ret) {
2125                                 leaf = p->nodes[0];
2126                                 if (p->slots[0] == btrfs_header_nritems(leaf))
2127                                         p->slots[0]--;
2128                                 return 0;
2129                         }
2130                         if (!return_any)
2131                                 return 1;
2132                         /*
2133                          * no lower item found, return the next
2134                          * higher instead
2135                          */
2136                         return_any = 0;
2137                         find_higher = 1;
2138                         btrfs_release_path(p);
2139                         goto again;
2140                 } else {
2141                         --p->slots[0];
2142                 }
2143         }
2144         return 0;
2145 }
2146
2147 /*
2148  * adjust the pointers going up the tree, starting at level
2149  * making sure the right key of each node is points to 'key'.
2150  * This is used after shifting pointers to the left, so it stops
2151  * fixing up pointers when a given leaf/node is not in slot 0 of the
2152  * higher levels
2153  *
2154  */
2155 static void fixup_low_keys(struct btrfs_path *path,
2156                            struct btrfs_disk_key *key, int level)
2157 {
2158         int i;
2159         struct extent_buffer *t;
2160         int ret;
2161
2162         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2163                 int tslot = path->slots[i];
2164
2165                 if (!path->nodes[i])
2166                         break;
2167                 t = path->nodes[i];
2168                 ret = btrfs_tree_mod_log_insert_key(t, tslot,
2169                                 BTRFS_MOD_LOG_KEY_REPLACE, GFP_ATOMIC);
2170                 BUG_ON(ret < 0);
2171                 btrfs_set_node_key(t, key, tslot);
2172                 btrfs_mark_buffer_dirty(path->nodes[i]);
2173                 if (tslot != 0)
2174                         break;
2175         }
2176 }
2177
2178 /*
2179  * update item key.
2180  *
2181  * This function isn't completely safe. It's the caller's responsibility
2182  * that the new key won't break the order
2183  */
2184 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
2185                              struct btrfs_path *path,
2186                              const struct btrfs_key *new_key)
2187 {
2188         struct btrfs_disk_key disk_key;
2189         struct extent_buffer *eb;
2190         int slot;
2191
2192         eb = path->nodes[0];
2193         slot = path->slots[0];
2194         if (slot > 0) {
2195                 btrfs_item_key(eb, &disk_key, slot - 1);
2196                 if (unlikely(comp_keys(&disk_key, new_key) >= 0)) {
2197                         btrfs_crit(fs_info,
2198                 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
2199                                    slot, btrfs_disk_key_objectid(&disk_key),
2200                                    btrfs_disk_key_type(&disk_key),
2201                                    btrfs_disk_key_offset(&disk_key),
2202                                    new_key->objectid, new_key->type,
2203                                    new_key->offset);
2204                         btrfs_print_leaf(eb);
2205                         BUG();
2206                 }
2207         }
2208         if (slot < btrfs_header_nritems(eb) - 1) {
2209                 btrfs_item_key(eb, &disk_key, slot + 1);
2210                 if (unlikely(comp_keys(&disk_key, new_key) <= 0)) {
2211                         btrfs_crit(fs_info,
2212                 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
2213                                    slot, btrfs_disk_key_objectid(&disk_key),
2214                                    btrfs_disk_key_type(&disk_key),
2215                                    btrfs_disk_key_offset(&disk_key),
2216                                    new_key->objectid, new_key->type,
2217                                    new_key->offset);
2218                         btrfs_print_leaf(eb);
2219                         BUG();
2220                 }
2221         }
2222
2223         btrfs_cpu_key_to_disk(&disk_key, new_key);
2224         btrfs_set_item_key(eb, &disk_key, slot);
2225         btrfs_mark_buffer_dirty(eb);
2226         if (slot == 0)
2227                 fixup_low_keys(path, &disk_key, 1);
2228 }
2229
2230 /*
2231  * Check key order of two sibling extent buffers.
2232  *
2233  * Return true if something is wrong.
2234  * Return false if everything is fine.
2235  *
2236  * Tree-checker only works inside one tree block, thus the following
2237  * corruption can not be detected by tree-checker:
2238  *
2239  * Leaf @left                   | Leaf @right
2240  * --------------------------------------------------------------
2241  * | 1 | 2 | 3 | 4 | 5 | f6 |   | 7 | 8 |
2242  *
2243  * Key f6 in leaf @left itself is valid, but not valid when the next
2244  * key in leaf @right is 7.
2245  * This can only be checked at tree block merge time.
2246  * And since tree checker has ensured all key order in each tree block
2247  * is correct, we only need to bother the last key of @left and the first
2248  * key of @right.
2249  */
2250 static bool check_sibling_keys(struct extent_buffer *left,
2251                                struct extent_buffer *right)
2252 {
2253         struct btrfs_key left_last;
2254         struct btrfs_key right_first;
2255         int level = btrfs_header_level(left);
2256         int nr_left = btrfs_header_nritems(left);
2257         int nr_right = btrfs_header_nritems(right);
2258
2259         /* No key to check in one of the tree blocks */
2260         if (!nr_left || !nr_right)
2261                 return false;
2262
2263         if (level) {
2264                 btrfs_node_key_to_cpu(left, &left_last, nr_left - 1);
2265                 btrfs_node_key_to_cpu(right, &right_first, 0);
2266         } else {
2267                 btrfs_item_key_to_cpu(left, &left_last, nr_left - 1);
2268                 btrfs_item_key_to_cpu(right, &right_first, 0);
2269         }
2270
2271         if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) {
2272                 btrfs_crit(left->fs_info,
2273 "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)",
2274                            left_last.objectid, left_last.type,
2275                            left_last.offset, right_first.objectid,
2276                            right_first.type, right_first.offset);
2277                 return true;
2278         }
2279         return false;
2280 }
2281
2282 /*
2283  * try to push data from one node into the next node left in the
2284  * tree.
2285  *
2286  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2287  * error, and > 0 if there was no room in the left hand block.
2288  */
2289 static int push_node_left(struct btrfs_trans_handle *trans,
2290                           struct extent_buffer *dst,
2291                           struct extent_buffer *src, int empty)
2292 {
2293         struct btrfs_fs_info *fs_info = trans->fs_info;
2294         int push_items = 0;
2295         int src_nritems;
2296         int dst_nritems;
2297         int ret = 0;
2298
2299         src_nritems = btrfs_header_nritems(src);
2300         dst_nritems = btrfs_header_nritems(dst);
2301         push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
2302         WARN_ON(btrfs_header_generation(src) != trans->transid);
2303         WARN_ON(btrfs_header_generation(dst) != trans->transid);
2304
2305         if (!empty && src_nritems <= 8)
2306                 return 1;
2307
2308         if (push_items <= 0)
2309                 return 1;
2310
2311         if (empty) {
2312                 push_items = min(src_nritems, push_items);
2313                 if (push_items < src_nritems) {
2314                         /* leave at least 8 pointers in the node if
2315                          * we aren't going to empty it
2316                          */
2317                         if (src_nritems - push_items < 8) {
2318                                 if (push_items <= 8)
2319                                         return 1;
2320                                 push_items -= 8;
2321                         }
2322                 }
2323         } else
2324                 push_items = min(src_nritems - 8, push_items);
2325
2326         /* dst is the left eb, src is the middle eb */
2327         if (check_sibling_keys(dst, src)) {
2328                 ret = -EUCLEAN;
2329                 btrfs_abort_transaction(trans, ret);
2330                 return ret;
2331         }
2332         ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
2333         if (ret) {
2334                 btrfs_abort_transaction(trans, ret);
2335                 return ret;
2336         }
2337         copy_extent_buffer(dst, src,
2338                            btrfs_node_key_ptr_offset(dst_nritems),
2339                            btrfs_node_key_ptr_offset(0),
2340                            push_items * sizeof(struct btrfs_key_ptr));
2341
2342         if (push_items < src_nritems) {
2343                 /*
2344                  * Don't call btrfs_tree_mod_log_insert_move() here, key removal
2345                  * was already fully logged by btrfs_tree_mod_log_eb_copy() above.
2346                  */
2347                 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
2348                                       btrfs_node_key_ptr_offset(push_items),
2349                                       (src_nritems - push_items) *
2350                                       sizeof(struct btrfs_key_ptr));
2351         }
2352         btrfs_set_header_nritems(src, src_nritems - push_items);
2353         btrfs_set_header_nritems(dst, dst_nritems + push_items);
2354         btrfs_mark_buffer_dirty(src);
2355         btrfs_mark_buffer_dirty(dst);
2356
2357         return ret;
2358 }
2359
2360 /*
2361  * try to push data from one node into the next node right in the
2362  * tree.
2363  *
2364  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
2365  * error, and > 0 if there was no room in the right hand block.
2366  *
2367  * this will  only push up to 1/2 the contents of the left node over
2368  */
2369 static int balance_node_right(struct btrfs_trans_handle *trans,
2370                               struct extent_buffer *dst,
2371                               struct extent_buffer *src)
2372 {
2373         struct btrfs_fs_info *fs_info = trans->fs_info;
2374         int push_items = 0;
2375         int max_push;
2376         int src_nritems;
2377         int dst_nritems;
2378         int ret = 0;
2379
2380         WARN_ON(btrfs_header_generation(src) != trans->transid);
2381         WARN_ON(btrfs_header_generation(dst) != trans->transid);
2382
2383         src_nritems = btrfs_header_nritems(src);
2384         dst_nritems = btrfs_header_nritems(dst);
2385         push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
2386         if (push_items <= 0)
2387                 return 1;
2388
2389         if (src_nritems < 4)
2390                 return 1;
2391
2392         max_push = src_nritems / 2 + 1;
2393         /* don't try to empty the node */
2394         if (max_push >= src_nritems)
2395                 return 1;
2396
2397         if (max_push < push_items)
2398                 push_items = max_push;
2399
2400         /* dst is the right eb, src is the middle eb */
2401         if (check_sibling_keys(src, dst)) {
2402                 ret = -EUCLEAN;
2403                 btrfs_abort_transaction(trans, ret);
2404                 return ret;
2405         }
2406         ret = btrfs_tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
2407         BUG_ON(ret < 0);
2408         memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2409                                       btrfs_node_key_ptr_offset(0),
2410                                       (dst_nritems) *
2411                                       sizeof(struct btrfs_key_ptr));
2412
2413         ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
2414                                          push_items);
2415         if (ret) {
2416                 btrfs_abort_transaction(trans, ret);
2417                 return ret;
2418         }
2419         copy_extent_buffer(dst, src,
2420                            btrfs_node_key_ptr_offset(0),
2421                            btrfs_node_key_ptr_offset(src_nritems - push_items),
2422                            push_items * sizeof(struct btrfs_key_ptr));
2423
2424         btrfs_set_header_nritems(src, src_nritems - push_items);
2425         btrfs_set_header_nritems(dst, dst_nritems + push_items);
2426
2427         btrfs_mark_buffer_dirty(src);
2428         btrfs_mark_buffer_dirty(dst);
2429
2430         return ret;
2431 }
2432
2433 /*
2434  * helper function to insert a new root level in the tree.
2435  * A new node is allocated, and a single item is inserted to
2436  * point to the existing root
2437  *
2438  * returns zero on success or < 0 on failure.
2439  */
2440 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2441                            struct btrfs_root *root,
2442                            struct btrfs_path *path, int level)
2443 {
2444         struct btrfs_fs_info *fs_info = root->fs_info;
2445         u64 lower_gen;
2446         struct extent_buffer *lower;
2447         struct extent_buffer *c;
2448         struct extent_buffer *old;
2449         struct btrfs_disk_key lower_key;
2450         int ret;
2451
2452         BUG_ON(path->nodes[level]);
2453         BUG_ON(path->nodes[level-1] != root->node);
2454
2455         lower = path->nodes[level-1];
2456         if (level == 1)
2457                 btrfs_item_key(lower, &lower_key, 0);
2458         else
2459                 btrfs_node_key(lower, &lower_key, 0);
2460
2461         c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level,
2462                                          root->node->start, 0,
2463                                          BTRFS_NESTING_NEW_ROOT);
2464         if (IS_ERR(c))
2465                 return PTR_ERR(c);
2466
2467         root_add_used(root, fs_info->nodesize);
2468
2469         btrfs_set_header_nritems(c, 1);
2470         btrfs_set_node_key(c, &lower_key, 0);
2471         btrfs_set_node_blockptr(c, 0, lower->start);
2472         lower_gen = btrfs_header_generation(lower);
2473         WARN_ON(lower_gen != trans->transid);
2474
2475         btrfs_set_node_ptr_generation(c, 0, lower_gen);
2476
2477         btrfs_mark_buffer_dirty(c);
2478
2479         old = root->node;
2480         ret = btrfs_tree_mod_log_insert_root(root->node, c, false);
2481         BUG_ON(ret < 0);
2482         rcu_assign_pointer(root->node, c);
2483
2484         /* the super has an extra ref to root->node */
2485         free_extent_buffer(old);
2486
2487         add_root_to_dirty_list(root);
2488         atomic_inc(&c->refs);
2489         path->nodes[level] = c;
2490         path->locks[level] = BTRFS_WRITE_LOCK;
2491         path->slots[level] = 0;
2492         return 0;
2493 }
2494
2495 /*
2496  * worker function to insert a single pointer in a node.
2497  * the node should have enough room for the pointer already
2498  *
2499  * slot and level indicate where you want the key to go, and
2500  * blocknr is the block the key points to.
2501  */
2502 static void insert_ptr(struct btrfs_trans_handle *trans,
2503                        struct btrfs_path *path,
2504                        struct btrfs_disk_key *key, u64 bytenr,
2505                        int slot, int level)
2506 {
2507         struct extent_buffer *lower;
2508         int nritems;
2509         int ret;
2510
2511         BUG_ON(!path->nodes[level]);
2512         btrfs_assert_tree_locked(path->nodes[level]);
2513         lower = path->nodes[level];
2514         nritems = btrfs_header_nritems(lower);
2515         BUG_ON(slot > nritems);
2516         BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
2517         if (slot != nritems) {
2518                 if (level) {
2519                         ret = btrfs_tree_mod_log_insert_move(lower, slot + 1,
2520                                         slot, nritems - slot);
2521                         BUG_ON(ret < 0);
2522                 }
2523                 memmove_extent_buffer(lower,
2524                               btrfs_node_key_ptr_offset(slot + 1),
2525                               btrfs_node_key_ptr_offset(slot),
2526                               (nritems - slot) * sizeof(struct btrfs_key_ptr));
2527         }
2528         if (level) {
2529                 ret = btrfs_tree_mod_log_insert_key(lower, slot,
2530                                             BTRFS_MOD_LOG_KEY_ADD, GFP_NOFS);
2531                 BUG_ON(ret < 0);
2532         }
2533         btrfs_set_node_key(lower, key, slot);
2534         btrfs_set_node_blockptr(lower, slot, bytenr);
2535         WARN_ON(trans->transid == 0);
2536         btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2537         btrfs_set_header_nritems(lower, nritems + 1);
2538         btrfs_mark_buffer_dirty(lower);
2539 }
2540
2541 /*
2542  * split the node at the specified level in path in two.
2543  * The path is corrected to point to the appropriate node after the split
2544  *
2545  * Before splitting this tries to make some room in the node by pushing
2546  * left and right, if either one works, it returns right away.
2547  *
2548  * returns 0 on success and < 0 on failure
2549  */
2550 static noinline int split_node(struct btrfs_trans_handle *trans,
2551                                struct btrfs_root *root,
2552                                struct btrfs_path *path, int level)
2553 {
2554         struct btrfs_fs_info *fs_info = root->fs_info;
2555         struct extent_buffer *c;
2556         struct extent_buffer *split;
2557         struct btrfs_disk_key disk_key;
2558         int mid;
2559         int ret;
2560         u32 c_nritems;
2561
2562         c = path->nodes[level];
2563         WARN_ON(btrfs_header_generation(c) != trans->transid);
2564         if (c == root->node) {
2565                 /*
2566                  * trying to split the root, lets make a new one
2567                  *
2568                  * tree mod log: We don't log_removal old root in
2569                  * insert_new_root, because that root buffer will be kept as a
2570                  * normal node. We are going to log removal of half of the
2571                  * elements below with btrfs_tree_mod_log_eb_copy(). We're
2572                  * holding a tree lock on the buffer, which is why we cannot
2573                  * race with other tree_mod_log users.
2574                  */
2575                 ret = insert_new_root(trans, root, path, level + 1);
2576                 if (ret)
2577                         return ret;
2578         } else {
2579                 ret = push_nodes_for_insert(trans, root, path, level);
2580                 c = path->nodes[level];
2581                 if (!ret && btrfs_header_nritems(c) <
2582                     BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
2583                         return 0;
2584                 if (ret < 0)
2585                         return ret;
2586         }
2587
2588         c_nritems = btrfs_header_nritems(c);
2589         mid = (c_nritems + 1) / 2;
2590         btrfs_node_key(c, &disk_key, mid);
2591
2592         split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level,
2593                                              c->start, 0, BTRFS_NESTING_SPLIT);
2594         if (IS_ERR(split))
2595                 return PTR_ERR(split);
2596
2597         root_add_used(root, fs_info->nodesize);
2598         ASSERT(btrfs_header_level(c) == level);
2599
2600         ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
2601         if (ret) {
2602                 btrfs_abort_transaction(trans, ret);
2603                 return ret;
2604         }
2605         copy_extent_buffer(split, c,
2606                            btrfs_node_key_ptr_offset(0),
2607                            btrfs_node_key_ptr_offset(mid),
2608                            (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2609         btrfs_set_header_nritems(split, c_nritems - mid);
2610         btrfs_set_header_nritems(c, mid);
2611
2612         btrfs_mark_buffer_dirty(c);
2613         btrfs_mark_buffer_dirty(split);
2614
2615         insert_ptr(trans, path, &disk_key, split->start,
2616                    path->slots[level + 1] + 1, level + 1);
2617
2618         if (path->slots[level] >= mid) {
2619                 path->slots[level] -= mid;
2620                 btrfs_tree_unlock(c);
2621                 free_extent_buffer(c);
2622                 path->nodes[level] = split;
2623                 path->slots[level + 1] += 1;
2624         } else {
2625                 btrfs_tree_unlock(split);
2626                 free_extent_buffer(split);
2627         }
2628         return 0;
2629 }
2630
2631 /*
2632  * how many bytes are required to store the items in a leaf.  start
2633  * and nr indicate which items in the leaf to check.  This totals up the
2634  * space used both by the item structs and the item data
2635  */
2636 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2637 {
2638         struct btrfs_item *start_item;
2639         struct btrfs_item *end_item;
2640         int data_len;
2641         int nritems = btrfs_header_nritems(l);
2642         int end = min(nritems, start + nr) - 1;
2643
2644         if (!nr)
2645                 return 0;
2646         start_item = btrfs_item_nr(start);
2647         end_item = btrfs_item_nr(end);
2648         data_len = btrfs_item_offset(l, start_item) +
2649                    btrfs_item_size(l, start_item);
2650         data_len = data_len - btrfs_item_offset(l, end_item);
2651         data_len += sizeof(struct btrfs_item) * nr;
2652         WARN_ON(data_len < 0);
2653         return data_len;
2654 }
2655
2656 /*
2657  * The space between the end of the leaf items and
2658  * the start of the leaf data.  IOW, how much room
2659  * the leaf has left for both items and data
2660  */
2661 noinline int btrfs_leaf_free_space(struct extent_buffer *leaf)
2662 {
2663         struct btrfs_fs_info *fs_info = leaf->fs_info;
2664         int nritems = btrfs_header_nritems(leaf);
2665         int ret;
2666
2667         ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
2668         if (ret < 0) {
2669                 btrfs_crit(fs_info,
2670                            "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
2671                            ret,
2672                            (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
2673                            leaf_space_used(leaf, 0, nritems), nritems);
2674         }
2675         return ret;
2676 }
2677
2678 /*
2679  * min slot controls the lowest index we're willing to push to the
2680  * right.  We'll push up to and including min_slot, but no lower
2681  */
2682 static noinline int __push_leaf_right(struct btrfs_path *path,
2683                                       int data_size, int empty,
2684                                       struct extent_buffer *right,
2685                                       int free_space, u32 left_nritems,
2686                                       u32 min_slot)
2687 {
2688         struct btrfs_fs_info *fs_info = right->fs_info;
2689         struct extent_buffer *left = path->nodes[0];
2690         struct extent_buffer *upper = path->nodes[1];
2691         struct btrfs_map_token token;
2692         struct btrfs_disk_key disk_key;
2693         int slot;
2694         u32 i;
2695         int push_space = 0;
2696         int push_items = 0;
2697         struct btrfs_item *item;
2698         u32 nr;
2699         u32 right_nritems;
2700         u32 data_end;
2701         u32 this_item_size;
2702
2703         if (empty)
2704                 nr = 0;
2705         else
2706                 nr = max_t(u32, 1, min_slot);
2707
2708         if (path->slots[0] >= left_nritems)
2709                 push_space += data_size;
2710
2711         slot = path->slots[1];
2712         i = left_nritems - 1;
2713         while (i >= nr) {
2714                 item = btrfs_item_nr(i);
2715
2716                 if (!empty && push_items > 0) {
2717                         if (path->slots[0] > i)
2718                                 break;
2719                         if (path->slots[0] == i) {
2720                                 int space = btrfs_leaf_free_space(left);
2721
2722                                 if (space + push_space * 2 > free_space)
2723                                         break;
2724                         }
2725                 }
2726
2727                 if (path->slots[0] == i)
2728                         push_space += data_size;
2729
2730                 this_item_size = btrfs_item_size(left, item);
2731                 if (this_item_size + sizeof(*item) + push_space > free_space)
2732                         break;
2733
2734                 push_items++;
2735                 push_space += this_item_size + sizeof(*item);
2736                 if (i == 0)
2737                         break;
2738                 i--;
2739         }
2740
2741         if (push_items == 0)
2742                 goto out_unlock;
2743
2744         WARN_ON(!empty && push_items == left_nritems);
2745
2746         /* push left to right */
2747         right_nritems = btrfs_header_nritems(right);
2748
2749         push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2750         push_space -= leaf_data_end(left);
2751
2752         /* make room in the right data area */
2753         data_end = leaf_data_end(right);
2754         memmove_extent_buffer(right,
2755                               BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
2756                               BTRFS_LEAF_DATA_OFFSET + data_end,
2757                               BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
2758
2759         /* copy from the left data area */
2760         copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
2761                      BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
2762                      BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left),
2763                      push_space);
2764
2765         memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2766                               btrfs_item_nr_offset(0),
2767                               right_nritems * sizeof(struct btrfs_item));
2768
2769         /* copy the items from left to right */
2770         copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2771                    btrfs_item_nr_offset(left_nritems - push_items),
2772                    push_items * sizeof(struct btrfs_item));
2773
2774         /* update the item pointers */
2775         btrfs_init_map_token(&token, right);
2776         right_nritems += push_items;
2777         btrfs_set_header_nritems(right, right_nritems);
2778         push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
2779         for (i = 0; i < right_nritems; i++) {
2780                 item = btrfs_item_nr(i);
2781                 push_space -= btrfs_token_item_size(&token, item);
2782                 btrfs_set_token_item_offset(&token, item, push_space);
2783         }
2784
2785         left_nritems -= push_items;
2786         btrfs_set_header_nritems(left, left_nritems);
2787
2788         if (left_nritems)
2789                 btrfs_mark_buffer_dirty(left);
2790         else
2791                 btrfs_clean_tree_block(left);
2792
2793         btrfs_mark_buffer_dirty(right);
2794
2795         btrfs_item_key(right, &disk_key, 0);
2796         btrfs_set_node_key(upper, &disk_key, slot + 1);
2797         btrfs_mark_buffer_dirty(upper);
2798
2799         /* then fixup the leaf pointer in the path */
2800         if (path->slots[0] >= left_nritems) {
2801                 path->slots[0] -= left_nritems;
2802                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2803                         btrfs_clean_tree_block(path->nodes[0]);
2804                 btrfs_tree_unlock(path->nodes[0]);
2805                 free_extent_buffer(path->nodes[0]);
2806                 path->nodes[0] = right;
2807                 path->slots[1] += 1;
2808         } else {
2809                 btrfs_tree_unlock(right);
2810                 free_extent_buffer(right);
2811         }
2812         return 0;
2813
2814 out_unlock:
2815         btrfs_tree_unlock(right);
2816         free_extent_buffer(right);
2817         return 1;
2818 }
2819
2820 /*
2821  * push some data in the path leaf to the right, trying to free up at
2822  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2823  *
2824  * returns 1 if the push failed because the other node didn't have enough
2825  * room, 0 if everything worked out and < 0 if there were major errors.
2826  *
2827  * this will push starting from min_slot to the end of the leaf.  It won't
2828  * push any slot lower than min_slot
2829  */
2830 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2831                            *root, struct btrfs_path *path,
2832                            int min_data_size, int data_size,
2833                            int empty, u32 min_slot)
2834 {
2835         struct extent_buffer *left = path->nodes[0];
2836         struct extent_buffer *right;
2837         struct extent_buffer *upper;
2838         int slot;
2839         int free_space;
2840         u32 left_nritems;
2841         int ret;
2842
2843         if (!path->nodes[1])
2844                 return 1;
2845
2846         slot = path->slots[1];
2847         upper = path->nodes[1];
2848         if (slot >= btrfs_header_nritems(upper) - 1)
2849                 return 1;
2850
2851         btrfs_assert_tree_locked(path->nodes[1]);
2852
2853         right = btrfs_read_node_slot(upper, slot + 1);
2854         /*
2855          * slot + 1 is not valid or we fail to read the right node,
2856          * no big deal, just return.
2857          */
2858         if (IS_ERR(right))
2859                 return 1;
2860
2861         __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
2862
2863         free_space = btrfs_leaf_free_space(right);
2864         if (free_space < data_size)
2865                 goto out_unlock;
2866
2867         /* cow and double check */
2868         ret = btrfs_cow_block(trans, root, right, upper,
2869                               slot + 1, &right, BTRFS_NESTING_RIGHT_COW);
2870         if (ret)
2871                 goto out_unlock;
2872
2873         free_space = btrfs_leaf_free_space(right);
2874         if (free_space < data_size)
2875                 goto out_unlock;
2876
2877         left_nritems = btrfs_header_nritems(left);
2878         if (left_nritems == 0)
2879                 goto out_unlock;
2880
2881         if (check_sibling_keys(left, right)) {
2882                 ret = -EUCLEAN;
2883                 btrfs_tree_unlock(right);
2884                 free_extent_buffer(right);
2885                 return ret;
2886         }
2887         if (path->slots[0] == left_nritems && !empty) {
2888                 /* Key greater than all keys in the leaf, right neighbor has
2889                  * enough room for it and we're not emptying our leaf to delete
2890                  * it, therefore use right neighbor to insert the new item and
2891                  * no need to touch/dirty our left leaf. */
2892                 btrfs_tree_unlock(left);
2893                 free_extent_buffer(left);
2894                 path->nodes[0] = right;
2895                 path->slots[0] = 0;
2896                 path->slots[1]++;
2897                 return 0;
2898         }
2899
2900         return __push_leaf_right(path, min_data_size, empty,
2901                                 right, free_space, left_nritems, min_slot);
2902 out_unlock:
2903         btrfs_tree_unlock(right);
2904         free_extent_buffer(right);
2905         return 1;
2906 }
2907
2908 /*
2909  * push some data in the path leaf to the left, trying to free up at
2910  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2911  *
2912  * max_slot can put a limit on how far into the leaf we'll push items.  The
2913  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us do all the
2914  * items
2915  */
2916 static noinline int __push_leaf_left(struct btrfs_path *path, int data_size,
2917                                      int empty, struct extent_buffer *left,
2918                                      int free_space, u32 right_nritems,
2919                                      u32 max_slot)
2920 {
2921         struct btrfs_fs_info *fs_info = left->fs_info;
2922         struct btrfs_disk_key disk_key;
2923         struct extent_buffer *right = path->nodes[0];
2924         int i;
2925         int push_space = 0;
2926         int push_items = 0;
2927         struct btrfs_item *item;
2928         u32 old_left_nritems;
2929         u32 nr;
2930         int ret = 0;
2931         u32 this_item_size;
2932         u32 old_left_item_size;
2933         struct btrfs_map_token token;
2934
2935         if (empty)
2936                 nr = min(right_nritems, max_slot);
2937         else
2938                 nr = min(right_nritems - 1, max_slot);
2939
2940         for (i = 0; i < nr; i++) {
2941                 item = btrfs_item_nr(i);
2942
2943                 if (!empty && push_items > 0) {
2944                         if (path->slots[0] < i)
2945                                 break;
2946                         if (path->slots[0] == i) {
2947                                 int space = btrfs_leaf_free_space(right);
2948
2949                                 if (space + push_space * 2 > free_space)
2950                                         break;
2951                         }
2952                 }
2953
2954                 if (path->slots[0] == i)
2955                         push_space += data_size;
2956
2957                 this_item_size = btrfs_item_size(right, item);
2958                 if (this_item_size + sizeof(*item) + push_space > free_space)
2959                         break;
2960
2961                 push_items++;
2962                 push_space += this_item_size + sizeof(*item);
2963         }
2964
2965         if (push_items == 0) {
2966                 ret = 1;
2967                 goto out;
2968         }
2969         WARN_ON(!empty && push_items == btrfs_header_nritems(right));
2970
2971         /* push data from right to left */
2972         copy_extent_buffer(left, right,
2973                            btrfs_item_nr_offset(btrfs_header_nritems(left)),
2974                            btrfs_item_nr_offset(0),
2975                            push_items * sizeof(struct btrfs_item));
2976
2977         push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
2978                      btrfs_item_offset_nr(right, push_items - 1);
2979
2980         copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
2981                      leaf_data_end(left) - push_space,
2982                      BTRFS_LEAF_DATA_OFFSET +
2983                      btrfs_item_offset_nr(right, push_items - 1),
2984                      push_space);
2985         old_left_nritems = btrfs_header_nritems(left);
2986         BUG_ON(old_left_nritems <= 0);
2987
2988         btrfs_init_map_token(&token, left);
2989         old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2990         for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2991                 u32 ioff;
2992
2993                 item = btrfs_item_nr(i);
2994
2995                 ioff = btrfs_token_item_offset(&token, item);
2996                 btrfs_set_token_item_offset(&token, item,
2997                       ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size));
2998         }
2999         btrfs_set_header_nritems(left, old_left_nritems + push_items);
3000
3001         /* fixup right node */
3002         if (push_items > right_nritems)
3003                 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3004                        right_nritems);
3005
3006         if (push_items < right_nritems) {
3007                 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3008                                                   leaf_data_end(right);
3009                 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3010                                       BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3011                                       BTRFS_LEAF_DATA_OFFSET +
3012                                       leaf_data_end(right), push_space);
3013
3014                 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3015                               btrfs_item_nr_offset(push_items),
3016                              (btrfs_header_nritems(right) - push_items) *
3017                              sizeof(struct btrfs_item));
3018         }
3019
3020         btrfs_init_map_token(&token, right);
3021         right_nritems -= push_items;
3022         btrfs_set_header_nritems(right, right_nritems);
3023         push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3024         for (i = 0; i < right_nritems; i++) {
3025                 item = btrfs_item_nr(i);
3026
3027                 push_space = push_space - btrfs_token_item_size(&token, item);
3028                 btrfs_set_token_item_offset(&token, item, push_space);
3029         }
3030
3031         btrfs_mark_buffer_dirty(left);
3032         if (right_nritems)
3033                 btrfs_mark_buffer_dirty(right);
3034         else
3035                 btrfs_clean_tree_block(right);
3036
3037         btrfs_item_key(right, &disk_key, 0);
3038         fixup_low_keys(path, &disk_key, 1);
3039
3040         /* then fixup the leaf pointer in the path */
3041         if (path->slots[0] < push_items) {
3042                 path->slots[0] += old_left_nritems;
3043                 btrfs_tree_unlock(path->nodes[0]);
3044                 free_extent_buffer(path->nodes[0]);
3045                 path->nodes[0] = left;
3046                 path->slots[1] -= 1;
3047         } else {
3048                 btrfs_tree_unlock(left);
3049                 free_extent_buffer(left);
3050                 path->slots[0] -= push_items;
3051         }
3052         BUG_ON(path->slots[0] < 0);
3053         return ret;
3054 out:
3055         btrfs_tree_unlock(left);
3056         free_extent_buffer(left);
3057         return ret;
3058 }
3059
3060 /*
3061  * push some data in the path leaf to the left, trying to free up at
3062  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3063  *
3064  * max_slot can put a limit on how far into the leaf we'll push items.  The
3065  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us push all the
3066  * items
3067  */
3068 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3069                           *root, struct btrfs_path *path, int min_data_size,
3070                           int data_size, int empty, u32 max_slot)
3071 {
3072         struct extent_buffer *right = path->nodes[0];
3073         struct extent_buffer *left;
3074         int slot;
3075         int free_space;
3076         u32 right_nritems;
3077         int ret = 0;
3078
3079         slot = path->slots[1];
3080         if (slot == 0)
3081                 return 1;
3082         if (!path->nodes[1])
3083                 return 1;
3084
3085         right_nritems = btrfs_header_nritems(right);
3086         if (right_nritems == 0)
3087                 return 1;
3088
3089         btrfs_assert_tree_locked(path->nodes[1]);
3090
3091         left = btrfs_read_node_slot(path->nodes[1], slot - 1);
3092         /*
3093          * slot - 1 is not valid or we fail to read the left node,
3094          * no big deal, just return.
3095          */
3096         if (IS_ERR(left))
3097                 return 1;
3098
3099         __btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
3100
3101         free_space = btrfs_leaf_free_space(left);
3102         if (free_space < data_size) {
3103                 ret = 1;
3104                 goto out;
3105         }
3106
3107         /* cow and double check */
3108         ret = btrfs_cow_block(trans, root, left,
3109                               path->nodes[1], slot - 1, &left,
3110                               BTRFS_NESTING_LEFT_COW);
3111         if (ret) {
3112                 /* we hit -ENOSPC, but it isn't fatal here */
3113                 if (ret == -ENOSPC)
3114                         ret = 1;
3115                 goto out;
3116         }
3117
3118         free_space = btrfs_leaf_free_space(left);
3119         if (free_space < data_size) {
3120                 ret = 1;
3121                 goto out;
3122         }
3123
3124         if (check_sibling_keys(left, right)) {
3125                 ret = -EUCLEAN;
3126                 goto out;
3127         }
3128         return __push_leaf_left(path, min_data_size,
3129                                empty, left, free_space, right_nritems,
3130                                max_slot);
3131 out:
3132         btrfs_tree_unlock(left);
3133         free_extent_buffer(left);
3134         return ret;
3135 }
3136
3137 /*
3138  * split the path's leaf in two, making sure there is at least data_size
3139  * available for the resulting leaf level of the path.
3140  */
3141 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
3142                                     struct btrfs_path *path,
3143                                     struct extent_buffer *l,
3144                                     struct extent_buffer *right,
3145                                     int slot, int mid, int nritems)
3146 {
3147         struct btrfs_fs_info *fs_info = trans->fs_info;
3148         int data_copy_size;
3149         int rt_data_off;
3150         int i;
3151         struct btrfs_disk_key disk_key;
3152         struct btrfs_map_token token;
3153
3154         nritems = nritems - mid;
3155         btrfs_set_header_nritems(right, nritems);
3156         data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l);
3157
3158         copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
3159                            btrfs_item_nr_offset(mid),
3160                            nritems * sizeof(struct btrfs_item));
3161
3162         copy_extent_buffer(right, l,
3163                      BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
3164                      data_copy_size, BTRFS_LEAF_DATA_OFFSET +
3165                      leaf_data_end(l), data_copy_size);
3166
3167         rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
3168
3169         btrfs_init_map_token(&token, right);
3170         for (i = 0; i < nritems; i++) {
3171                 struct btrfs_item *item = btrfs_item_nr(i);
3172                 u32 ioff;
3173
3174                 ioff = btrfs_token_item_offset(&token, item);
3175                 btrfs_set_token_item_offset(&token, item, ioff + rt_data_off);
3176         }
3177
3178         btrfs_set_header_nritems(l, mid);
3179         btrfs_item_key(right, &disk_key, 0);
3180         insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
3181
3182         btrfs_mark_buffer_dirty(right);
3183         btrfs_mark_buffer_dirty(l);
3184         BUG_ON(path->slots[0] != slot);
3185
3186         if (mid <= slot) {
3187                 btrfs_tree_unlock(path->nodes[0]);
3188                 free_extent_buffer(path->nodes[0]);
3189                 path->nodes[0] = right;
3190                 path->slots[0] -= mid;
3191                 path->slots[1] += 1;
3192         } else {
3193                 btrfs_tree_unlock(right);
3194                 free_extent_buffer(right);
3195         }
3196
3197         BUG_ON(path->slots[0] < 0);
3198 }
3199
3200 /*
3201  * double splits happen when we need to insert a big item in the middle
3202  * of a leaf.  A double split can leave us with 3 mostly empty leaves:
3203  * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3204  *          A                 B                 C
3205  *
3206  * We avoid this by trying to push the items on either side of our target
3207  * into the adjacent leaves.  If all goes well we can avoid the double split
3208  * completely.
3209  */
3210 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
3211                                           struct btrfs_root *root,
3212                                           struct btrfs_path *path,
3213                                           int data_size)
3214 {
3215         int ret;
3216         int progress = 0;
3217         int slot;
3218         u32 nritems;
3219         int space_needed = data_size;
3220
3221         slot = path->slots[0];
3222         if (slot < btrfs_header_nritems(path->nodes[0]))
3223                 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
3224
3225         /*
3226          * try to push all the items after our slot into the
3227          * right leaf
3228          */
3229         ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
3230         if (ret < 0)
3231                 return ret;
3232
3233         if (ret == 0)
3234                 progress++;
3235
3236         nritems = btrfs_header_nritems(path->nodes[0]);
3237         /*
3238          * our goal is to get our slot at the start or end of a leaf.  If
3239          * we've done so we're done
3240          */
3241         if (path->slots[0] == 0 || path->slots[0] == nritems)
3242                 return 0;
3243
3244         if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
3245                 return 0;
3246
3247         /* try to push all the items before our slot into the next leaf */
3248         slot = path->slots[0];
3249         space_needed = data_size;
3250         if (slot > 0)
3251                 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
3252         ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
3253         if (ret < 0)
3254                 return ret;
3255
3256         if (ret == 0)
3257                 progress++;
3258
3259         if (progress)
3260                 return 0;
3261         return 1;
3262 }
3263
3264 /*
3265  * split the path's leaf in two, making sure there is at least data_size
3266  * available for the resulting leaf level of the path.
3267  *
3268  * returns 0 if all went well and < 0 on failure.
3269  */
3270 static noinline int split_leaf(struct btrfs_trans_handle *trans,
3271                                struct btrfs_root *root,
3272                                const struct btrfs_key *ins_key,
3273                                struct btrfs_path *path, int data_size,
3274                                int extend)
3275 {
3276         struct btrfs_disk_key disk_key;
3277         struct extent_buffer *l;
3278         u32 nritems;
3279         int mid;
3280         int slot;
3281         struct extent_buffer *right;
3282         struct btrfs_fs_info *fs_info = root->fs_info;
3283         int ret = 0;
3284         int wret;
3285         int split;
3286         int num_doubles = 0;
3287         int tried_avoid_double = 0;
3288
3289         l = path->nodes[0];
3290         slot = path->slots[0];
3291         if (extend && data_size + btrfs_item_size_nr(l, slot) +
3292             sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
3293                 return -EOVERFLOW;
3294
3295         /* first try to make some room by pushing left and right */
3296         if (data_size && path->nodes[1]) {
3297                 int space_needed = data_size;
3298
3299                 if (slot < btrfs_header_nritems(l))
3300                         space_needed -= btrfs_leaf_free_space(l);
3301
3302                 wret = push_leaf_right(trans, root, path, space_needed,
3303                                        space_needed, 0, 0);
3304                 if (wret < 0)
3305                         return wret;
3306                 if (wret) {
3307                         space_needed = data_size;
3308                         if (slot > 0)
3309                                 space_needed -= btrfs_leaf_free_space(l);
3310                         wret = push_leaf_left(trans, root, path, space_needed,
3311                                               space_needed, 0, (u32)-1);
3312                         if (wret < 0)
3313                                 return wret;
3314                 }
3315                 l = path->nodes[0];
3316
3317                 /* did the pushes work? */
3318                 if (btrfs_leaf_free_space(l) >= data_size)
3319                         return 0;
3320         }
3321
3322         if (!path->nodes[1]) {
3323                 ret = insert_new_root(trans, root, path, 1);
3324                 if (ret)
3325                         return ret;
3326         }
3327 again:
3328         split = 1;
3329         l = path->nodes[0];
3330         slot = path->slots[0];
3331         nritems = btrfs_header_nritems(l);
3332         mid = (nritems + 1) / 2;
3333
3334         if (mid <= slot) {
3335                 if (nritems == 1 ||
3336                     leaf_space_used(l, mid, nritems - mid) + data_size >
3337                         BTRFS_LEAF_DATA_SIZE(fs_info)) {
3338                         if (slot >= nritems) {
3339                                 split = 0;
3340                         } else {
3341                                 mid = slot;
3342                                 if (mid != nritems &&
3343                                     leaf_space_used(l, mid, nritems - mid) +
3344                                     data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
3345                                         if (data_size && !tried_avoid_double)
3346                                                 goto push_for_double;
3347                                         split = 2;
3348                                 }
3349                         }
3350                 }
3351         } else {
3352                 if (leaf_space_used(l, 0, mid) + data_size >
3353                         BTRFS_LEAF_DATA_SIZE(fs_info)) {
3354                         if (!extend && data_size && slot == 0) {
3355                                 split = 0;
3356                         } else if ((extend || !data_size) && slot == 0) {
3357                                 mid = 1;
3358                         } else {
3359                                 mid = slot;
3360                                 if (mid != nritems &&
3361                                     leaf_space_used(l, mid, nritems - mid) +
3362                                     data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
3363                                         if (data_size && !tried_avoid_double)
3364                                                 goto push_for_double;
3365                                         split = 2;
3366                                 }
3367                         }
3368                 }
3369         }
3370
3371         if (split == 0)
3372                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
3373         else
3374                 btrfs_item_key(l, &disk_key, mid);
3375
3376         /*
3377          * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double
3378          * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES
3379          * subclasses, which is 8 at the time of this patch, and we've maxed it
3380          * out.  In the future we could add a
3381          * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just
3382          * use BTRFS_NESTING_NEW_ROOT.
3383          */
3384         right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0,
3385                                              l->start, 0, num_doubles ?
3386                                              BTRFS_NESTING_NEW_ROOT :
3387                                              BTRFS_NESTING_SPLIT);
3388         if (IS_ERR(right))
3389                 return PTR_ERR(right);
3390
3391         root_add_used(root, fs_info->nodesize);
3392
3393         if (split == 0) {
3394                 if (mid <= slot) {
3395                         btrfs_set_header_nritems(right, 0);
3396                         insert_ptr(trans, path, &disk_key,
3397                                    right->start, path->slots[1] + 1, 1);
3398                         btrfs_tree_unlock(path->nodes[0]);
3399                         free_extent_buffer(path->nodes[0]);
3400                         path->nodes[0] = right;
3401                         path->slots[0] = 0;
3402                         path->slots[1] += 1;
3403                 } else {
3404                         btrfs_set_header_nritems(right, 0);
3405                         insert_ptr(trans, path, &disk_key,
3406                                    right->start, path->slots[1], 1);
3407                         btrfs_tree_unlock(path->nodes[0]);
3408                         free_extent_buffer(path->nodes[0]);
3409                         path->nodes[0] = right;
3410                         path->slots[0] = 0;
3411                         if (path->slots[1] == 0)
3412                                 fixup_low_keys(path, &disk_key, 1);
3413                 }
3414                 /*
3415                  * We create a new leaf 'right' for the required ins_len and
3416                  * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
3417                  * the content of ins_len to 'right'.
3418                  */
3419                 return ret;
3420         }
3421
3422         copy_for_split(trans, path, l, right, slot, mid, nritems);
3423
3424         if (split == 2) {
3425                 BUG_ON(num_doubles != 0);
3426                 num_doubles++;
3427                 goto again;
3428         }
3429
3430         return 0;
3431
3432 push_for_double:
3433         push_for_double_split(trans, root, path, data_size);
3434         tried_avoid_double = 1;
3435         if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
3436                 return 0;
3437         goto again;
3438 }
3439
3440 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3441                                          struct btrfs_root *root,
3442                                          struct btrfs_path *path, int ins_len)
3443 {
3444         struct btrfs_key key;
3445         struct extent_buffer *leaf;
3446         struct btrfs_file_extent_item *fi;
3447         u64 extent_len = 0;
3448         u32 item_size;
3449         int ret;
3450
3451         leaf = path->nodes[0];
3452         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3453
3454         BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3455                key.type != BTRFS_EXTENT_CSUM_KEY);
3456
3457         if (btrfs_leaf_free_space(leaf) >= ins_len)
3458                 return 0;
3459
3460         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3461         if (key.type == BTRFS_EXTENT_DATA_KEY) {
3462                 fi = btrfs_item_ptr(leaf, path->slots[0],
3463                                     struct btrfs_file_extent_item);
3464                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3465         }
3466         btrfs_release_path(path);
3467
3468         path->keep_locks = 1;
3469         path->search_for_split = 1;
3470         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3471         path->search_for_split = 0;
3472         if (ret > 0)
3473                 ret = -EAGAIN;
3474         if (ret < 0)
3475                 goto err;
3476
3477         ret = -EAGAIN;
3478         leaf = path->nodes[0];
3479         /* if our item isn't there, return now */
3480         if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3481                 goto err;
3482
3483         /* the leaf has  changed, it now has room.  return now */
3484         if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
3485                 goto err;
3486
3487         if (key.type == BTRFS_EXTENT_DATA_KEY) {
3488                 fi = btrfs_item_ptr(leaf, path->slots[0],
3489                                     struct btrfs_file_extent_item);
3490                 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3491                         goto err;
3492         }
3493
3494         ret = split_leaf(trans, root, &key, path, ins_len, 1);
3495         if (ret)
3496                 goto err;
3497
3498         path->keep_locks = 0;
3499         btrfs_unlock_up_safe(path, 1);
3500         return 0;
3501 err:
3502         path->keep_locks = 0;
3503         return ret;
3504 }
3505
3506 static noinline int split_item(struct btrfs_path *path,
3507                                const struct btrfs_key *new_key,
3508                                unsigned long split_offset)
3509 {
3510         struct extent_buffer *leaf;
3511         struct btrfs_item *item;
3512         struct btrfs_item *new_item;
3513         int slot;
3514         char *buf;
3515         u32 nritems;
3516         u32 item_size;
3517         u32 orig_offset;
3518         struct btrfs_disk_key disk_key;
3519
3520         leaf = path->nodes[0];
3521         BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item));
3522
3523         item = btrfs_item_nr(path->slots[0]);
3524         orig_offset = btrfs_item_offset(leaf, item);
3525         item_size = btrfs_item_size(leaf, item);
3526
3527         buf = kmalloc(item_size, GFP_NOFS);
3528         if (!buf)
3529                 return -ENOMEM;
3530
3531         read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3532                             path->slots[0]), item_size);
3533
3534         slot = path->slots[0] + 1;
3535         nritems = btrfs_header_nritems(leaf);
3536         if (slot != nritems) {
3537                 /* shift the items */
3538                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3539                                 btrfs_item_nr_offset(slot),
3540                                 (nritems - slot) * sizeof(struct btrfs_item));
3541         }
3542
3543         btrfs_cpu_key_to_disk(&disk_key, new_key);
3544         btrfs_set_item_key(leaf, &disk_key, slot);
3545
3546         new_item = btrfs_item_nr(slot);
3547
3548         btrfs_set_item_offset(leaf, new_item, orig_offset);
3549         btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3550
3551         btrfs_set_item_offset(leaf, item,
3552                               orig_offset + item_size - split_offset);
3553         btrfs_set_item_size(leaf, item, split_offset);
3554
3555         btrfs_set_header_nritems(leaf, nritems + 1);
3556
3557         /* write the data for the start of the original item */
3558         write_extent_buffer(leaf, buf,
3559                             btrfs_item_ptr_offset(leaf, path->slots[0]),
3560                             split_offset);
3561
3562         /* write the data for the new item */
3563         write_extent_buffer(leaf, buf + split_offset,
3564                             btrfs_item_ptr_offset(leaf, slot),
3565                             item_size - split_offset);
3566         btrfs_mark_buffer_dirty(leaf);
3567
3568         BUG_ON(btrfs_leaf_free_space(leaf) < 0);
3569         kfree(buf);
3570         return 0;
3571 }
3572
3573 /*
3574  * This function splits a single item into two items,
3575  * giving 'new_key' to the new item and splitting the
3576  * old one at split_offset (from the start of the item).
3577  *
3578  * The path may be released by this operation.  After
3579  * the split, the path is pointing to the old item.  The
3580  * new item is going to be in the same node as the old one.
3581  *
3582  * Note, the item being split must be smaller enough to live alone on
3583  * a tree block with room for one extra struct btrfs_item
3584  *
3585  * This allows us to split the item in place, keeping a lock on the
3586  * leaf the entire time.
3587  */
3588 int btrfs_split_item(struct btrfs_trans_handle *trans,
3589                      struct btrfs_root *root,
3590                      struct btrfs_path *path,
3591                      const struct btrfs_key *new_key,
3592                      unsigned long split_offset)
3593 {
3594         int ret;
3595         ret = setup_leaf_for_split(trans, root, path,
3596                                    sizeof(struct btrfs_item));
3597         if (ret)
3598                 return ret;
3599
3600         ret = split_item(path, new_key, split_offset);
3601         return ret;
3602 }
3603
3604 /*
3605  * This function duplicate a item, giving 'new_key' to the new item.
3606  * It guarantees both items live in the same tree leaf and the new item
3607  * is contiguous with the original item.
3608  *
3609  * This allows us to split file extent in place, keeping a lock on the
3610  * leaf the entire time.
3611  */
3612 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
3613                          struct btrfs_root *root,
3614                          struct btrfs_path *path,
3615                          const struct btrfs_key *new_key)
3616 {
3617         struct extent_buffer *leaf;
3618         int ret;
3619         u32 item_size;
3620
3621         leaf = path->nodes[0];
3622         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3623         ret = setup_leaf_for_split(trans, root, path,
3624                                    item_size + sizeof(struct btrfs_item));
3625         if (ret)
3626                 return ret;
3627
3628         path->slots[0]++;
3629         setup_items_for_insert(root, path, new_key, &item_size, 1);
3630         leaf = path->nodes[0];
3631         memcpy_extent_buffer(leaf,
3632                              btrfs_item_ptr_offset(leaf, path->slots[0]),
3633                              btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
3634                              item_size);
3635         return 0;
3636 }
3637
3638 /*
3639  * make the item pointed to by the path smaller.  new_size indicates
3640  * how small to make it, and from_end tells us if we just chop bytes
3641  * off the end of the item or if we shift the item to chop bytes off
3642  * the front.
3643  */
3644 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end)
3645 {
3646         int slot;
3647         struct extent_buffer *leaf;
3648         struct btrfs_item *item;
3649         u32 nritems;
3650         unsigned int data_end;
3651         unsigned int old_data_start;
3652         unsigned int old_size;
3653         unsigned int size_diff;
3654         int i;
3655         struct btrfs_map_token token;
3656
3657         leaf = path->nodes[0];
3658         slot = path->slots[0];
3659
3660         old_size = btrfs_item_size_nr(leaf, slot);
3661         if (old_size == new_size)
3662                 return;
3663
3664         nritems = btrfs_header_nritems(leaf);
3665         data_end = leaf_data_end(leaf);
3666
3667         old_data_start = btrfs_item_offset_nr(leaf, slot);
3668
3669         size_diff = old_size - new_size;
3670
3671         BUG_ON(slot < 0);
3672         BUG_ON(slot >= nritems);
3673
3674         /*
3675          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3676          */
3677         /* first correct the data pointers */
3678         btrfs_init_map_token(&token, leaf);
3679         for (i = slot; i < nritems; i++) {
3680                 u32 ioff;
3681                 item = btrfs_item_nr(i);
3682
3683                 ioff = btrfs_token_item_offset(&token, item);
3684                 btrfs_set_token_item_offset(&token, item, ioff + size_diff);
3685         }
3686
3687         /* shift the data */
3688         if (from_end) {
3689                 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
3690                               data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
3691                               data_end, old_data_start + new_size - data_end);
3692         } else {
3693                 struct btrfs_disk_key disk_key;
3694                 u64 offset;
3695
3696                 btrfs_item_key(leaf, &disk_key, slot);
3697
3698                 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3699                         unsigned long ptr;
3700                         struct btrfs_file_extent_item *fi;
3701
3702                         fi = btrfs_item_ptr(leaf, slot,
3703                                             struct btrfs_file_extent_item);
3704                         fi = (struct btrfs_file_extent_item *)(
3705                              (unsigned long)fi - size_diff);
3706
3707                         if (btrfs_file_extent_type(leaf, fi) ==
3708                             BTRFS_FILE_EXTENT_INLINE) {
3709                                 ptr = btrfs_item_ptr_offset(leaf, slot);
3710                                 memmove_extent_buffer(leaf, ptr,
3711                                       (unsigned long)fi,
3712                                       BTRFS_FILE_EXTENT_INLINE_DATA_START);
3713                         }
3714                 }
3715
3716                 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
3717                               data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
3718                               data_end, old_data_start - data_end);
3719
3720                 offset = btrfs_disk_key_offset(&disk_key);
3721                 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3722                 btrfs_set_item_key(leaf, &disk_key, slot);
3723                 if (slot == 0)
3724                         fixup_low_keys(path, &disk_key, 1);
3725         }
3726
3727         item = btrfs_item_nr(slot);
3728         btrfs_set_item_size(leaf, item, new_size);
3729         btrfs_mark_buffer_dirty(leaf);
3730
3731         if (btrfs_leaf_free_space(leaf) < 0) {
3732                 btrfs_print_leaf(leaf);
3733                 BUG();
3734         }
3735 }
3736
3737 /*
3738  * make the item pointed to by the path bigger, data_size is the added size.
3739  */
3740 void btrfs_extend_item(struct btrfs_path *path, u32 data_size)
3741 {
3742         int slot;
3743         struct extent_buffer *leaf;
3744         struct btrfs_item *item;
3745         u32 nritems;
3746         unsigned int data_end;
3747         unsigned int old_data;
3748         unsigned int old_size;
3749         int i;
3750         struct btrfs_map_token token;
3751
3752         leaf = path->nodes[0];
3753
3754         nritems = btrfs_header_nritems(leaf);
3755         data_end = leaf_data_end(leaf);
3756
3757         if (btrfs_leaf_free_space(leaf) < data_size) {
3758                 btrfs_print_leaf(leaf);
3759                 BUG();
3760         }
3761         slot = path->slots[0];
3762         old_data = btrfs_item_end_nr(leaf, slot);
3763
3764         BUG_ON(slot < 0);
3765         if (slot >= nritems) {
3766                 btrfs_print_leaf(leaf);
3767                 btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d",
3768                            slot, nritems);
3769                 BUG();
3770         }
3771
3772         /*
3773          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3774          */
3775         /* first correct the data pointers */
3776         btrfs_init_map_token(&token, leaf);
3777         for (i = slot; i < nritems; i++) {
3778                 u32 ioff;
3779                 item = btrfs_item_nr(i);
3780
3781                 ioff = btrfs_token_item_offset(&token, item);
3782                 btrfs_set_token_item_offset(&token, item, ioff - data_size);
3783         }
3784
3785         /* shift the data */
3786         memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
3787                       data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
3788                       data_end, old_data - data_end);
3789
3790         data_end = old_data;
3791         old_size = btrfs_item_size_nr(leaf, slot);
3792         item = btrfs_item_nr(slot);
3793         btrfs_set_item_size(leaf, item, old_size + data_size);
3794         btrfs_mark_buffer_dirty(leaf);
3795
3796         if (btrfs_leaf_free_space(leaf) < 0) {
3797                 btrfs_print_leaf(leaf);
3798                 BUG();
3799         }
3800 }
3801
3802 /**
3803  * setup_items_for_insert - Helper called before inserting one or more items
3804  * to a leaf. Main purpose is to save stack depth by doing the bulk of the work
3805  * in a function that doesn't call btrfs_search_slot
3806  *
3807  * @root:       root we are inserting items to
3808  * @path:       points to the leaf/slot where we are going to insert new items
3809  * @cpu_key:    array of keys for items to be inserted
3810  * @data_size:  size of the body of each item we are going to insert
3811  * @nr:         size of @cpu_key/@data_size arrays
3812  */
3813 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
3814                             const struct btrfs_key *cpu_key, u32 *data_size,
3815                             int nr)
3816 {
3817         struct btrfs_fs_info *fs_info = root->fs_info;
3818         struct btrfs_item *item;
3819         int i;
3820         u32 nritems;
3821         unsigned int data_end;
3822         struct btrfs_disk_key disk_key;
3823         struct extent_buffer *leaf;
3824         int slot;
3825         struct btrfs_map_token token;
3826         u32 total_size;
3827         u32 total_data = 0;
3828
3829         for (i = 0; i < nr; i++)
3830                 total_data += data_size[i];
3831         total_size = total_data + (nr * sizeof(struct btrfs_item));
3832
3833         if (path->slots[0] == 0) {
3834                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3835                 fixup_low_keys(path, &disk_key, 1);
3836         }
3837         btrfs_unlock_up_safe(path, 1);
3838
3839         leaf = path->nodes[0];
3840         slot = path->slots[0];
3841
3842         nritems = btrfs_header_nritems(leaf);
3843         data_end = leaf_data_end(leaf);
3844
3845         if (btrfs_leaf_free_space(leaf) < total_size) {
3846                 btrfs_print_leaf(leaf);
3847                 btrfs_crit(fs_info, "not enough freespace need %u have %d",
3848                            total_size, btrfs_leaf_free_space(leaf));
3849                 BUG();
3850         }
3851
3852         btrfs_init_map_token(&token, leaf);
3853         if (slot != nritems) {
3854                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3855
3856                 if (old_data < data_end) {
3857                         btrfs_print_leaf(leaf);
3858                         btrfs_crit(fs_info,
3859                 "item at slot %d with data offset %u beyond data end of leaf %u",
3860                                    slot, old_data, data_end);
3861                         BUG();
3862                 }
3863                 /*
3864                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3865                  */
3866                 /* first correct the data pointers */
3867                 for (i = slot; i < nritems; i++) {
3868                         u32 ioff;
3869
3870                         item = btrfs_item_nr(i);
3871                         ioff = btrfs_token_item_offset(&token, item);
3872                         btrfs_set_token_item_offset(&token, item,
3873                                                     ioff - total_data);
3874                 }
3875                 /* shift the items */
3876                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3877                               btrfs_item_nr_offset(slot),
3878                               (nritems - slot) * sizeof(struct btrfs_item));
3879
3880                 /* shift the data */
3881                 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
3882                               data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
3883                               data_end, old_data - data_end);
3884                 data_end = old_data;
3885         }
3886
3887         /* setup the item for the new data */
3888         for (i = 0; i < nr; i++) {
3889                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3890                 btrfs_set_item_key(leaf, &disk_key, slot + i);
3891                 item = btrfs_item_nr(slot + i);
3892                 data_end -= data_size[i];
3893                 btrfs_set_token_item_offset(&token, item, data_end);
3894                 btrfs_set_token_item_size(&token, item, data_size[i]);
3895         }
3896
3897         btrfs_set_header_nritems(leaf, nritems + nr);
3898         btrfs_mark_buffer_dirty(leaf);
3899
3900         if (btrfs_leaf_free_space(leaf) < 0) {
3901                 btrfs_print_leaf(leaf);
3902                 BUG();
3903         }
3904 }
3905
3906 /*
3907  * Given a key and some data, insert items into the tree.
3908  * This does all the path init required, making room in the tree if needed.
3909  */
3910 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3911                             struct btrfs_root *root,
3912                             struct btrfs_path *path,
3913                             const struct btrfs_key *cpu_key, u32 *data_size,
3914                             int nr)
3915 {
3916         int ret = 0;
3917         int slot;
3918         int i;
3919         u32 total_size = 0;
3920         u32 total_data = 0;
3921
3922         for (i = 0; i < nr; i++)
3923                 total_data += data_size[i];
3924
3925         total_size = total_data + (nr * sizeof(struct btrfs_item));
3926         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3927         if (ret == 0)
3928                 return -EEXIST;
3929         if (ret < 0)
3930                 return ret;
3931
3932         slot = path->slots[0];
3933         BUG_ON(slot < 0);
3934
3935         setup_items_for_insert(root, path, cpu_key, data_size, nr);
3936         return 0;
3937 }
3938
3939 /*
3940  * Given a key and some data, insert an item into the tree.
3941  * This does all the path init required, making room in the tree if needed.
3942  */
3943 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3944                       const struct btrfs_key *cpu_key, void *data,
3945                       u32 data_size)
3946 {
3947         int ret = 0;
3948         struct btrfs_path *path;
3949         struct extent_buffer *leaf;
3950         unsigned long ptr;
3951
3952         path = btrfs_alloc_path();
3953         if (!path)
3954                 return -ENOMEM;
3955         ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3956         if (!ret) {
3957                 leaf = path->nodes[0];
3958                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3959                 write_extent_buffer(leaf, data, ptr, data_size);
3960                 btrfs_mark_buffer_dirty(leaf);
3961         }
3962         btrfs_free_path(path);
3963         return ret;
3964 }
3965
3966 /*
3967  * delete the pointer from a given node.
3968  *
3969  * the tree should have been previously balanced so the deletion does not
3970  * empty a node.
3971  */
3972 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
3973                     int level, int slot)
3974 {
3975         struct extent_buffer *parent = path->nodes[level];
3976         u32 nritems;
3977         int ret;
3978
3979         nritems = btrfs_header_nritems(parent);
3980         if (slot != nritems - 1) {
3981                 if (level) {
3982                         ret = btrfs_tree_mod_log_insert_move(parent, slot,
3983                                         slot + 1, nritems - slot - 1);
3984                         BUG_ON(ret < 0);
3985                 }
3986                 memmove_extent_buffer(parent,
3987                               btrfs_node_key_ptr_offset(slot),
3988                               btrfs_node_key_ptr_offset(slot + 1),
3989                               sizeof(struct btrfs_key_ptr) *
3990                               (nritems - slot - 1));
3991         } else if (level) {
3992                 ret = btrfs_tree_mod_log_insert_key(parent, slot,
3993                                 BTRFS_MOD_LOG_KEY_REMOVE, GFP_NOFS);
3994                 BUG_ON(ret < 0);
3995         }
3996
3997         nritems--;
3998         btrfs_set_header_nritems(parent, nritems);
3999         if (nritems == 0 && parent == root->node) {
4000                 BUG_ON(btrfs_header_level(root->node) != 1);
4001                 /* just turn the root into a leaf and break */
4002                 btrfs_set_header_level(root->node, 0);
4003         } else if (slot == 0) {
4004                 struct btrfs_disk_key disk_key;
4005
4006                 btrfs_node_key(parent, &disk_key, 0);
4007                 fixup_low_keys(path, &disk_key, level + 1);
4008         }
4009         btrfs_mark_buffer_dirty(parent);
4010 }
4011
4012 /*
4013  * a helper function to delete the leaf pointed to by path->slots[1] and
4014  * path->nodes[1].
4015  *
4016  * This deletes the pointer in path->nodes[1] and frees the leaf
4017  * block extent.  zero is returned if it all worked out, < 0 otherwise.
4018  *
4019  * The path must have already been setup for deleting the leaf, including
4020  * all the proper balancing.  path->nodes[1] must be locked.
4021  */
4022 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4023                                     struct btrfs_root *root,
4024                                     struct btrfs_path *path,
4025                                     struct extent_buffer *leaf)
4026 {
4027         WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4028         del_ptr(root, path, 1, path->slots[1]);
4029
4030         /*
4031          * btrfs_free_extent is expensive, we want to make sure we
4032          * aren't holding any locks when we call it
4033          */
4034         btrfs_unlock_up_safe(path, 0);
4035
4036         root_sub_used(root, leaf->len);
4037
4038         atomic_inc(&leaf->refs);
4039         btrfs_free_tree_block(trans, root, leaf, 0, 1);
4040         free_extent_buffer_stale(leaf);
4041 }
4042 /*
4043  * delete the item at the leaf level in path.  If that empties
4044  * the leaf, remove it from the tree
4045  */
4046 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4047                     struct btrfs_path *path, int slot, int nr)
4048 {
4049         struct btrfs_fs_info *fs_info = root->fs_info;
4050         struct extent_buffer *leaf;
4051         struct btrfs_item *item;
4052         u32 last_off;
4053         u32 dsize = 0;
4054         int ret = 0;
4055         int wret;
4056         int i;
4057         u32 nritems;
4058
4059         leaf = path->nodes[0];
4060         last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4061
4062         for (i = 0; i < nr; i++)
4063                 dsize += btrfs_item_size_nr(leaf, slot + i);
4064
4065         nritems = btrfs_header_nritems(leaf);
4066
4067         if (slot + nr != nritems) {
4068                 int data_end = leaf_data_end(leaf);
4069                 struct btrfs_map_token token;
4070
4071                 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4072                               data_end + dsize,
4073                               BTRFS_LEAF_DATA_OFFSET + data_end,
4074                               last_off - data_end);
4075
4076                 btrfs_init_map_token(&token, leaf);
4077                 for (i = slot + nr; i < nritems; i++) {
4078                         u32 ioff;
4079
4080                         item = btrfs_item_nr(i);
4081                         ioff = btrfs_token_item_offset(&token, item);
4082                         btrfs_set_token_item_offset(&token, item, ioff + dsize);
4083                 }
4084
4085                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4086                               btrfs_item_nr_offset(slot + nr),
4087                               sizeof(struct btrfs_item) *
4088                               (nritems - slot - nr));
4089         }
4090         btrfs_set_header_nritems(leaf, nritems - nr);
4091         nritems -= nr;
4092
4093         /* delete the leaf if we've emptied it */
4094         if (nritems == 0) {
4095                 if (leaf == root->node) {
4096                         btrfs_set_header_level(leaf, 0);
4097                 } else {
4098                         btrfs_clean_tree_block(leaf);
4099                         btrfs_del_leaf(trans, root, path, leaf);
4100                 }
4101         } else {
4102                 int used = leaf_space_used(leaf, 0, nritems);
4103                 if (slot == 0) {
4104                         struct btrfs_disk_key disk_key;
4105
4106                         btrfs_item_key(leaf, &disk_key, 0);
4107                         fixup_low_keys(path, &disk_key, 1);
4108                 }
4109
4110                 /* delete the leaf if it is mostly empty */
4111                 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
4112                         /* push_leaf_left fixes the path.
4113                          * make sure the path still points to our leaf
4114                          * for possible call to del_ptr below
4115                          */
4116                         slot = path->slots[1];
4117                         atomic_inc(&leaf->refs);
4118
4119                         wret = push_leaf_left(trans, root, path, 1, 1,
4120                                               1, (u32)-1);
4121                         if (wret < 0 && wret != -ENOSPC)
4122                                 ret = wret;
4123
4124                         if (path->nodes[0] == leaf &&
4125                             btrfs_header_nritems(leaf)) {
4126                                 wret = push_leaf_right(trans, root, path, 1,
4127                                                        1, 1, 0);
4128                                 if (wret < 0 && wret != -ENOSPC)
4129                                         ret = wret;
4130                         }
4131
4132                         if (btrfs_header_nritems(leaf) == 0) {
4133                                 path->slots[1] = slot;
4134                                 btrfs_del_leaf(trans, root, path, leaf);
4135                                 free_extent_buffer(leaf);
4136                                 ret = 0;
4137                         } else {
4138                                 /* if we're still in the path, make sure
4139                                  * we're dirty.  Otherwise, one of the
4140                                  * push_leaf functions must have already
4141                                  * dirtied this buffer
4142                                  */
4143                                 if (path->nodes[0] == leaf)
4144                                         btrfs_mark_buffer_dirty(leaf);
4145                                 free_extent_buffer(leaf);
4146                         }
4147                 } else {
4148                         btrfs_mark_buffer_dirty(leaf);
4149                 }
4150         }
4151         return ret;
4152 }
4153
4154 /*
4155  * search the tree again to find a leaf with lesser keys
4156  * returns 0 if it found something or 1 if there are no lesser leaves.
4157  * returns < 0 on io errors.
4158  *
4159  * This may release the path, and so you may lose any locks held at the
4160  * time you call it.
4161  */
4162 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
4163 {
4164         struct btrfs_key key;
4165         struct btrfs_disk_key found_key;
4166         int ret;
4167
4168         btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
4169
4170         if (key.offset > 0) {
4171                 key.offset--;
4172         } else if (key.type > 0) {
4173                 key.type--;
4174                 key.offset = (u64)-1;
4175         } else if (key.objectid > 0) {
4176                 key.objectid--;
4177                 key.type = (u8)-1;
4178                 key.offset = (u64)-1;
4179         } else {
4180                 return 1;
4181         }
4182
4183         btrfs_release_path(path);
4184         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4185         if (ret < 0)
4186                 return ret;
4187         btrfs_item_key(path->nodes[0], &found_key, 0);
4188         ret = comp_keys(&found_key, &key);
4189         /*
4190          * We might have had an item with the previous key in the tree right
4191          * before we released our path. And after we released our path, that
4192          * item might have been pushed to the first slot (0) of the leaf we
4193          * were holding due to a tree balance. Alternatively, an item with the
4194          * previous key can exist as the only element of a leaf (big fat item).
4195          * Therefore account for these 2 cases, so that our callers (like
4196          * btrfs_previous_item) don't miss an existing item with a key matching
4197          * the previous key we computed above.
4198          */
4199         if (ret <= 0)
4200                 return 0;
4201         return 1;
4202 }
4203
4204 /*
4205  * A helper function to walk down the tree starting at min_key, and looking
4206  * for nodes or leaves that are have a minimum transaction id.
4207  * This is used by the btree defrag code, and tree logging
4208  *
4209  * This does not cow, but it does stuff the starting key it finds back
4210  * into min_key, so you can call btrfs_search_slot with cow=1 on the
4211  * key and get a writable path.
4212  *
4213  * This honors path->lowest_level to prevent descent past a given level
4214  * of the tree.
4215  *
4216  * min_trans indicates the oldest transaction that you are interested
4217  * in walking through.  Any nodes or leaves older than min_trans are
4218  * skipped over (without reading them).
4219  *
4220  * returns zero if something useful was found, < 0 on error and 1 if there
4221  * was nothing in the tree that matched the search criteria.
4222  */
4223 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
4224                          struct btrfs_path *path,
4225                          u64 min_trans)
4226 {
4227         struct extent_buffer *cur;
4228         struct btrfs_key found_key;
4229         int slot;
4230         int sret;
4231         u32 nritems;
4232         int level;
4233         int ret = 1;
4234         int keep_locks = path->keep_locks;
4235
4236         path->keep_locks = 1;
4237 again:
4238         cur = btrfs_read_lock_root_node(root);
4239         level = btrfs_header_level(cur);
4240         WARN_ON(path->nodes[level]);
4241         path->nodes[level] = cur;
4242         path->locks[level] = BTRFS_READ_LOCK;
4243
4244         if (btrfs_header_generation(cur) < min_trans) {
4245                 ret = 1;
4246                 goto out;
4247         }
4248         while (1) {
4249                 nritems = btrfs_header_nritems(cur);
4250                 level = btrfs_header_level(cur);
4251                 sret = btrfs_bin_search(cur, min_key, &slot);
4252                 if (sret < 0) {
4253                         ret = sret;
4254                         goto out;
4255                 }
4256
4257                 /* at the lowest level, we're done, setup the path and exit */
4258                 if (level == path->lowest_level) {
4259                         if (slot >= nritems)
4260                                 goto find_next_key;
4261                         ret = 0;
4262                         path->slots[level] = slot;
4263                         btrfs_item_key_to_cpu(cur, &found_key, slot);
4264                         goto out;
4265                 }
4266                 if (sret && slot > 0)
4267                         slot--;
4268                 /*
4269                  * check this node pointer against the min_trans parameters.
4270                  * If it is too old, skip to the next one.
4271                  */
4272                 while (slot < nritems) {
4273                         u64 gen;
4274
4275                         gen = btrfs_node_ptr_generation(cur, slot);
4276                         if (gen < min_trans) {
4277                                 slot++;
4278                                 continue;
4279                         }
4280                         break;
4281                 }
4282 find_next_key:
4283                 /*
4284                  * we didn't find a candidate key in this node, walk forward
4285                  * and find another one
4286                  */
4287                 if (slot >= nritems) {
4288                         path->slots[level] = slot;
4289                         sret = btrfs_find_next_key(root, path, min_key, level,
4290                                                   min_trans);
4291                         if (sret == 0) {
4292                                 btrfs_release_path(path);
4293                                 goto again;
4294                         } else {
4295                                 goto out;
4296                         }
4297                 }
4298                 /* save our key for returning back */
4299                 btrfs_node_key_to_cpu(cur, &found_key, slot);
4300                 path->slots[level] = slot;
4301                 if (level == path->lowest_level) {
4302                         ret = 0;
4303                         goto out;
4304                 }
4305                 cur = btrfs_read_node_slot(cur, slot);
4306                 if (IS_ERR(cur)) {
4307                         ret = PTR_ERR(cur);
4308                         goto out;
4309                 }
4310
4311                 btrfs_tree_read_lock(cur);
4312
4313                 path->locks[level - 1] = BTRFS_READ_LOCK;
4314                 path->nodes[level - 1] = cur;
4315                 unlock_up(path, level, 1, 0, NULL);
4316         }
4317 out:
4318         path->keep_locks = keep_locks;
4319         if (ret == 0) {
4320                 btrfs_unlock_up_safe(path, path->lowest_level + 1);
4321                 memcpy(min_key, &found_key, sizeof(found_key));
4322         }
4323         return ret;
4324 }
4325
4326 /*
4327  * this is similar to btrfs_next_leaf, but does not try to preserve
4328  * and fixup the path.  It looks for and returns the next key in the
4329  * tree based on the current path and the min_trans parameters.
4330  *
4331  * 0 is returned if another key is found, < 0 if there are any errors
4332  * and 1 is returned if there are no higher keys in the tree
4333  *
4334  * path->keep_locks should be set to 1 on the search made before
4335  * calling this function.
4336  */
4337 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4338                         struct btrfs_key *key, int level, u64 min_trans)
4339 {
4340         int slot;
4341         struct extent_buffer *c;
4342
4343         WARN_ON(!path->keep_locks && !path->skip_locking);
4344         while (level < BTRFS_MAX_LEVEL) {
4345                 if (!path->nodes[level])
4346                         return 1;
4347
4348                 slot = path->slots[level] + 1;
4349                 c = path->nodes[level];
4350 next:
4351                 if (slot >= btrfs_header_nritems(c)) {
4352                         int ret;
4353                         int orig_lowest;
4354                         struct btrfs_key cur_key;
4355                         if (level + 1 >= BTRFS_MAX_LEVEL ||
4356                             !path->nodes[level + 1])
4357                                 return 1;
4358
4359                         if (path->locks[level + 1] || path->skip_locking) {
4360                                 level++;
4361                                 continue;
4362                         }
4363
4364                         slot = btrfs_header_nritems(c) - 1;
4365                         if (level == 0)
4366                                 btrfs_item_key_to_cpu(c, &cur_key, slot);
4367                         else
4368                                 btrfs_node_key_to_cpu(c, &cur_key, slot);
4369
4370                         orig_lowest = path->lowest_level;
4371                         btrfs_release_path(path);
4372                         path->lowest_level = level;
4373                         ret = btrfs_search_slot(NULL, root, &cur_key, path,
4374                                                 0, 0);
4375                         path->lowest_level = orig_lowest;
4376                         if (ret < 0)
4377                                 return ret;
4378
4379                         c = path->nodes[level];
4380                         slot = path->slots[level];
4381                         if (ret == 0)
4382                                 slot++;
4383                         goto next;
4384                 }
4385
4386                 if (level == 0)
4387                         btrfs_item_key_to_cpu(c, key, slot);
4388                 else {
4389                         u64 gen = btrfs_node_ptr_generation(c, slot);
4390
4391                         if (gen < min_trans) {
4392                                 slot++;
4393                                 goto next;
4394                         }
4395                         btrfs_node_key_to_cpu(c, key, slot);
4396                 }
4397                 return 0;
4398         }
4399         return 1;
4400 }
4401
4402 /*
4403  * search the tree again to find a leaf with greater keys
4404  * returns 0 if it found something or 1 if there are no greater leaves.
4405  * returns < 0 on io errors.
4406  */
4407 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4408 {
4409         return btrfs_next_old_leaf(root, path, 0);
4410 }
4411
4412 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
4413                         u64 time_seq)
4414 {
4415         int slot;
4416         int level;
4417         struct extent_buffer *c;
4418         struct extent_buffer *next;
4419         struct btrfs_key key;
4420         u32 nritems;
4421         int ret;
4422         int i;
4423
4424         nritems = btrfs_header_nritems(path->nodes[0]);
4425         if (nritems == 0)
4426                 return 1;
4427
4428         btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4429 again:
4430         level = 1;
4431         next = NULL;
4432         btrfs_release_path(path);
4433
4434         path->keep_locks = 1;
4435
4436         if (time_seq)
4437                 ret = btrfs_search_old_slot(root, &key, path, time_seq);
4438         else
4439                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4440         path->keep_locks = 0;
4441
4442         if (ret < 0)
4443                 return ret;
4444
4445         nritems = btrfs_header_nritems(path->nodes[0]);
4446         /*
4447          * by releasing the path above we dropped all our locks.  A balance
4448          * could have added more items next to the key that used to be
4449          * at the very end of the block.  So, check again here and
4450          * advance the path if there are now more items available.
4451          */
4452         if (nritems > 0 && path->slots[0] < nritems - 1) {
4453                 if (ret == 0)
4454                         path->slots[0]++;
4455                 ret = 0;
4456                 goto done;
4457         }
4458         /*
4459          * So the above check misses one case:
4460          * - after releasing the path above, someone has removed the item that
4461          *   used to be at the very end of the block, and balance between leafs
4462          *   gets another one with bigger key.offset to replace it.
4463          *
4464          * This one should be returned as well, or we can get leaf corruption
4465          * later(esp. in __btrfs_drop_extents()).
4466          *
4467          * And a bit more explanation about this check,
4468          * with ret > 0, the key isn't found, the path points to the slot
4469          * where it should be inserted, so the path->slots[0] item must be the
4470          * bigger one.
4471          */
4472         if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
4473                 ret = 0;
4474                 goto done;
4475         }
4476
4477         while (level < BTRFS_MAX_LEVEL) {
4478                 if (!path->nodes[level]) {
4479                         ret = 1;
4480                         goto done;
4481                 }
4482
4483                 slot = path->slots[level] + 1;
4484                 c = path->nodes[level];
4485                 if (slot >= btrfs_header_nritems(c)) {
4486                         level++;
4487                         if (level == BTRFS_MAX_LEVEL) {
4488                                 ret = 1;
4489                                 goto done;
4490                         }
4491                         continue;
4492                 }
4493
4494
4495                 /*
4496                  * Our current level is where we're going to start from, and to
4497                  * make sure lockdep doesn't complain we need to drop our locks
4498                  * and nodes from 0 to our current level.
4499                  */
4500                 for (i = 0; i < level; i++) {
4501                         if (path->locks[level]) {
4502                                 btrfs_tree_read_unlock(path->nodes[i]);
4503                                 path->locks[i] = 0;
4504                         }
4505                         free_extent_buffer(path->nodes[i]);
4506                         path->nodes[i] = NULL;
4507                 }
4508
4509                 next = c;
4510                 ret = read_block_for_search(root, path, &next, level,
4511                                             slot, &key);
4512                 if (ret == -EAGAIN)
4513                         goto again;
4514
4515                 if (ret < 0) {
4516                         btrfs_release_path(path);
4517                         goto done;
4518                 }
4519
4520                 if (!path->skip_locking) {
4521                         ret = btrfs_try_tree_read_lock(next);
4522                         if (!ret && time_seq) {
4523                                 /*
4524                                  * If we don't get the lock, we may be racing
4525                                  * with push_leaf_left, holding that lock while
4526                                  * itself waiting for the leaf we've currently
4527                                  * locked. To solve this situation, we give up
4528                                  * on our lock and cycle.
4529                                  */
4530                                 free_extent_buffer(next);
4531                                 btrfs_release_path(path);
4532                                 cond_resched();
4533                                 goto again;
4534                         }
4535                         if (!ret)
4536                                 btrfs_tree_read_lock(next);
4537                 }
4538                 break;
4539         }
4540         path->slots[level] = slot;
4541         while (1) {
4542                 level--;
4543                 path->nodes[level] = next;
4544                 path->slots[level] = 0;
4545                 if (!path->skip_locking)
4546                         path->locks[level] = BTRFS_READ_LOCK;
4547                 if (!level)
4548                         break;
4549
4550                 ret = read_block_for_search(root, path, &next, level,
4551                                             0, &key);
4552                 if (ret == -EAGAIN)
4553                         goto again;
4554
4555                 if (ret < 0) {
4556                         btrfs_release_path(path);
4557                         goto done;
4558                 }
4559
4560                 if (!path->skip_locking)
4561                         btrfs_tree_read_lock(next);
4562         }
4563         ret = 0;
4564 done:
4565         unlock_up(path, 0, 1, 0, NULL);
4566
4567         return ret;
4568 }
4569
4570 /*
4571  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4572  * searching until it gets past min_objectid or finds an item of 'type'
4573  *
4574  * returns 0 if something is found, 1 if nothing was found and < 0 on error
4575  */
4576 int btrfs_previous_item(struct btrfs_root *root,
4577                         struct btrfs_path *path, u64 min_objectid,
4578                         int type)
4579 {
4580         struct btrfs_key found_key;
4581         struct extent_buffer *leaf;
4582         u32 nritems;
4583         int ret;
4584
4585         while (1) {
4586                 if (path->slots[0] == 0) {
4587                         ret = btrfs_prev_leaf(root, path);
4588                         if (ret != 0)
4589                                 return ret;
4590                 } else {
4591                         path->slots[0]--;
4592                 }
4593                 leaf = path->nodes[0];
4594                 nritems = btrfs_header_nritems(leaf);
4595                 if (nritems == 0)
4596                         return 1;
4597                 if (path->slots[0] == nritems)
4598                         path->slots[0]--;
4599
4600                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4601                 if (found_key.objectid < min_objectid)
4602                         break;
4603                 if (found_key.type == type)
4604                         return 0;
4605                 if (found_key.objectid == min_objectid &&
4606                     found_key.type < type)
4607                         break;
4608         }
4609         return 1;
4610 }
4611
4612 /*
4613  * search in extent tree to find a previous Metadata/Data extent item with
4614  * min objecitd.
4615  *
4616  * returns 0 if something is found, 1 if nothing was found and < 0 on error
4617  */
4618 int btrfs_previous_extent_item(struct btrfs_root *root,
4619                         struct btrfs_path *path, u64 min_objectid)
4620 {
4621         struct btrfs_key found_key;
4622         struct extent_buffer *leaf;
4623         u32 nritems;
4624         int ret;
4625
4626         while (1) {
4627                 if (path->slots[0] == 0) {
4628                         ret = btrfs_prev_leaf(root, path);
4629                         if (ret != 0)
4630                                 return ret;
4631                 } else {
4632                         path->slots[0]--;
4633                 }
4634                 leaf = path->nodes[0];
4635                 nritems = btrfs_header_nritems(leaf);
4636                 if (nritems == 0)
4637                         return 1;
4638                 if (path->slots[0] == nritems)
4639                         path->slots[0]--;
4640
4641                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4642                 if (found_key.objectid < min_objectid)
4643                         break;
4644                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
4645                     found_key.type == BTRFS_METADATA_ITEM_KEY)
4646                         return 0;
4647                 if (found_key.objectid == min_objectid &&
4648                     found_key.type < BTRFS_EXTENT_ITEM_KEY)
4649                         break;
4650         }
4651         return 1;
4652 }