fs/btrfs/ctree.c

   1 /*
   2  * Copyright (C) 2007,2008 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/sched.h>
  20 #include "ctree.h"
  21 #include "disk-io.h"
  22 #include "transaction.h"
  23 #include "print-tree.h"
  24 #include "locking.h"
  25
  26 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
  27                       *root, struct btrfs_path *path, int level);
  28 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
  29                       *root, struct btrfs_key *ins_key,
  30                       struct btrfs_path *path, int data_size, int extend);
  31 static int push_node_left(struct btrfs_trans_handle *trans,
  32                           struct btrfs_root *root, struct extent_buffer *dst,
  33                           struct extent_buffer *src, int empty);
  34 static int balance_node_right(struct btrfs_trans_handle *trans,
  35                               struct btrfs_root *root,
  36                               struct extent_buffer *dst_buf,
  37                               struct extent_buffer *src_buf);
  38 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  39                    struct btrfs_path *path, int level, int slot);
  40
  41 struct btrfs_path *btrfs_alloc_path(void)
  42 {
  43         struct btrfs_path *path;
  44         path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
  45         if (path)
  46                 path->reada = 1;
  47         return path;
  48 }
  49
  50 /*
  51  * set all locked nodes in the path to blocking locks.  This should
  52  * be done before scheduling
  53  */
  54 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
  55 {
  56         int i;
  57         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
  58                 if (p->nodes[i] && p->locks[i])
  59                         btrfs_set_lock_blocking(p->nodes[i]);
  60         }
  61 }
  62
  63 /*
  64  * reset all the locked nodes in the patch to spinning locks.
  65  *
  66  * held is used to keep lockdep happy, when lockdep is enabled
  67  * we set held to a blocking lock before we go around and
  68  * retake all the spinlocks in the path.  You can safely use NULL
  69  * for held
  70  */
  71 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
  72                                         struct extent_buffer *held)
  73 {
  74         int i;
  75
  76 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  77         /* lockdep really cares that we take all of these spinlocks
  78          * in the right order.  If any of the locks in the path are not
  79          * currently blocking, it is going to complain.  So, make really
  80          * really sure by forcing the path to blocking before we clear
  81          * the path blocking.
  82          */
  83         if (held)
  84                 btrfs_set_lock_blocking(held);
  85         btrfs_set_path_blocking(p);
  86 #endif
  87
  88         for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
  89                 if (p->nodes[i] && p->locks[i])
  90                         btrfs_clear_lock_blocking(p->nodes[i]);
  91         }
  92
  93 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  94         if (held)
  95                 btrfs_clear_lock_blocking(held);
  96 #endif
  97 }
  98
  99 /* this also releases the path */
 100 void btrfs_free_path(struct btrfs_path *p)
 101 {
 102         btrfs_release_path(NULL, p);
 103         kmem_cache_free(btrfs_path_cachep, p);
 104 }
 105
 106 /*
 107  * path release drops references on the extent buffers in the path
 108  * and it drops any locks held by this path
 109  *
 110  * It is safe to call this on paths that no locks or extent buffers held.
 111  */
 112 noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
 113 {
 114         int i;
 115
 116         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
 117                 p->slots[i] = 0;
 118                 if (!p->nodes[i])
 119                         continue;
 120                 if (p->locks[i]) {
 121                         btrfs_tree_unlock(p->nodes[i]);
 122                         p->locks[i] = 0;
 123                 }
 124                 free_extent_buffer(p->nodes[i]);
 125                 p->nodes[i] = NULL;
 126         }
 127 }
 128
 129 /*
 130  * safely gets a reference on the root node of a tree.  A lock
 131  * is not taken, so a concurrent writer may put a different node
 132  * at the root of the tree.  See btrfs_lock_root_node for the
 133  * looping required.
 134  *
 135  * The extent buffer returned by this has a reference taken, so
 136  * it won't disappear.  It may stop being the root of the tree
 137  * at any time because there are no locks held.
 138  */
 139 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
 140 {
 141         struct extent_buffer *eb;
 142         spin_lock(&root->node_lock);
 143         eb = root->node;
 144         extent_buffer_get(eb);
 145         spin_unlock(&root->node_lock);
 146         return eb;
 147 }
 148
 149 /* loop around taking references on and locking the root node of the
 150  * tree until you end up with a lock on the root.  A locked buffer
 151  * is returned, with a reference held.
 152  */
 153 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
 154 {
 155         struct extent_buffer *eb;
 156
 157         while (1) {
 158                 eb = btrfs_root_node(root);
 159                 btrfs_tree_lock(eb);
 160
 161                 spin_lock(&root->node_lock);
 162                 if (eb == root->node) {
 163                         spin_unlock(&root->node_lock);
 164                         break;
 165                 }
 166                 spin_unlock(&root->node_lock);
 167
 168                 btrfs_tree_unlock(eb);
 169                 free_extent_buffer(eb);
 170         }
 171         return eb;
 172 }
 173
 174 /* cowonly root (everything not a reference counted cow subvolume), just get
 175  * put onto a simple dirty list.  transaction.c walks this to make sure they
 176  * get properly updated on disk.
 177  */
 178 static void add_root_to_dirty_list(struct btrfs_root *root)
 179 {
 180         if (root->track_dirty && list_empty(&root->dirty_list)) {
 181                 list_add(&root->dirty_list,
 182                          &root->fs_info->dirty_cowonly_roots);
 183         }
 184 }
 185
 186 /*
 187  * used by snapshot creation to make a copy of a root for a tree with
 188  * a given objectid.  The buffer with the new root node is returned in
 189  * cow_ret, and this func returns zero on success or a negative error code.
 190  */
 191 int btrfs_copy_root(struct btrfs_trans_handle *trans,
 192                       struct btrfs_root *root,
 193                       struct extent_buffer *buf,
 194                       struct extent_buffer **cow_ret, u64 new_root_objectid)
 195 {
 196         struct extent_buffer *cow;
 197         u32 nritems;
 198         int ret = 0;
 199         int level;
 200         struct btrfs_disk_key disk_key;
 201
 202         WARN_ON(root->ref_cows && trans->transid !=
 203                 root->fs_info->running_transaction->transid);
 204         WARN_ON(root->ref_cows && trans->transid != root->last_trans);
 205
 206         level = btrfs_header_level(buf);
 207         nritems = btrfs_header_nritems(buf);
 208         if (level == 0)
 209                 btrfs_item_key(buf, &disk_key, 0);
 210         else
 211                 btrfs_node_key(buf, &disk_key, 0);
 212
 213         cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
 214                                      new_root_objectid, &disk_key, level,
 215                                      buf->start, 0);
 216         if (IS_ERR(cow))
 217                 return PTR_ERR(cow);
 218
 219         copy_extent_buffer(cow, buf, 0, 0, cow->len);
 220         btrfs_set_header_bytenr(cow, cow->start);
 221         btrfs_set_header_generation(cow, trans->transid);
 222         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
 223         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
 224                                      BTRFS_HEADER_FLAG_RELOC);
 225         if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
 226                 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
 227         else
 228                 btrfs_set_header_owner(cow, new_root_objectid);
 229
 230         write_extent_buffer(cow, root->fs_info->fsid,
 231                             (unsigned long)btrfs_header_fsid(cow),
 232                             BTRFS_FSID_SIZE);
 233
 234         WARN_ON(btrfs_header_generation(buf) > trans->transid);
 235         if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
 236                 ret = btrfs_inc_ref(trans, root, cow, 1);
 237         else
 238                 ret = btrfs_inc_ref(trans, root, cow, 0);
 239
 240         if (ret)
 241                 return ret;
 242
 243         btrfs_mark_buffer_dirty(cow);
 244         *cow_ret = cow;
 245         return 0;
 246 }
 247
 248 /*
 249  * check if the tree block can be shared by multiple trees
 250  */
 251 int btrfs_block_can_be_shared(struct btrfs_root *root,
 252                               struct extent_buffer *buf)
 253 {
 254         /*
 255          * Tree blocks not in refernece counted trees and tree roots
 256          * are never shared. If a block was allocated after the last
 257          * snapshot and the block was not allocated by tree relocation,
 258          * we know the block is not shared.
 259          */
 260         if (root->ref_cows &&
 261             buf != root->node && buf != root->commit_root &&
 262             (btrfs_header_generation(buf) <=
 263              btrfs_root_last_snapshot(&root->root_item) ||
 264              btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
 265                 return 1;
 266 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 267         if (root->ref_cows &&
 268             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
 269                 return 1;
 270 #endif
 271         return 0;
 272 }
 273
 274 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
 275                                        struct btrfs_root *root,
 276                                        struct extent_buffer *buf,
 277                                        struct extent_buffer *cow)
 278 {
 279         u64 refs;
 280         u64 owner;
 281         u64 flags;
 282         u64 new_flags = 0;
 283         int ret;
 284
 285         /*
 286          * Backrefs update rules:
 287          *
 288          * Always use full backrefs for extent pointers in tree block
 289          * allocated by tree relocation.
 290          *
 291          * If a shared tree block is no longer referenced by its owner
 292          * tree (btrfs_header_owner(buf) == root->root_key.objectid),
 293          * use full backrefs for extent pointers in tree block.
 294          *
 295          * If a tree block is been relocating
 296          * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
 297          * use full backrefs for extent pointers in tree block.
 298          * The reason for this is some operations (such as drop tree)
 299          * are only allowed for blocks use full backrefs.
 300          */
 301
 302         if (btrfs_block_can_be_shared(root, buf)) {
 303                 ret = btrfs_lookup_extent_info(trans, root, buf->start,
 304                                                buf->len, &refs, &flags);
 305                 BUG_ON(ret);
 306                 BUG_ON(refs == 0);
 307         } else {
 308                 refs = 1;
 309                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
 310                     btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
 311                         flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 312                 else
 313                         flags = 0;
 314         }
 315
 316         owner = btrfs_header_owner(buf);
 317         BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
 318                !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
 319
 320         if (refs > 1) {
 321                 if ((owner == root->root_key.objectid ||
 322                      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
 323                     !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
 324                         ret = btrfs_inc_ref(trans, root, buf, 1);
 325                         BUG_ON(ret);
 326
 327                         if (root->root_key.objectid ==
 328                             BTRFS_TREE_RELOC_OBJECTID) {
 329                                 ret = btrfs_dec_ref(trans, root, buf, 0);
 330                                 BUG_ON(ret);
 331                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 332                                 BUG_ON(ret);
 333                         }
 334                         new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
 335                 } else {
 336
 337                         if (root->root_key.objectid ==
 338                             BTRFS_TREE_RELOC_OBJECTID)
 339                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 340                         else
 341                                 ret = btrfs_inc_ref(trans, root, cow, 0);
 342                         BUG_ON(ret);
 343                 }
 344                 if (new_flags != 0) {
 345                         ret = btrfs_set_disk_extent_flags(trans, root,
 346                                                           buf->start,
 347                                                           buf->len,
 348                                                           new_flags, 0);
 349                         BUG_ON(ret);
 350                 }
 351         } else {
 352                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
 353                         if (root->root_key.objectid ==
 354                             BTRFS_TREE_RELOC_OBJECTID)
 355                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 356                         else
 357                                 ret = btrfs_inc_ref(trans, root, cow, 0);
 358                         BUG_ON(ret);
 359                         ret = btrfs_dec_ref(trans, root, buf, 1);
 360                         BUG_ON(ret);
 361                 }
 362                 clean_tree_block(trans, root, buf);
 363         }
 364         return 0;
 365 }
 366
 367 /*
 368  * does the dirty work in cow of a single block.  The parent block (if
 369  * supplied) is updated to point to the new cow copy.  The new buffer is marked
 370  * dirty and returned locked.  If you modify the block it needs to be marked
 371  * dirty again.
 372  *
 373  * search_start -- an allocation hint for the new block
 374  *
 375  * empty_size -- a hint that you plan on doing more cow.  This is the size in
 376  * bytes the allocator should try to find free next to the block it returns.
 377  * This is just a hint and may be ignored by the allocator.
 378  */
 379 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 380                              struct btrfs_root *root,
 381                              struct extent_buffer *buf,
 382                              struct extent_buffer *parent, int parent_slot,
 383                              struct extent_buffer **cow_ret,
 384                              u64 search_start, u64 empty_size)
 385 {
 386         struct btrfs_disk_key disk_key;
 387         struct extent_buffer *cow;
 388         int level;
 389         int unlock_orig = 0;
 390         u64 parent_start;
 391
 392         if (*cow_ret == buf)
 393                 unlock_orig = 1;
 394
 395         btrfs_assert_tree_locked(buf);
 396
 397         WARN_ON(root->ref_cows && trans->transid !=
 398                 root->fs_info->running_transaction->transid);
 399         WARN_ON(root->ref_cows && trans->transid != root->last_trans);
 400
 401         level = btrfs_header_level(buf);
 402
 403         if (level == 0)
 404                 btrfs_item_key(buf, &disk_key, 0);
 405         else
 406                 btrfs_node_key(buf, &disk_key, 0);
 407
 408         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
 409                 if (parent)
 410                         parent_start = parent->start;
 411                 else
 412                         parent_start = 0;
 413         } else
 414                 parent_start = 0;
 415
 416         cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
 417                                      root->root_key.objectid, &disk_key,
 418                                      level, search_start, empty_size);
 419         if (IS_ERR(cow))
 420                 return PTR_ERR(cow);
 421
 422         /* cow is set to blocking by btrfs_init_new_buffer */
 423
 424         copy_extent_buffer(cow, buf, 0, 0, cow->len);
 425         btrfs_set_header_bytenr(cow, cow->start);
 426         btrfs_set_header_generation(cow, trans->transid);
 427         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
 428         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
 429                                      BTRFS_HEADER_FLAG_RELOC);
 430         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
 431                 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
 432         else
 433                 btrfs_set_header_owner(cow, root->root_key.objectid);
 434
 435         write_extent_buffer(cow, root->fs_info->fsid,
 436                             (unsigned long)btrfs_header_fsid(cow),
 437                             BTRFS_FSID_SIZE);
 438
 439         update_ref_for_cow(trans, root, buf, cow);
 440
 441         if (buf == root->node) {
 442                 WARN_ON(parent && parent != buf);
 443                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
 444                     btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
 445                         parent_start = buf->start;
 446                 else
 447                         parent_start = 0;
 448
 449                 spin_lock(&root->node_lock);
 450                 root->node = cow;
 451                 extent_buffer_get(cow);
 452                 spin_unlock(&root->node_lock);
 453
 454                 btrfs_free_extent(trans, root, buf->start, buf->len,
 455                                   parent_start, root->root_key.objectid,
 456                                   level, 0);
 457                 free_extent_buffer(buf);
 458                 add_root_to_dirty_list(root);
 459         } else {
 460                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
 461                         parent_start = parent->start;
 462                 else
 463                         parent_start = 0;
 464
 465                 WARN_ON(trans->transid != btrfs_header_generation(parent));
 466                 btrfs_set_node_blockptr(parent, parent_slot,
 467                                         cow->start);
 468                 btrfs_set_node_ptr_generation(parent, parent_slot,
 469                                               trans->transid);
 470                 btrfs_mark_buffer_dirty(parent);
 471                 btrfs_free_extent(trans, root, buf->start, buf->len,
 472                                   parent_start, root->root_key.objectid,
 473                                   level, 0);
 474         }
 475         if (unlock_orig)
 476                 btrfs_tree_unlock(buf);
 477         free_extent_buffer(buf);
 478         btrfs_mark_buffer_dirty(cow);
 479         *cow_ret = cow;
 480         return 0;
 481 }
 482
 483 static inline int should_cow_block(struct btrfs_trans_handle *trans,
 484                                    struct btrfs_root *root,
 485                                    struct extent_buffer *buf)
 486 {
 487         if (btrfs_header_generation(buf) == trans->transid &&
 488             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
 489             !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
 490               btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
 491                 return 0;
 492         return 1;
 493 }
 494
 495 /*
 496  * cows a single block, see __btrfs_cow_block for the real work.
 497  * This version of it has extra checks so that a block isn't cow'd more than
 498  * once per transaction, as long as it hasn't been written yet
 499  */
 500 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
 501                     struct btrfs_root *root, struct extent_buffer *buf,
 502                     struct extent_buffer *parent, int parent_slot,
 503                     struct extent_buffer **cow_ret)
 504 {
 505         u64 search_start;
 506         int ret;
 507
 508         if (trans->transaction != root->fs_info->running_transaction) {
 509                 printk(KERN_CRIT "trans %llu running %llu\n",
 510                        (unsigned long long)trans->transid,
 511                        (unsigned long long)
 512                        root->fs_info->running_transaction->transid);
 513                 WARN_ON(1);
 514         }
 515         if (trans->transid != root->fs_info->generation) {
 516                 printk(KERN_CRIT "trans %llu running %llu\n",
 517                        (unsigned long long)trans->transid,
 518                        (unsigned long long)root->fs_info->generation);
 519                 WARN_ON(1);
 520         }
 521
 522         if (!should_cow_block(trans, root, buf)) {
 523                 *cow_ret = buf;
 524                 return 0;
 525         }
 526
 527         search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
 528
 529         if (parent)
 530                 btrfs_set_lock_blocking(parent);
 531         btrfs_set_lock_blocking(buf);
 532
 533         ret = __btrfs_cow_block(trans, root, buf, parent,
 534                                  parent_slot, cow_ret, search_start, 0);
 535         return ret;
 536 }
 537
 538 /*
 539  * helper function for defrag to decide if two blocks pointed to by a
 540  * node are actually close by
 541  */
 542 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
 543 {
 544         if (blocknr < other && other - (blocknr + blocksize) < 32768)
 545                 return 1;
 546         if (blocknr > other && blocknr - (other + blocksize) < 32768)
 547                 return 1;
 548         return 0;
 549 }
 550
 551 /*
 552  * compare two keys in a memcmp fashion
 553  */
 554 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
 555 {
 556         struct btrfs_key k1;
 557
 558         btrfs_disk_key_to_cpu(&k1, disk);
 559
 560         return btrfs_comp_cpu_keys(&k1, k2);
 561 }
 562
 563 /*
 564  * same as comp_keys only with two btrfs_key's
 565  */
 566 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
 567 {
 568         if (k1->objectid > k2->objectid)
 569                 return 1;
 570         if (k1->objectid < k2->objectid)
 571                 return -1;
 572         if (k1->type > k2->type)
 573                 return 1;
 574         if (k1->type < k2->type)
 575                 return -1;
 576         if (k1->offset > k2->offset)
 577                 return 1;
 578         if (k1->offset < k2->offset)
 579                 return -1;
 580         return 0;
 581 }
 582
 583 /*
 584  * this is used by the defrag code to go through all the
 585  * leaves pointed to by a node and reallocate them so that
 586  * disk order is close to key order
 587  */
 588 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
 589                        struct btrfs_root *root, struct extent_buffer *parent,
 590                        int start_slot, int cache_only, u64 *last_ret,
 591                        struct btrfs_key *progress)
 592 {
 593         struct extent_buffer *cur;
 594         u64 blocknr;
 595         u64 gen;
 596         u64 search_start = *last_ret;
 597         u64 last_block = 0;
 598         u64 other;
 599         u32 parent_nritems;
 600         int end_slot;
 601         int i;
 602         int err = 0;
 603         int parent_level;
 604         int uptodate;
 605         u32 blocksize;
 606         int progress_passed = 0;
 607         struct btrfs_disk_key disk_key;
 608
 609         parent_level = btrfs_header_level(parent);
 610         if (cache_only && parent_level != 1)
 611                 return 0;
 612
 613         if (trans->transaction != root->fs_info->running_transaction)
 614                 WARN_ON(1);
 615         if (trans->transid != root->fs_info->generation)
 616                 WARN_ON(1);
 617
 618         parent_nritems = btrfs_header_nritems(parent);
 619         blocksize = btrfs_level_size(root, parent_level - 1);
 620         end_slot = parent_nritems;
 621
 622         if (parent_nritems == 1)
 623                 return 0;
 624
 625         btrfs_set_lock_blocking(parent);
 626
 627         for (i = start_slot; i < end_slot; i++) {
 628                 int close = 1;
 629
 630                 if (!parent->map_token) {
 631                         map_extent_buffer(parent,
 632                                         btrfs_node_key_ptr_offset(i),
 633                                         sizeof(struct btrfs_key_ptr),
 634                                         &parent->map_token, &parent->kaddr,
 635                                         &parent->map_start, &parent->map_len,
 636                                         KM_USER1);
 637                 }
 638                 btrfs_node_key(parent, &disk_key, i);
 639                 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
 640                         continue;
 641
 642                 progress_passed = 1;
 643                 blocknr = btrfs_node_blockptr(parent, i);
 644                 gen = btrfs_node_ptr_generation(parent, i);
 645                 if (last_block == 0)
 646                         last_block = blocknr;
 647
 648                 if (i > 0) {
 649                         other = btrfs_node_blockptr(parent, i - 1);
 650                         close = close_blocks(blocknr, other, blocksize);
 651                 }
 652                 if (!close && i < end_slot - 2) {
 653                         other = btrfs_node_blockptr(parent, i + 1);
 654                         close = close_blocks(blocknr, other, blocksize);
 655                 }
 656                 if (close) {
 657                         last_block = blocknr;
 658                         continue;
 659                 }
 660                 if (parent->map_token) {
 661                         unmap_extent_buffer(parent, parent->map_token,
 662                                             KM_USER1);
 663                         parent->map_token = NULL;
 664                 }
 665
 666                 cur = btrfs_find_tree_block(root, blocknr, blocksize);
 667                 if (cur)
 668                         uptodate = btrfs_buffer_uptodate(cur, gen);
 669                 else
 670                         uptodate = 0;
 671                 if (!cur || !uptodate) {
 672                         if (cache_only) {
 673                                 free_extent_buffer(cur);
 674                                 continue;
 675                         }
 676                         if (!cur) {
 677                                 cur = read_tree_block(root, blocknr,
 678                                                          blocksize, gen);
 679                         } else if (!uptodate) {
 680                                 btrfs_read_buffer(cur, gen);
 681                         }
 682                 }
 683                 if (search_start == 0)
 684                         search_start = last_block;
 685
 686                 btrfs_tree_lock(cur);
 687                 btrfs_set_lock_blocking(cur);
 688                 err = __btrfs_cow_block(trans, root, cur, parent, i,
 689                                         &cur, search_start,
 690                                         min(16 * blocksize,
 691                                             (end_slot - i) * blocksize));
 692                 if (err) {
 693                         btrfs_tree_unlock(cur);
 694                         free_extent_buffer(cur);
 695                         break;
 696                 }
 697                 search_start = cur->start;
 698                 last_block = cur->start;
 699                 *last_ret = search_start;
 700                 btrfs_tree_unlock(cur);
 701                 free_extent_buffer(cur);
 702         }
 703         if (parent->map_token) {
 704                 unmap_extent_buffer(parent, parent->map_token,
 705                                     KM_USER1);
 706                 parent->map_token = NULL;
 707         }
 708         return err;
 709 }
 710
 711 /*
 712  * The leaf data grows from end-to-front in the node.
 713  * this returns the address of the start of the last item,
 714  * which is the stop of the leaf data stack
 715  */
 716 static inline unsigned int leaf_data_end(struct btrfs_root *root,
 717                                          struct extent_buffer *leaf)
 718 {
 719         u32 nr = btrfs_header_nritems(leaf);
 720         if (nr == 0)
 721                 return BTRFS_LEAF_DATA_SIZE(root);
 722         return btrfs_item_offset_nr(leaf, nr - 1);
 723 }
 724
 725 /*
 726  * extra debugging checks to make sure all the items in a key are
 727  * well formed and in the proper order
 728  */
 729 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
 730                       int level)
 731 {
 732         struct extent_buffer *parent = NULL;
 733         struct extent_buffer *node = path->nodes[level];
 734         struct btrfs_disk_key parent_key;
 735         struct btrfs_disk_key node_key;
 736         int parent_slot;
 737         int slot;
 738         struct btrfs_key cpukey;
 739         u32 nritems = btrfs_header_nritems(node);
 740
 741         if (path->nodes[level + 1])
 742                 parent = path->nodes[level + 1];
 743
 744         slot = path->slots[level];
 745         BUG_ON(nritems == 0);
 746         if (parent) {
 747                 parent_slot = path->slots[level + 1];
 748                 btrfs_node_key(parent, &parent_key, parent_slot);
 749                 btrfs_node_key(node, &node_key, 0);
 750                 BUG_ON(memcmp(&parent_key, &node_key,
 751                               sizeof(struct btrfs_disk_key)));
 752                 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
 753                        btrfs_header_bytenr(node));
 754         }
 755         BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
 756         if (slot != 0) {
 757                 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
 758                 btrfs_node_key(node, &node_key, slot);
 759                 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
 760         }
 761         if (slot < nritems - 1) {
 762                 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
 763                 btrfs_node_key(node, &node_key, slot);
 764                 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
 765         }
 766         return 0;
 767 }
 768
 769 /*
 770  * extra checking to make sure all the items in a leaf are
 771  * well formed and in the proper order
 772  */
 773 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
 774                       int level)
 775 {
 776         struct extent_buffer *leaf = path->nodes[level];
 777         struct extent_buffer *parent = NULL;
 778         int parent_slot;
 779         struct btrfs_key cpukey;
 780         struct btrfs_disk_key parent_key;
 781         struct btrfs_disk_key leaf_key;
 782         int slot = path->slots[0];
 783
 784         u32 nritems = btrfs_header_nritems(leaf);
 785
 786         if (path->nodes[level + 1])
 787                 parent = path->nodes[level + 1];
 788
 789         if (nritems == 0)
 790                 return 0;
 791
 792         if (parent) {
 793                 parent_slot = path->slots[level + 1];
 794                 btrfs_node_key(parent, &parent_key, parent_slot);
 795                 btrfs_item_key(leaf, &leaf_key, 0);
 796
 797                 BUG_ON(memcmp(&parent_key, &leaf_key,
 798                        sizeof(struct btrfs_disk_key)));
 799                 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
 800                        btrfs_header_bytenr(leaf));
 801         }
 802         if (slot != 0 && slot < nritems - 1) {
 803                 btrfs_item_key(leaf, &leaf_key, slot);
 804                 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
 805                 if (comp_keys(&leaf_key, &cpukey) <= 0) {
 806                         btrfs_print_leaf(root, leaf);
 807                         printk(KERN_CRIT "slot %d offset bad key\n", slot);
 808                         BUG_ON(1);
 809                 }
 810                 if (btrfs_item_offset_nr(leaf, slot - 1) !=
 811                        btrfs_item_end_nr(leaf, slot)) {
 812                         btrfs_print_leaf(root, leaf);
 813                         printk(KERN_CRIT "slot %d offset bad\n", slot);
 814                         BUG_ON(1);
 815                 }
 816         }
 817         if (slot < nritems - 1) {
 818                 btrfs_item_key(leaf, &leaf_key, slot);
 819                 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
 820                 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
 821                 if (btrfs_item_offset_nr(leaf, slot) !=
 822                         btrfs_item_end_nr(leaf, slot + 1)) {
 823                         btrfs_print_leaf(root, leaf);
 824                         printk(KERN_CRIT "slot %d offset bad\n", slot);
 825                         BUG_ON(1);
 826                 }
 827         }
 828         BUG_ON(btrfs_item_offset_nr(leaf, 0) +
 829                btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
 830         return 0;
 831 }
 832
 833 static noinline int check_block(struct btrfs_root *root,
 834                                 struct btrfs_path *path, int level)
 835 {
 836         return 0;
 837         if (level == 0)
 838                 return check_leaf(root, path, level);
 839         return check_node(root, path, level);
 840 }
 841
 842 /*
 843  * search for key in the extent_buffer.  The items start at offset p,
 844  * and they are item_size apart.  There are 'max' items in p.
 845  *
 846  * the slot in the array is returned via slot, and it points to
 847  * the place where you would insert key if it is not found in
 848  * the array.
 849  *
 850  * slot may point to max if the key is bigger than all of the keys
 851  */
 852 static noinline int generic_bin_search(struct extent_buffer *eb,
 853                                        unsigned long p,
 854                                        int item_size, struct btrfs_key *key,
 855                                        int max, int *slot)
 856 {
 857         int low = 0;
 858         int high = max;
 859         int mid;
 860         int ret;
 861         struct btrfs_disk_key *tmp = NULL;
 862         struct btrfs_disk_key unaligned;
 863         unsigned long offset;
 864         char *map_token = NULL;
 865         char *kaddr = NULL;
 866         unsigned long map_start = 0;
 867         unsigned long map_len = 0;
 868         int err;
 869
 870         while (low < high) {
 871                 mid = (low + high) / 2;
 872                 offset = p + mid * item_size;
 873
 874                 if (!map_token || offset < map_start ||
 875                     (offset + sizeof(struct btrfs_disk_key)) >
 876                     map_start + map_len) {
 877                         if (map_token) {
 878                                 unmap_extent_buffer(eb, map_token, KM_USER0);
 879                                 map_token = NULL;
 880                         }
 881
 882                         err = map_private_extent_buffer(eb, offset,
 883                                                 sizeof(struct btrfs_disk_key),
 884                                                 &map_token, &kaddr,
 885                                                 &map_start, &map_len, KM_USER0);
 886
 887                         if (!err) {
 888                                 tmp = (struct btrfs_disk_key *)(kaddr + offset -
 889                                                         map_start);
 890                         } else {
 891                                 read_extent_buffer(eb, &unaligned,
 892                                                    offset, sizeof(unaligned));
 893                                 tmp = &unaligned;
 894                         }
 895
 896                 } else {
 897                         tmp = (struct btrfs_disk_key *)(kaddr + offset -
 898                                                         map_start);
 899                 }
 900                 ret = comp_keys(tmp, key);
 901
 902                 if (ret < 0)
 903                         low = mid + 1;
 904                 else if (ret > 0)
 905                         high = mid;
 906                 else {
 907                         *slot = mid;
 908                         if (map_token)
 909                                 unmap_extent_buffer(eb, map_token, KM_USER0);
 910                         return 0;
 911                 }
 912         }
 913         *slot = low;
 914         if (map_token)
 915                 unmap_extent_buffer(eb, map_token, KM_USER0);
 916         return 1;
 917 }
 918
 919 /*
 920  * simple bin_search frontend that does the right thing for
 921  * leaves vs nodes
 922  */
 923 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
 924                       int level, int *slot)
 925 {
 926         if (level == 0) {
 927                 return generic_bin_search(eb,
 928                                           offsetof(struct btrfs_leaf, items),
 929                                           sizeof(struct btrfs_item),
 930                                           key, btrfs_header_nritems(eb),
 931                                           slot);
 932         } else {
 933                 return generic_bin_search(eb,
 934                                           offsetof(struct btrfs_node, ptrs),
 935                                           sizeof(struct btrfs_key_ptr),
 936                                           key, btrfs_header_nritems(eb),
 937                                           slot);
 938         }
 939         return -1;
 940 }
 941
 942 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
 943                      int level, int *slot)
 944 {
 945         return bin_search(eb, key, level, slot);
 946 }
 947
 948 /* given a node and slot number, this reads the blocks it points to.  The
 949  * extent buffer is returned with a reference taken (but unlocked).
 950  * NULL is returned on error.
 951  */
 952 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
 953                                    struct extent_buffer *parent, int slot)
 954 {
 955         int level = btrfs_header_level(parent);
 956         if (slot < 0)
 957                 return NULL;
 958         if (slot >= btrfs_header_nritems(parent))
 959                 return NULL;
 960
 961         BUG_ON(level == 0);
 962
 963         return read_tree_block(root, btrfs_node_blockptr(parent, slot),
 964                        btrfs_level_size(root, level - 1),
 965                        btrfs_node_ptr_generation(parent, slot));
 966 }
 967
 968 /*
 969  * node level balancing, used to make sure nodes are in proper order for
 970  * item deletion.  We balance from the top down, so we have to make sure
 971  * that a deletion won't leave an node completely empty later on.
 972  */
 973 static noinline int balance_level(struct btrfs_trans_handle *trans,
 974                          struct btrfs_root *root,
 975                          struct btrfs_path *path, int level)
 976 {
 977         struct extent_buffer *right = NULL;
 978         struct extent_buffer *mid;
 979         struct extent_buffer *left = NULL;
 980         struct extent_buffer *parent = NULL;
 981         int ret = 0;
 982         int wret;
 983         int pslot;
 984         int orig_slot = path->slots[level];
 985         int err_on_enospc = 0;
 986         u64 orig_ptr;
 987
 988         if (level == 0)
 989                 return 0;
 990
 991         mid = path->nodes[level];
 992
 993         WARN_ON(!path->locks[level]);
 994         WARN_ON(btrfs_header_generation(mid) != trans->transid);
 995
 996         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
 997
 998         if (level < BTRFS_MAX_LEVEL - 1)
 999                 parent = path->nodes[level + 1];
1000         pslot = path->slots[level + 1];
1001
1002         /*
1003          * deal with the case where there is only one pointer in the root
1004          * by promoting the node below to a root
1005          */
1006         if (!parent) {
1007                 struct extent_buffer *child;
1008
1009                 if (btrfs_header_nritems(mid) != 1)
1010                         return 0;
1011
1012                 /* promote the child to a root */
1013                 child = read_node_slot(root, mid, 0);
1014                 BUG_ON(!child);
1015                 btrfs_tree_lock(child);
1016                 btrfs_set_lock_blocking(child);
1017                 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1018                 BUG_ON(ret);
1019
1020                 spin_lock(&root->node_lock);
1021                 root->node = child;
1022                 spin_unlock(&root->node_lock);
1023
1024                 add_root_to_dirty_list(root);
1025                 btrfs_tree_unlock(child);
1026
1027                 path->locks[level] = 0;
1028                 path->nodes[level] = NULL;
1029                 clean_tree_block(trans, root, mid);
1030                 btrfs_tree_unlock(mid);
1031                 /* once for the path */
1032                 free_extent_buffer(mid);
1033                 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
1034                                         0, root->root_key.objectid, level, 1);
1035                 /* once for the root ptr */
1036                 free_extent_buffer(mid);
1037                 return ret;
1038         }
1039         if (btrfs_header_nritems(mid) >
1040             BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1041                 return 0;
1042
1043         if (btrfs_header_nritems(mid) < 2)
1044                 err_on_enospc = 1;
1045
1046         left = read_node_slot(root, parent, pslot - 1);
1047         if (left) {
1048                 btrfs_tree_lock(left);
1049                 btrfs_set_lock_blocking(left);
1050                 wret = btrfs_cow_block(trans, root, left,
1051                                        parent, pslot - 1, &left);
1052                 if (wret) {
1053                         ret = wret;
1054                         goto enospc;
1055                 }
1056         }
1057         right = read_node_slot(root, parent, pslot + 1);
1058         if (right) {
1059                 btrfs_tree_lock(right);
1060                 btrfs_set_lock_blocking(right);
1061                 wret = btrfs_cow_block(trans, root, right,
1062                                        parent, pslot + 1, &right);
1063                 if (wret) {
1064                         ret = wret;
1065                         goto enospc;
1066                 }
1067         }
1068
1069         /* first, try to make some room in the middle buffer */
1070         if (left) {
1071                 orig_slot += btrfs_header_nritems(left);
1072                 wret = push_node_left(trans, root, left, mid, 1);
1073                 if (wret < 0)
1074                         ret = wret;
1075                 if (btrfs_header_nritems(mid) < 2)
1076                         err_on_enospc = 1;
1077         }
1078
1079         /*
1080          * then try to empty the right most buffer into the middle
1081          */
1082         if (right) {
1083                 wret = push_node_left(trans, root, mid, right, 1);
1084                 if (wret < 0 && wret != -ENOSPC)
1085                         ret = wret;
1086                 if (btrfs_header_nritems(right) == 0) {
1087                         u64 bytenr = right->start;
1088                         u32 blocksize = right->len;
1089
1090                         clean_tree_block(trans, root, right);
1091                         btrfs_tree_unlock(right);
1092                         free_extent_buffer(right);
1093                         right = NULL;
1094                         wret = del_ptr(trans, root, path, level + 1, pslot +
1095                                        1);
1096                         if (wret)
1097                                 ret = wret;
1098                         wret = btrfs_free_extent(trans, root, bytenr,
1099                                                  blocksize, 0,
1100                                                  root->root_key.objectid,
1101                                                  level, 0);
1102                         if (wret)
1103                                 ret = wret;
1104                 } else {
1105                         struct btrfs_disk_key right_key;
1106                         btrfs_node_key(right, &right_key, 0);
1107                         btrfs_set_node_key(parent, &right_key, pslot + 1);
1108                         btrfs_mark_buffer_dirty(parent);
1109                 }
1110         }
1111         if (btrfs_header_nritems(mid) == 1) {
1112                 /*
1113                  * we're not allowed to leave a node with one item in the
1114                  * tree during a delete.  A deletion from lower in the tree
1115                  * could try to delete the only pointer in this node.
1116                  * So, pull some keys from the left.
1117                  * There has to be a left pointer at this point because
1118                  * otherwise we would have pulled some pointers from the
1119                  * right
1120                  */
1121                 BUG_ON(!left);
1122                 wret = balance_node_right(trans, root, mid, left);
1123                 if (wret < 0) {
1124                         ret = wret;
1125                         goto enospc;
1126                 }
1127                 if (wret == 1) {
1128                         wret = push_node_left(trans, root, left, mid, 1);
1129                         if (wret < 0)
1130                                 ret = wret;
1131                 }
1132                 BUG_ON(wret == 1);
1133         }
1134         if (btrfs_header_nritems(mid) == 0) {
1135                 /* we've managed to empty the middle node, drop it */
1136                 u64 bytenr = mid->start;
1137                 u32 blocksize = mid->len;
1138
1139                 clean_tree_block(trans, root, mid);
1140                 btrfs_tree_unlock(mid);
1141                 free_extent_buffer(mid);
1142                 mid = NULL;
1143                 wret = del_ptr(trans, root, path, level + 1, pslot);
1144                 if (wret)
1145                         ret = wret;
1146                 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
1147                                          0, root->root_key.objectid,
1148                                          level, 0);
1149                 if (wret)
1150                         ret = wret;
1151         } else {
1152                 /* update the parent key to reflect our changes */
1153                 struct btrfs_disk_key mid_key;
1154                 btrfs_node_key(mid, &mid_key, 0);
1155                 btrfs_set_node_key(parent, &mid_key, pslot);
1156                 btrfs_mark_buffer_dirty(parent);
1157         }
1158
1159         /* update the path */
1160         if (left) {
1161                 if (btrfs_header_nritems(left) > orig_slot) {
1162                         extent_buffer_get(left);
1163                         /* left was locked after cow */
1164                         path->nodes[level] = left;
1165                         path->slots[level + 1] -= 1;
1166                         path->slots[level] = orig_slot;
1167                         if (mid) {
1168                                 btrfs_tree_unlock(mid);
1169                                 free_extent_buffer(mid);
1170                         }
1171                 } else {
1172                         orig_slot -= btrfs_header_nritems(left);
1173                         path->slots[level] = orig_slot;
1174                 }
1175         }
1176         /* double check we haven't messed things up */
1177         check_block(root, path, level);
1178         if (orig_ptr !=
1179             btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1180                 BUG();
1181 enospc:
1182         if (right) {
1183                 btrfs_tree_unlock(right);
1184                 free_extent_buffer(right);
1185         }
1186         if (left) {
1187                 if (path->nodes[level] != left)
1188                         btrfs_tree_unlock(left);
1189                 free_extent_buffer(left);
1190         }
1191         return ret;
1192 }
1193
1194 /* Node balancing for insertion.  Here we only split or push nodes around
1195  * when they are completely full.  This is also done top down, so we
1196  * have to be pessimistic.
1197  */
1198 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1199                                           struct btrfs_root *root,
1200                                           struct btrfs_path *path, int level)
1201 {
1202         struct extent_buffer *right = NULL;
1203         struct extent_buffer *mid;
1204         struct extent_buffer *left = NULL;
1205         struct extent_buffer *parent = NULL;
1206         int ret = 0;
1207         int wret;
1208         int pslot;
1209         int orig_slot = path->slots[level];
1210         u64 orig_ptr;
1211
1212         if (level == 0)
1213                 return 1;
1214
1215         mid = path->nodes[level];
1216         WARN_ON(btrfs_header_generation(mid) != trans->transid);
1217         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1218
1219         if (level < BTRFS_MAX_LEVEL - 1)
1220                 parent = path->nodes[level + 1];
1221         pslot = path->slots[level + 1];
1222
1223         if (!parent)
1224                 return 1;
1225
1226         left = read_node_slot(root, parent, pslot - 1);
1227
1228         /* first, try to make some room in the middle buffer */
1229         if (left) {
1230                 u32 left_nr;
1231
1232                 btrfs_tree_lock(left);
1233                 btrfs_set_lock_blocking(left);
1234
1235                 left_nr = btrfs_header_nritems(left);
1236                 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1237                         wret = 1;
1238                 } else {
1239                         ret = btrfs_cow_block(trans, root, left, parent,
1240                                               pslot - 1, &left);
1241                         if (ret)
1242                                 wret = 1;
1243                         else {
1244                                 wret = push_node_left(trans, root,
1245                                                       left, mid, 0);
1246                         }
1247                 }
1248                 if (wret < 0)
1249                         ret = wret;
1250                 if (wret == 0) {
1251                         struct btrfs_disk_key disk_key;
1252                         orig_slot += left_nr;
1253                         btrfs_node_key(mid, &disk_key, 0);
1254                         btrfs_set_node_key(parent, &disk_key, pslot);
1255                         btrfs_mark_buffer_dirty(parent);
1256                         if (btrfs_header_nritems(left) > orig_slot) {
1257                                 path->nodes[level] = left;
1258                                 path->slots[level + 1] -= 1;
1259                                 path->slots[level] = orig_slot;
1260                                 btrfs_tree_unlock(mid);
1261                                 free_extent_buffer(mid);
1262                         } else {
1263                                 orig_slot -=
1264                                         btrfs_header_nritems(left);
1265                                 path->slots[level] = orig_slot;
1266                                 btrfs_tree_unlock(left);
1267                                 free_extent_buffer(left);
1268                         }
1269                         return 0;
1270                 }
1271                 btrfs_tree_unlock(left);
1272                 free_extent_buffer(left);
1273         }
1274         right = read_node_slot(root, parent, pslot + 1);
1275
1276         /*
1277          * then try to empty the right most buffer into the middle
1278          */
1279         if (right) {
1280                 u32 right_nr;
1281
1282                 btrfs_tree_lock(right);
1283                 btrfs_set_lock_blocking(right);
1284
1285                 right_nr = btrfs_header_nritems(right);
1286                 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1287                         wret = 1;
1288                 } else {
1289                         ret = btrfs_cow_block(trans, root, right,
1290                                               parent, pslot + 1,
1291                                               &right);
1292                         if (ret)
1293                                 wret = 1;
1294                         else {
1295                                 wret = balance_node_right(trans, root,
1296                                                           right, mid);
1297                         }
1298                 }
1299                 if (wret < 0)
1300                         ret = wret;
1301                 if (wret == 0) {
1302                         struct btrfs_disk_key disk_key;
1303
1304                         btrfs_node_key(right, &disk_key, 0);
1305                         btrfs_set_node_key(parent, &disk_key, pslot + 1);
1306                         btrfs_mark_buffer_dirty(parent);
1307
1308                         if (btrfs_header_nritems(mid) <= orig_slot) {
1309                                 path->nodes[level] = right;
1310                                 path->slots[level + 1] += 1;
1311                                 path->slots[level] = orig_slot -
1312                                         btrfs_header_nritems(mid);
1313                                 btrfs_tree_unlock(mid);
1314                                 free_extent_buffer(mid);
1315                         } else {
1316                                 btrfs_tree_unlock(right);
1317                                 free_extent_buffer(right);
1318                         }
1319                         return 0;
1320                 }
1321                 btrfs_tree_unlock(right);
1322                 free_extent_buffer(right);
1323         }
1324         return 1;
1325 }
1326
1327 /*
1328  * readahead one full node of leaves, finding things that are close
1329  * to the block in 'slot', and triggering ra on them.
1330  */
1331 static void reada_for_search(struct btrfs_root *root,
1332                              struct btrfs_path *path,
1333                              int level, int slot, u64 objectid)
1334 {
1335         struct extent_buffer *node;
1336         struct btrfs_disk_key disk_key;
1337         u32 nritems;
1338         u64 search;
1339         u64 target;
1340         u64 nread = 0;
1341         int direction = path->reada;
1342         struct extent_buffer *eb;
1343         u32 nr;
1344         u32 blocksize;
1345         u32 nscan = 0;
1346
1347         if (level != 1)
1348                 return;
1349
1350         if (!path->nodes[level])
1351                 return;
1352
1353         node = path->nodes[level];
1354
1355         search = btrfs_node_blockptr(node, slot);
1356         blocksize = btrfs_level_size(root, level - 1);
1357         eb = btrfs_find_tree_block(root, search, blocksize);
1358         if (eb) {
1359                 free_extent_buffer(eb);
1360                 return;
1361         }
1362
1363         target = search;
1364
1365         nritems = btrfs_header_nritems(node);
1366         nr = slot;
1367         while (1) {
1368                 if (direction < 0) {
1369                         if (nr == 0)
1370                                 break;
1371                         nr--;
1372                 } else if (direction > 0) {
1373                         nr++;
1374                         if (nr >= nritems)
1375                                 break;
1376                 }
1377                 if (path->reada < 0 && objectid) {
1378                         btrfs_node_key(node, &disk_key, nr);
1379                         if (btrfs_disk_key_objectid(&disk_key) != objectid)
1380                                 break;
1381                 }
1382                 search = btrfs_node_blockptr(node, nr);
1383                 if ((search <= target && target - search <= 65536) ||
1384                     (search > target && search - target <= 65536)) {
1385                         readahead_tree_block(root, search, blocksize,
1386                                      btrfs_node_ptr_generation(node, nr));
1387                         nread += blocksize;
1388                 }
1389                 nscan++;
1390                 if ((nread > 65536 || nscan > 32))
1391                         break;
1392         }
1393 }
1394
1395 /*
1396  * returns -EAGAIN if it had to drop the path, or zero if everything was in
1397  * cache
1398  */
1399 static noinline int reada_for_balance(struct btrfs_root *root,
1400                                       struct btrfs_path *path, int level)
1401 {
1402         int slot;
1403         int nritems;
1404         struct extent_buffer *parent;
1405         struct extent_buffer *eb;
1406         u64 gen;
1407         u64 block1 = 0;
1408         u64 block2 = 0;
1409         int ret = 0;
1410         int blocksize;
1411
1412         parent = path->nodes[level + 1];
1413         if (!parent)
1414                 return 0;
1415
1416         nritems = btrfs_header_nritems(parent);
1417         slot = path->slots[level + 1];
1418         blocksize = btrfs_level_size(root, level);
1419
1420         if (slot > 0) {
1421                 block1 = btrfs_node_blockptr(parent, slot - 1);
1422                 gen = btrfs_node_ptr_generation(parent, slot - 1);
1423                 eb = btrfs_find_tree_block(root, block1, blocksize);
1424                 if (eb && btrfs_buffer_uptodate(eb, gen))
1425                         block1 = 0;
1426                 free_extent_buffer(eb);
1427         }
1428         if (slot + 1 < nritems) {
1429                 block2 = btrfs_node_blockptr(parent, slot + 1);
1430                 gen = btrfs_node_ptr_generation(parent, slot + 1);
1431                 eb = btrfs_find_tree_block(root, block2, blocksize);
1432                 if (eb && btrfs_buffer_uptodate(eb, gen))
1433                         block2 = 0;
1434                 free_extent_buffer(eb);
1435         }
1436         if (block1 || block2) {
1437                 ret = -EAGAIN;
1438
1439                 /* release the whole path */
1440                 btrfs_release_path(root, path);
1441
1442                 /* read the blocks */
1443                 if (block1)
1444                         readahead_tree_block(root, block1, blocksize, 0);
1445                 if (block2)
1446                         readahead_tree_block(root, block2, blocksize, 0);
1447
1448                 if (block1) {
1449                         eb = read_tree_block(root, block1, blocksize, 0);
1450                         free_extent_buffer(eb);
1451                 }
1452                 if (block2) {
1453                         eb = read_tree_block(root, block2, blocksize, 0);
1454                         free_extent_buffer(eb);
1455                 }
1456         }
1457         return ret;
1458 }
1459
1460
1461 /*
1462  * when we walk down the tree, it is usually safe to unlock the higher layers
1463  * in the tree.  The exceptions are when our path goes through slot 0, because
1464  * operations on the tree might require changing key pointers higher up in the
1465  * tree.
1466  *
1467  * callers might also have set path->keep_locks, which tells this code to keep
1468  * the lock if the path points to the last slot in the block.  This is part of
1469  * walking through the tree, and selecting the next slot in the higher block.
1470  *
1471  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
1472  * if lowest_unlock is 1, level 0 won't be unlocked
1473  */
1474 static noinline void unlock_up(struct btrfs_path *path, int level,
1475                                int lowest_unlock)
1476 {
1477         int i;
1478         int skip_level = level;
1479         int no_skips = 0;
1480         struct extent_buffer *t;
1481
1482         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1483                 if (!path->nodes[i])
1484                         break;
1485                 if (!path->locks[i])
1486                         break;
1487                 if (!no_skips && path->slots[i] == 0) {
1488                         skip_level = i + 1;
1489                         continue;
1490                 }
1491                 if (!no_skips && path->keep_locks) {
1492                         u32 nritems;
1493                         t = path->nodes[i];
1494                         nritems = btrfs_header_nritems(t);
1495                         if (nritems < 1 || path->slots[i] >= nritems - 1) {
1496                                 skip_level = i + 1;
1497                                 continue;
1498                         }
1499                 }
1500                 if (skip_level < i && i >= lowest_unlock)
1501                         no_skips = 1;
1502
1503                 t = path->nodes[i];
1504                 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1505                         btrfs_tree_unlock(t);
1506                         path->locks[i] = 0;
1507                 }
1508         }
1509 }
1510
1511 /*
1512  * This releases any locks held in the path starting at level and
1513  * going all the way up to the root.
1514  *
1515  * btrfs_search_slot will keep the lock held on higher nodes in a few
1516  * corner cases, such as COW of the block at slot zero in the node.  This
1517  * ignores those rules, and it should only be called when there are no
1518  * more updates to be done higher up in the tree.
1519  */
1520 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1521 {
1522         int i;
1523
1524         if (path->keep_locks)
1525                 return;
1526
1527         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1528                 if (!path->nodes[i])
1529                         continue;
1530                 if (!path->locks[i])
1531                         continue;
1532                 btrfs_tree_unlock(path->nodes[i]);
1533                 path->locks[i] = 0;
1534         }
1535 }
1536
1537 /*
1538  * helper function for btrfs_search_slot.  The goal is to find a block
1539  * in cache without setting the path to blocking.  If we find the block
1540  * we return zero and the path is unchanged.
1541  *
1542  * If we can't find the block, we set the path blocking and do some
1543  * reada.  -EAGAIN is returned and the search must be repeated.
1544  */
1545 static int
1546 read_block_for_search(struct btrfs_trans_handle *trans,
1547                        struct btrfs_root *root, struct btrfs_path *p,
1548                        struct extent_buffer **eb_ret, int level, int slot,
1549                        struct btrfs_key *key)
1550 {
1551         u64 blocknr;
1552         u64 gen;
1553         u32 blocksize;
1554         struct extent_buffer *b = *eb_ret;
1555         struct extent_buffer *tmp;
1556         int ret;
1557
1558         blocknr = btrfs_node_blockptr(b, slot);
1559         gen = btrfs_node_ptr_generation(b, slot);
1560         blocksize = btrfs_level_size(root, level - 1);
1561
1562         tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1563         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1564                 /*
1565                  * we found an up to date block without sleeping, return
1566                  * right away
1567                  */
1568                 *eb_ret = tmp;
1569                 return 0;
1570         }
1571
1572         /*
1573          * reduce lock contention at high levels
1574          * of the btree by dropping locks before
1575          * we read.  Don't release the lock on the current
1576          * level because we need to walk this node to figure
1577          * out which blocks to read.
1578          */
1579         btrfs_unlock_up_safe(p, level + 1);
1580         btrfs_set_path_blocking(p);
1581
1582         if (tmp)
1583                 free_extent_buffer(tmp);
1584         if (p->reada)
1585                 reada_for_search(root, p, level, slot, key->objectid);
1586
1587         btrfs_release_path(NULL, p);
1588
1589         ret = -EAGAIN;
1590         tmp = read_tree_block(root, blocknr, blocksize, gen);
1591         if (tmp) {
1592                 /*
1593                  * If the read above didn't mark this buffer up to date,
1594                  * it will never end up being up to date.  Set ret to EIO now
1595                  * and give up so that our caller doesn't loop forever
1596                  * on our EAGAINs.
1597                  */
1598                 if (!btrfs_buffer_uptodate(tmp, 0))
1599                         ret = -EIO;
1600                 free_extent_buffer(tmp);
1601         }
1602         return ret;
1603 }
1604
1605 /*
1606  * helper function for btrfs_search_slot.  This does all of the checks
1607  * for node-level blocks and does any balancing required based on
1608  * the ins_len.
1609  *
1610  * If no extra work was required, zero is returned.  If we had to
1611  * drop the path, -EAGAIN is returned and btrfs_search_slot must
1612  * start over
1613  */
1614 static int
1615 setup_nodes_for_search(struct btrfs_trans_handle *trans,
1616                        struct btrfs_root *root, struct btrfs_path *p,
1617                        struct extent_buffer *b, int level, int ins_len)
1618 {
1619         int ret;
1620         if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1621             BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1622                 int sret;
1623
1624                 sret = reada_for_balance(root, p, level);
1625                 if (sret)
1626                         goto again;
1627
1628                 btrfs_set_path_blocking(p);
1629                 sret = split_node(trans, root, p, level);
1630                 btrfs_clear_path_blocking(p, NULL);
1631
1632                 BUG_ON(sret > 0);
1633                 if (sret) {
1634                         ret = sret;
1635                         goto done;
1636                 }
1637                 b = p->nodes[level];
1638         } else if (ins_len < 0 && btrfs_header_nritems(b) <
1639                    BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
1640                 int sret;
1641
1642                 sret = reada_for_balance(root, p, level);
1643                 if (sret)
1644                         goto again;
1645
1646                 btrfs_set_path_blocking(p);
1647                 sret = balance_level(trans, root, p, level);
1648                 btrfs_clear_path_blocking(p, NULL);
1649
1650                 if (sret) {
1651                         ret = sret;
1652                         goto done;
1653                 }
1654                 b = p->nodes[level];
1655                 if (!b) {
1656                         btrfs_release_path(NULL, p);
1657                         goto again;
1658                 }
1659                 BUG_ON(btrfs_header_nritems(b) == 1);
1660         }
1661         return 0;
1662
1663 again:
1664         ret = -EAGAIN;
1665 done:
1666         return ret;
1667 }
1668
1669 /*
1670  * look for key in the tree.  path is filled in with nodes along the way
1671  * if key is found, we return zero and you can find the item in the leaf
1672  * level of the path (level 0)
1673  *
1674  * If the key isn't found, the path points to the slot where it should
1675  * be inserted, and 1 is returned.  If there are other errors during the
1676  * search a negative error number is returned.
1677  *
1678  * if ins_len > 0, nodes and leaves will be split as we walk down the
1679  * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
1680  * possible)
1681  */
1682 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1683                       *root, struct btrfs_key *key, struct btrfs_path *p, int
1684                       ins_len, int cow)
1685 {
1686         struct extent_buffer *b;
1687         int slot;
1688         int ret;
1689         int err;
1690         int level;
1691         int lowest_unlock = 1;
1692         u8 lowest_level = 0;
1693
1694         lowest_level = p->lowest_level;
1695         WARN_ON(lowest_level && ins_len > 0);
1696         WARN_ON(p->nodes[0] != NULL);
1697
1698         if (ins_len < 0)
1699                 lowest_unlock = 2;
1700
1701 again:
1702         if (p->search_commit_root) {
1703                 b = root->commit_root;
1704                 extent_buffer_get(b);
1705                 if (!p->skip_locking)
1706                         btrfs_tree_lock(b);
1707         } else {
1708                 if (p->skip_locking)
1709                         b = btrfs_root_node(root);
1710                 else
1711                         b = btrfs_lock_root_node(root);
1712         }
1713
1714         while (b) {
1715                 level = btrfs_header_level(b);
1716
1717                 /*
1718                  * setup the path here so we can release it under lock
1719                  * contention with the cow code
1720                  */
1721                 p->nodes[level] = b;
1722                 if (!p->skip_locking)
1723                         p->locks[level] = 1;
1724
1725                 if (cow) {
1726                         /*
1727                          * if we don't really need to cow this block
1728                          * then we don't want to set the path blocking,
1729                          * so we test it here
1730                          */
1731                         if (!should_cow_block(trans, root, b))
1732                                 goto cow_done;
1733
1734                         btrfs_set_path_blocking(p);
1735
1736                         err = btrfs_cow_block(trans, root, b,
1737                                               p->nodes[level + 1],
1738                                               p->slots[level + 1], &b);
1739                         if (err) {
1740                                 free_extent_buffer(b);
1741                                 ret = err;
1742                                 goto done;
1743                         }
1744                 }
1745 cow_done:
1746                 BUG_ON(!cow && ins_len);
1747                 if (level != btrfs_header_level(b))
1748                         WARN_ON(1);
1749                 level = btrfs_header_level(b);
1750
1751                 p->nodes[level] = b;
1752                 if (!p->skip_locking)
1753                         p->locks[level] = 1;
1754
1755                 btrfs_clear_path_blocking(p, NULL);
1756
1757                 /*
1758                  * we have a lock on b and as long as we aren't changing
1759                  * the tree, there is no way to for the items in b to change.
1760                  * It is safe to drop the lock on our parent before we
1761                  * go through the expensive btree search on b.
1762                  *
1763                  * If cow is true, then we might be changing slot zero,
1764                  * which may require changing the parent.  So, we can't
1765                  * drop the lock until after we know which slot we're
1766                  * operating on.
1767                  */
1768                 if (!cow)
1769                         btrfs_unlock_up_safe(p, level + 1);
1770
1771                 ret = check_block(root, p, level);
1772                 if (ret) {
1773                         ret = -1;
1774                         goto done;
1775                 }
1776
1777                 ret = bin_search(b, key, level, &slot);
1778
1779                 if (level != 0) {
1780                         int dec = 0;
1781                         if (ret && slot > 0) {
1782                                 dec = 1;
1783                                 slot -= 1;
1784                         }
1785                         p->slots[level] = slot;
1786                         err = setup_nodes_for_search(trans, root, p, b, level,
1787                                                      ins_len);
1788                         if (err == -EAGAIN)
1789                                 goto again;
1790                         if (err) {
1791                                 ret = err;
1792                                 goto done;
1793                         }
1794                         b = p->nodes[level];
1795                         slot = p->slots[level];
1796
1797                         unlock_up(p, level, lowest_unlock);
1798
1799                         if (level == lowest_level) {
1800                                 if (dec)
1801                                         p->slots[level]++;
1802                                 goto done;
1803                         }
1804
1805                         err = read_block_for_search(trans, root, p,
1806                                                     &b, level, slot, key);
1807                         if (err == -EAGAIN)
1808                                 goto again;
1809                         if (err) {
1810                                 ret = err;
1811                                 goto done;
1812                         }
1813
1814                         if (!p->skip_locking) {
1815                                 btrfs_clear_path_blocking(p, NULL);
1816                                 err = btrfs_try_spin_lock(b);
1817
1818                                 if (!err) {
1819                                         btrfs_set_path_blocking(p);
1820                                         btrfs_tree_lock(b);
1821                                         btrfs_clear_path_blocking(p, b);
1822                                 }
1823                         }
1824                 } else {
1825                         p->slots[level] = slot;
1826                         if (ins_len > 0 &&
1827                             btrfs_leaf_free_space(root, b) < ins_len) {
1828                                 btrfs_set_path_blocking(p);
1829                                 err = split_leaf(trans, root, key,
1830                                                  p, ins_len, ret == 0);
1831                                 btrfs_clear_path_blocking(p, NULL);
1832
1833                                 BUG_ON(err > 0);
1834                                 if (err) {
1835                                         ret = err;
1836                                         goto done;
1837                                 }
1838                         }
1839                         if (!p->search_for_split)
1840                                 unlock_up(p, level, lowest_unlock);
1841                         goto done;
1842                 }
1843         }
1844         ret = 1;
1845 done:
1846         /*
1847          * we don't really know what they plan on doing with the path
1848          * from here on, so for now just mark it as blocking
1849          */
1850         if (!p->leave_spinning)
1851                 btrfs_set_path_blocking(p);
1852         if (ret < 0)
1853                 btrfs_release_path(root, p);
1854         return ret;
1855 }
1856
1857 /*
1858  * adjust the pointers going up the tree, starting at level
1859  * making sure the right key of each node is points to 'key'.
1860  * This is used after shifting pointers to the left, so it stops
1861  * fixing up pointers when a given leaf/node is not in slot 0 of the
1862  * higher levels
1863  *
1864  * If this fails to write a tree block, it returns -1, but continues
1865  * fixing up the blocks in ram so the tree is consistent.
1866  */
1867 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1868                           struct btrfs_root *root, struct btrfs_path *path,
1869                           struct btrfs_disk_key *key, int level)
1870 {
1871         int i;
1872         int ret = 0;
1873         struct extent_buffer *t;
1874
1875         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1876                 int tslot = path->slots[i];
1877                 if (!path->nodes[i])
1878                         break;
1879                 t = path->nodes[i];
1880                 btrfs_set_node_key(t, key, tslot);
1881                 btrfs_mark_buffer_dirty(path->nodes[i]);
1882                 if (tslot != 0)
1883                         break;
1884         }
1885         return ret;
1886 }
1887
1888 /*
1889  * update item key.
1890  *
1891  * This function isn't completely safe. It's the caller's responsibility
1892  * that the new key won't break the order
1893  */
1894 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1895                             struct btrfs_root *root, struct btrfs_path *path,
1896                             struct btrfs_key *new_key)
1897 {
1898         struct btrfs_disk_key disk_key;
1899         struct extent_buffer *eb;
1900         int slot;
1901
1902         eb = path->nodes[0];
1903         slot = path->slots[0];
1904         if (slot > 0) {
1905                 btrfs_item_key(eb, &disk_key, slot - 1);
1906                 if (comp_keys(&disk_key, new_key) >= 0)
1907                         return -1;
1908         }
1909         if (slot < btrfs_header_nritems(eb) - 1) {
1910                 btrfs_item_key(eb, &disk_key, slot + 1);
1911                 if (comp_keys(&disk_key, new_key) <= 0)
1912                         return -1;
1913         }
1914
1915         btrfs_cpu_key_to_disk(&disk_key, new_key);
1916         btrfs_set_item_key(eb, &disk_key, slot);
1917         btrfs_mark_buffer_dirty(eb);
1918         if (slot == 0)
1919                 fixup_low_keys(trans, root, path, &disk_key, 1);
1920         return 0;
1921 }
1922
1923 /*
1924  * try to push data from one node into the next node left in the
1925  * tree.
1926  *
1927  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1928  * error, and > 0 if there was no room in the left hand block.
1929  */
1930 static int push_node_left(struct btrfs_trans_handle *trans,
1931                           struct btrfs_root *root, struct extent_buffer *dst,
1932                           struct extent_buffer *src, int empty)
1933 {
1934         int push_items = 0;
1935         int src_nritems;
1936         int dst_nritems;
1937         int ret = 0;
1938
1939         src_nritems = btrfs_header_nritems(src);
1940         dst_nritems = btrfs_header_nritems(dst);
1941         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1942         WARN_ON(btrfs_header_generation(src) != trans->transid);
1943         WARN_ON(btrfs_header_generation(dst) != trans->transid);
1944
1945         if (!empty && src_nritems <= 8)
1946                 return 1;
1947
1948         if (push_items <= 0)
1949                 return 1;
1950
1951         if (empty) {
1952                 push_items = min(src_nritems, push_items);
1953                 if (push_items < src_nritems) {
1954                         /* leave at least 8 pointers in the node if
1955                          * we aren't going to empty it
1956                          */
1957                         if (src_nritems - push_items < 8) {
1958                                 if (push_items <= 8)
1959                                         return 1;
1960                                 push_items -= 8;
1961                         }
1962                 }
1963         } else
1964                 push_items = min(src_nritems - 8, push_items);
1965
1966         copy_extent_buffer(dst, src,
1967                            btrfs_node_key_ptr_offset(dst_nritems),
1968                            btrfs_node_key_ptr_offset(0),
1969                            push_items * sizeof(struct btrfs_key_ptr));
1970
1971         if (push_items < src_nritems) {
1972                 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1973                                       btrfs_node_key_ptr_offset(push_items),
1974                                       (src_nritems - push_items) *
1975                                       sizeof(struct btrfs_key_ptr));
1976         }
1977         btrfs_set_header_nritems(src, src_nritems - push_items);
1978         btrfs_set_header_nritems(dst, dst_nritems + push_items);
1979         btrfs_mark_buffer_dirty(src);
1980         btrfs_mark_buffer_dirty(dst);
1981
1982         return ret;
1983 }
1984
1985 /*
1986  * try to push data from one node into the next node right in the
1987  * tree.
1988  *
1989  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1990  * error, and > 0 if there was no room in the right hand block.
1991  *
1992  * this will  only push up to 1/2 the contents of the left node over
1993  */
1994 static int balance_node_right(struct btrfs_trans_handle *trans,
1995                               struct btrfs_root *root,
1996                               struct extent_buffer *dst,
1997                               struct extent_buffer *src)
1998 {
1999         int push_items = 0;
2000         int max_push;
2001         int src_nritems;
2002         int dst_nritems;
2003         int ret = 0;
2004
2005         WARN_ON(btrfs_header_generation(src) != trans->transid);
2006         WARN_ON(btrfs_header_generation(dst) != trans->transid);
2007
2008         src_nritems = btrfs_header_nritems(src);
2009         dst_nritems = btrfs_header_nritems(dst);
2010         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2011         if (push_items <= 0)
2012                 return 1;
2013
2014         if (src_nritems < 4)
2015                 return 1;
2016
2017         max_push = src_nritems / 2 + 1;
2018         /* don't try to empty the node */
2019         if (max_push >= src_nritems)
2020                 return 1;
2021
2022         if (max_push < push_items)
2023                 push_items = max_push;
2024
2025         memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2026                                       btrfs_node_key_ptr_offset(0),
2027                                       (dst_nritems) *
2028                                       sizeof(struct btrfs_key_ptr));
2029
2030         copy_extent_buffer(dst, src,
2031                            btrfs_node_key_ptr_offset(0),
2032                            btrfs_node_key_ptr_offset(src_nritems - push_items),
2033                            push_items * sizeof(struct btrfs_key_ptr));
2034
2035         btrfs_set_header_nritems(src, src_nritems - push_items);
2036         btrfs_set_header_nritems(dst, dst_nritems + push_items);
2037
2038         btrfs_mark_buffer_dirty(src);
2039         btrfs_mark_buffer_dirty(dst);
2040
2041         return ret;
2042 }
2043
2044 /*
2045  * helper function to insert a new root level in the tree.
2046  * A new node is allocated, and a single item is inserted to
2047  * point to the existing root
2048  *
2049  * returns zero on success or < 0 on failure.
2050  */
2051 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2052                            struct btrfs_root *root,
2053                            struct btrfs_path *path, int level)
2054 {
2055         u64 lower_gen;
2056         struct extent_buffer *lower;
2057         struct extent_buffer *c;
2058         struct extent_buffer *old;
2059         struct btrfs_disk_key lower_key;
2060
2061         BUG_ON(path->nodes[level]);
2062         BUG_ON(path->nodes[level-1] != root->node);
2063
2064         lower = path->nodes[level-1];
2065         if (level == 1)
2066                 btrfs_item_key(lower, &lower_key, 0);
2067         else
2068                 btrfs_node_key(lower, &lower_key, 0);
2069
2070         c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2071                                    root->root_key.objectid, &lower_key,
2072                                    level, root->node->start, 0);
2073         if (IS_ERR(c))
2074                 return PTR_ERR(c);
2075
2076         memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
2077         btrfs_set_header_nritems(c, 1);
2078         btrfs_set_header_level(c, level);
2079         btrfs_set_header_bytenr(c, c->start);
2080         btrfs_set_header_generation(c, trans->transid);
2081         btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
2082         btrfs_set_header_owner(c, root->root_key.objectid);
2083
2084         write_extent_buffer(c, root->fs_info->fsid,
2085                             (unsigned long)btrfs_header_fsid(c),
2086                             BTRFS_FSID_SIZE);
2087
2088         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2089                             (unsigned long)btrfs_header_chunk_tree_uuid(c),
2090                             BTRFS_UUID_SIZE);
2091
2092         btrfs_set_node_key(c, &lower_key, 0);
2093         btrfs_set_node_blockptr(c, 0, lower->start);
2094         lower_gen = btrfs_header_generation(lower);
2095         WARN_ON(lower_gen != trans->transid);
2096
2097         btrfs_set_node_ptr_generation(c, 0, lower_gen);
2098
2099         btrfs_mark_buffer_dirty(c);
2100
2101         spin_lock(&root->node_lock);
2102         old = root->node;
2103         root->node = c;
2104         spin_unlock(&root->node_lock);
2105
2106         /* the super has an extra ref to root->node */
2107         free_extent_buffer(old);
2108
2109         add_root_to_dirty_list(root);
2110         extent_buffer_get(c);
2111         path->nodes[level] = c;
2112         path->locks[level] = 1;
2113         path->slots[level] = 0;
2114         return 0;
2115 }
2116
2117 /*
2118  * worker function to insert a single pointer in a node.
2119  * the node should have enough room for the pointer already
2120  *
2121  * slot and level indicate where you want the key to go, and
2122  * blocknr is the block the key points to.
2123  *
2124  * returns zero on success and < 0 on any error
2125  */
2126 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2127                       *root, struct btrfs_path *path, struct btrfs_disk_key
2128                       *key, u64 bytenr, int slot, int level)
2129 {
2130         struct extent_buffer *lower;
2131         int nritems;
2132
2133         BUG_ON(!path->nodes[level]);
2134         lower = path->nodes[level];
2135         nritems = btrfs_header_nritems(lower);
2136         BUG_ON(slot > nritems);
2137         if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2138                 BUG();
2139         if (slot != nritems) {
2140                 memmove_extent_buffer(lower,
2141                               btrfs_node_key_ptr_offset(slot + 1),
2142                               btrfs_node_key_ptr_offset(slot),
2143                               (nritems - slot) * sizeof(struct btrfs_key_ptr));
2144         }
2145         btrfs_set_node_key(lower, key, slot);
2146         btrfs_set_node_blockptr(lower, slot, bytenr);
2147         WARN_ON(trans->transid == 0);
2148         btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2149         btrfs_set_header_nritems(lower, nritems + 1);
2150         btrfs_mark_buffer_dirty(lower);
2151         return 0;
2152 }
2153
2154 /*
2155  * split the node at the specified level in path in two.
2156  * The path is corrected to point to the appropriate node after the split
2157  *
2158  * Before splitting this tries to make some room in the node by pushing
2159  * left and right, if either one works, it returns right away.
2160  *
2161  * returns 0 on success and < 0 on failure
2162  */
2163 static noinline int split_node(struct btrfs_trans_handle *trans,
2164                                struct btrfs_root *root,
2165                                struct btrfs_path *path, int level)
2166 {
2167         struct extent_buffer *c;
2168         struct extent_buffer *split;
2169         struct btrfs_disk_key disk_key;
2170         int mid;
2171         int ret;
2172         int wret;
2173         u32 c_nritems;
2174
2175         c = path->nodes[level];
2176         WARN_ON(btrfs_header_generation(c) != trans->transid);
2177         if (c == root->node) {
2178                 /* trying to split the root, lets make a new one */
2179                 ret = insert_new_root(trans, root, path, level + 1);
2180                 if (ret)
2181                         return ret;
2182         } else {
2183                 ret = push_nodes_for_insert(trans, root, path, level);
2184                 c = path->nodes[level];
2185                 if (!ret && btrfs_header_nritems(c) <
2186                     BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2187                         return 0;
2188                 if (ret < 0)
2189                         return ret;
2190         }
2191
2192         c_nritems = btrfs_header_nritems(c);
2193         mid = (c_nritems + 1) / 2;
2194         btrfs_node_key(c, &disk_key, mid);
2195
2196         split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2197                                         root->root_key.objectid,
2198                                         &disk_key, level, c->start, 0);
2199         if (IS_ERR(split))
2200                 return PTR_ERR(split);
2201
2202         memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
2203         btrfs_set_header_level(split, btrfs_header_level(c));
2204         btrfs_set_header_bytenr(split, split->start);
2205         btrfs_set_header_generation(split, trans->transid);
2206         btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
2207         btrfs_set_header_owner(split, root->root_key.objectid);
2208         write_extent_buffer(split, root->fs_info->fsid,
2209                             (unsigned long)btrfs_header_fsid(split),
2210                             BTRFS_FSID_SIZE);
2211         write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2212                             (unsigned long)btrfs_header_chunk_tree_uuid(split),
2213                             BTRFS_UUID_SIZE);
2214
2215
2216         copy_extent_buffer(split, c,
2217                            btrfs_node_key_ptr_offset(0),
2218                            btrfs_node_key_ptr_offset(mid),
2219                            (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2220         btrfs_set_header_nritems(split, c_nritems - mid);
2221         btrfs_set_header_nritems(c, mid);
2222         ret = 0;
2223
2224         btrfs_mark_buffer_dirty(c);
2225         btrfs_mark_buffer_dirty(split);
2226
2227         wret = insert_ptr(trans, root, path, &disk_key, split->start,
2228                           path->slots[level + 1] + 1,
2229                           level + 1);
2230         if (wret)
2231                 ret = wret;
2232
2233         if (path->slots[level] >= mid) {
2234                 path->slots[level] -= mid;
2235                 btrfs_tree_unlock(c);
2236                 free_extent_buffer(c);
2237                 path->nodes[level] = split;
2238                 path->slots[level + 1] += 1;
2239         } else {
2240                 btrfs_tree_unlock(split);
2241                 free_extent_buffer(split);
2242         }
2243         return ret;
2244 }
2245
2246 /*
2247  * how many bytes are required to store the items in a leaf.  start
2248  * and nr indicate which items in the leaf to check.  This totals up the
2249  * space used both by the item structs and the item data
2250  */
2251 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2252 {
2253         int data_len;
2254         int nritems = btrfs_header_nritems(l);
2255         int end = min(nritems, start + nr) - 1;
2256
2257         if (!nr)
2258                 return 0;
2259         data_len = btrfs_item_end_nr(l, start);
2260         data_len = data_len - btrfs_item_offset_nr(l, end);
2261         data_len += sizeof(struct btrfs_item) * nr;
2262         WARN_ON(data_len < 0);
2263         return data_len;
2264 }
2265
2266 /*
2267  * The space between the end of the leaf items and
2268  * the start of the leaf data.  IOW, how much room
2269  * the leaf has left for both items and data
2270  */
2271 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2272                                    struct extent_buffer *leaf)
2273 {
2274         int nritems = btrfs_header_nritems(leaf);
2275         int ret;
2276         ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2277         if (ret < 0) {
2278                 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2279                        "used %d nritems %d\n",
2280                        ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2281                        leaf_space_used(leaf, 0, nritems), nritems);
2282         }
2283         return ret;
2284 }
2285
2286 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2287                                       struct btrfs_root *root,
2288                                       struct btrfs_path *path,
2289                                       int data_size, int empty,
2290                                       struct extent_buffer *right,
2291                                       int free_space, u32 left_nritems)
2292 {
2293         struct extent_buffer *left = path->nodes[0];
2294         struct extent_buffer *upper = path->nodes[1];
2295         struct btrfs_disk_key disk_key;
2296         int slot;
2297         u32 i;
2298         int push_space = 0;
2299         int push_items = 0;
2300         struct btrfs_item *item;
2301         u32 nr;
2302         u32 right_nritems;
2303         u32 data_end;
2304         u32 this_item_size;
2305
2306         if (empty)
2307                 nr = 0;
2308         else
2309                 nr = 1;
2310
2311         if (path->slots[0] >= left_nritems)
2312                 push_space += data_size;
2313
2314         slot = path->slots[1];
2315         i = left_nritems - 1;
2316         while (i >= nr) {
2317                 item = btrfs_item_nr(left, i);
2318
2319                 if (!empty && push_items > 0) {
2320                         if (path->slots[0] > i)
2321                                 break;
2322                         if (path->slots[0] == i) {
2323                                 int space = btrfs_leaf_free_space(root, left);
2324                                 if (space + push_space * 2 > free_space)
2325                                         break;
2326                         }
2327                 }
2328
2329                 if (path->slots[0] == i)
2330                         push_space += data_size;
2331
2332                 if (!left->map_token) {
2333                         map_extent_buffer(left, (unsigned long)item,
2334                                         sizeof(struct btrfs_item),
2335                                         &left->map_token, &left->kaddr,
2336                                         &left->map_start, &left->map_len,
2337                                         KM_USER1);
2338                 }
2339
2340                 this_item_size = btrfs_item_size(left, item);
2341                 if (this_item_size + sizeof(*item) + push_space > free_space)
2342                         break;
2343
2344                 push_items++;
2345                 push_space += this_item_size + sizeof(*item);
2346                 if (i == 0)
2347                         break;
2348                 i--;
2349         }
2350         if (left->map_token) {
2351                 unmap_extent_buffer(left, left->map_token, KM_USER1);
2352                 left->map_token = NULL;
2353         }
2354
2355         if (push_items == 0)
2356                 goto out_unlock;
2357
2358         if (!empty && push_items == left_nritems)
2359                 WARN_ON(1);
2360
2361         /* push left to right */
2362         right_nritems = btrfs_header_nritems(right);
2363
2364         push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2365         push_space -= leaf_data_end(root, left);
2366
2367         /* make room in the right data area */
2368         data_end = leaf_data_end(root, right);
2369         memmove_extent_buffer(right,
2370                               btrfs_leaf_data(right) + data_end - push_space,
2371                               btrfs_leaf_data(right) + data_end,
2372                               BTRFS_LEAF_DATA_SIZE(root) - data_end);
2373
2374         /* copy from the left data area */
2375         copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2376                      BTRFS_LEAF_DATA_SIZE(root) - push_space,
2377                      btrfs_leaf_data(left) + leaf_data_end(root, left),
2378                      push_space);
2379
2380         memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2381                               btrfs_item_nr_offset(0),
2382                               right_nritems * sizeof(struct btrfs_item));
2383
2384         /* copy the items from left to right */
2385         copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2386                    btrfs_item_nr_offset(left_nritems - push_items),
2387                    push_items * sizeof(struct btrfs_item));
2388
2389         /* update the item pointers */
2390         right_nritems += push_items;
2391         btrfs_set_header_nritems(right, right_nritems);
2392         push_space = BTRFS_LEAF_DATA_SIZE(root);
2393         for (i = 0; i < right_nritems; i++) {
2394                 item = btrfs_item_nr(right, i);
2395                 if (!right->map_token) {
2396                         map_extent_buffer(right, (unsigned long)item,
2397                                         sizeof(struct btrfs_item),
2398                                         &right->map_token, &right->kaddr,
2399                                         &right->map_start, &right->map_len,
2400                                         KM_USER1);
2401                 }
2402                 push_space -= btrfs_item_size(right, item);
2403                 btrfs_set_item_offset(right, item, push_space);
2404         }
2405
2406         if (right->map_token) {
2407                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2408                 right->map_token = NULL;
2409         }
2410         left_nritems -= push_items;
2411         btrfs_set_header_nritems(left, left_nritems);
2412
2413         if (left_nritems)
2414                 btrfs_mark_buffer_dirty(left);
2415         btrfs_mark_buffer_dirty(right);
2416
2417         btrfs_item_key(right, &disk_key, 0);
2418         btrfs_set_node_key(upper, &disk_key, slot + 1);
2419         btrfs_mark_buffer_dirty(upper);
2420
2421         /* then fixup the leaf pointer in the path */
2422         if (path->slots[0] >= left_nritems) {
2423                 path->slots[0] -= left_nritems;
2424                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2425                         clean_tree_block(trans, root, path->nodes[0]);
2426                 btrfs_tree_unlock(path->nodes[0]);
2427                 free_extent_buffer(path->nodes[0]);
2428                 path->nodes[0] = right;
2429                 path->slots[1] += 1;
2430         } else {
2431                 btrfs_tree_unlock(right);
2432                 free_extent_buffer(right);
2433         }
2434         return 0;
2435
2436 out_unlock:
2437         btrfs_tree_unlock(right);
2438         free_extent_buffer(right);
2439         return 1;
2440 }
2441
2442 /*
2443  * push some data in the path leaf to the right, trying to free up at
2444  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2445  *
2446  * returns 1 if the push failed because the other node didn't have enough
2447  * room, 0 if everything worked out and < 0 if there were major errors.
2448  */
2449 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2450                            *root, struct btrfs_path *path, int data_size,
2451                            int empty)
2452 {
2453         struct extent_buffer *left = path->nodes[0];
2454         struct extent_buffer *right;
2455         struct extent_buffer *upper;
2456         int slot;
2457         int free_space;
2458         u32 left_nritems;
2459         int ret;
2460
2461         if (!path->nodes[1])
2462                 return 1;
2463
2464         slot = path->slots[1];
2465         upper = path->nodes[1];
2466         if (slot >= btrfs_header_nritems(upper) - 1)
2467                 return 1;
2468
2469         btrfs_assert_tree_locked(path->nodes[1]);
2470
2471         right = read_node_slot(root, upper, slot + 1);
2472         btrfs_tree_lock(right);
2473         btrfs_set_lock_blocking(right);
2474
2475         free_space = btrfs_leaf_free_space(root, right);
2476         if (free_space < data_size)
2477                 goto out_unlock;
2478
2479         /* cow and double check */
2480         ret = btrfs_cow_block(trans, root, right, upper,
2481                               slot + 1, &right);
2482         if (ret)
2483                 goto out_unlock;
2484
2485         free_space = btrfs_leaf_free_space(root, right);
2486         if (free_space < data_size)
2487                 goto out_unlock;
2488
2489         left_nritems = btrfs_header_nritems(left);
2490         if (left_nritems == 0)
2491                 goto out_unlock;
2492
2493         return __push_leaf_right(trans, root, path, data_size, empty,
2494                                 right, free_space, left_nritems);
2495 out_unlock:
2496         btrfs_tree_unlock(right);
2497         free_extent_buffer(right);
2498         return 1;
2499 }
2500
2501 /*
2502  * push some data in the path leaf to the left, trying to free up at
2503  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2504  */
2505 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2506                                      struct btrfs_root *root,
2507                                      struct btrfs_path *path, int data_size,
2508                                      int empty, struct extent_buffer *left,
2509                                      int free_space, int right_nritems)
2510 {
2511         struct btrfs_disk_key disk_key;
2512         struct extent_buffer *right = path->nodes[0];
2513         int slot;
2514         int i;
2515         int push_space = 0;
2516         int push_items = 0;
2517         struct btrfs_item *item;
2518         u32 old_left_nritems;
2519         u32 nr;
2520         int ret = 0;
2521         int wret;
2522         u32 this_item_size;
2523         u32 old_left_item_size;
2524
2525         slot = path->slots[1];
2526
2527         if (empty)
2528                 nr = right_nritems;
2529         else
2530                 nr = right_nritems - 1;
2531
2532         for (i = 0; i < nr; i++) {
2533                 item = btrfs_item_nr(right, i);
2534                 if (!right->map_token) {
2535                         map_extent_buffer(right, (unsigned long)item,
2536                                         sizeof(struct btrfs_item),
2537                                         &right->map_token, &right->kaddr,
2538                                         &right->map_start, &right->map_len,
2539                                         KM_USER1);
2540                 }
2541
2542                 if (!empty && push_items > 0) {
2543                         if (path->slots[0] < i)
2544                                 break;
2545                         if (path->slots[0] == i) {
2546                                 int space = btrfs_leaf_free_space(root, right);
2547                                 if (space + push_space * 2 > free_space)
2548                                         break;
2549                         }
2550                 }
2551
2552                 if (path->slots[0] == i)
2553                         push_space += data_size;
2554
2555                 this_item_size = btrfs_item_size(right, item);
2556                 if (this_item_size + sizeof(*item) + push_space > free_space)
2557                         break;
2558
2559                 push_items++;
2560                 push_space += this_item_size + sizeof(*item);
2561         }
2562
2563         if (right->map_token) {
2564                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2565                 right->map_token = NULL;
2566         }
2567
2568         if (push_items == 0) {
2569                 ret = 1;
2570                 goto out;
2571         }
2572         if (!empty && push_items == btrfs_header_nritems(right))
2573                 WARN_ON(1);
2574
2575         /* push data from right to left */
2576         copy_extent_buffer(left, right,
2577                            btrfs_item_nr_offset(btrfs_header_nritems(left)),
2578                            btrfs_item_nr_offset(0),
2579                            push_items * sizeof(struct btrfs_item));
2580
2581         push_space = BTRFS_LEAF_DATA_SIZE(root) -
2582                      btrfs_item_offset_nr(right, push_items - 1);
2583
2584         copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2585                      leaf_data_end(root, left) - push_space,
2586                      btrfs_leaf_data(right) +
2587                      btrfs_item_offset_nr(right, push_items - 1),
2588                      push_space);
2589         old_left_nritems = btrfs_header_nritems(left);
2590         BUG_ON(old_left_nritems <= 0);
2591
2592         old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2593         for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2594                 u32 ioff;
2595
2596                 item = btrfs_item_nr(left, i);
2597                 if (!left->map_token) {
2598                         map_extent_buffer(left, (unsigned long)item,
2599                                         sizeof(struct btrfs_item),
2600                                         &left->map_token, &left->kaddr,
2601                                         &left->map_start, &left->map_len,
2602                                         KM_USER1);
2603                 }
2604
2605                 ioff = btrfs_item_offset(left, item);
2606                 btrfs_set_item_offset(left, item,
2607                       ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2608         }
2609         btrfs_set_header_nritems(left, old_left_nritems + push_items);
2610         if (left->map_token) {
2611                 unmap_extent_buffer(left, left->map_token, KM_USER1);
2612                 left->map_token = NULL;
2613         }
2614
2615         /* fixup right node */
2616         if (push_items > right_nritems) {
2617                 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2618                        right_nritems);
2619                 WARN_ON(1);
2620         }
2621
2622         if (push_items < right_nritems) {
2623                 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2624                                                   leaf_data_end(root, right);
2625                 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2626                                       BTRFS_LEAF_DATA_SIZE(root) - push_space,
2627                                       btrfs_leaf_data(right) +
2628                                       leaf_data_end(root, right), push_space);
2629
2630                 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2631                               btrfs_item_nr_offset(push_items),
2632                              (btrfs_header_nritems(right) - push_items) *
2633                              sizeof(struct btrfs_item));
2634         }
2635         right_nritems -= push_items;
2636         btrfs_set_header_nritems(right, right_nritems);
2637         push_space = BTRFS_LEAF_DATA_SIZE(root);
2638         for (i = 0; i < right_nritems; i++) {
2639                 item = btrfs_item_nr(right, i);
2640
2641                 if (!right->map_token) {
2642                         map_extent_buffer(right, (unsigned long)item,
2643                                         sizeof(struct btrfs_item),
2644                                         &right->map_token, &right->kaddr,
2645                                         &right->map_start, &right->map_len,
2646                                         KM_USER1);
2647                 }
2648
2649                 push_space = push_space - btrfs_item_size(right, item);
2650                 btrfs_set_item_offset(right, item, push_space);
2651         }
2652         if (right->map_token) {
2653                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2654                 right->map_token = NULL;
2655         }
2656
2657         btrfs_mark_buffer_dirty(left);
2658         if (right_nritems)
2659                 btrfs_mark_buffer_dirty(right);
2660
2661         btrfs_item_key(right, &disk_key, 0);
2662         wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2663         if (wret)
2664                 ret = wret;
2665
2666         /* then fixup the leaf pointer in the path */
2667         if (path->slots[0] < push_items) {
2668                 path->slots[0] += old_left_nritems;
2669                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2670                         clean_tree_block(trans, root, path->nodes[0]);
2671                 btrfs_tree_unlock(path->nodes[0]);
2672                 free_extent_buffer(path->nodes[0]);
2673                 path->nodes[0] = left;
2674                 path->slots[1] -= 1;
2675         } else {
2676                 btrfs_tree_unlock(left);
2677                 free_extent_buffer(left);
2678                 path->slots[0] -= push_items;
2679         }
2680         BUG_ON(path->slots[0] < 0);
2681         return ret;
2682 out:
2683         btrfs_tree_unlock(left);
2684         free_extent_buffer(left);
2685         return ret;
2686 }
2687
2688 /*
2689  * push some data in the path leaf to the left, trying to free up at
2690  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2691  */
2692 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2693                           *root, struct btrfs_path *path, int data_size,
2694                           int empty)
2695 {
2696         struct extent_buffer *right = path->nodes[0];
2697         struct extent_buffer *left;
2698         int slot;
2699         int free_space;
2700         u32 right_nritems;
2701         int ret = 0;
2702
2703         slot = path->slots[1];
2704         if (slot == 0)
2705                 return 1;
2706         if (!path->nodes[1])
2707                 return 1;
2708
2709         right_nritems = btrfs_header_nritems(right);
2710         if (right_nritems == 0)
2711                 return 1;
2712
2713         btrfs_assert_tree_locked(path->nodes[1]);
2714
2715         left = read_node_slot(root, path->nodes[1], slot - 1);
2716         btrfs_tree_lock(left);
2717         btrfs_set_lock_blocking(left);
2718
2719         free_space = btrfs_leaf_free_space(root, left);
2720         if (free_space < data_size) {
2721                 ret = 1;
2722                 goto out;
2723         }
2724
2725         /* cow and double check */
2726         ret = btrfs_cow_block(trans, root, left,
2727                               path->nodes[1], slot - 1, &left);
2728         if (ret) {
2729                 /* we hit -ENOSPC, but it isn't fatal here */
2730                 ret = 1;
2731                 goto out;
2732         }
2733
2734         free_space = btrfs_leaf_free_space(root, left);
2735         if (free_space < data_size) {
2736                 ret = 1;
2737                 goto out;
2738         }
2739
2740         return __push_leaf_left(trans, root, path, data_size,
2741                                empty, left, free_space, right_nritems);
2742 out:
2743         btrfs_tree_unlock(left);
2744         free_extent_buffer(left);
2745         return ret;
2746 }
2747
2748 /*
2749  * split the path's leaf in two, making sure there is at least data_size
2750  * available for the resulting leaf level of the path.
2751  *
2752  * returns 0 if all went well and < 0 on failure.
2753  */
2754 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2755                                struct btrfs_root *root,
2756                                struct btrfs_path *path,
2757                                struct extent_buffer *l,
2758                                struct extent_buffer *right,
2759                                int slot, int mid, int nritems)
2760 {
2761         int data_copy_size;
2762         int rt_data_off;
2763         int i;
2764         int ret = 0;
2765         int wret;
2766         struct btrfs_disk_key disk_key;
2767
2768         nritems = nritems - mid;
2769         btrfs_set_header_nritems(right, nritems);
2770         data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2771
2772         copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2773                            btrfs_item_nr_offset(mid),
2774                            nritems * sizeof(struct btrfs_item));
2775
2776         copy_extent_buffer(right, l,
2777                      btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2778                      data_copy_size, btrfs_leaf_data(l) +
2779                      leaf_data_end(root, l), data_copy_size);
2780
2781         rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2782                       btrfs_item_end_nr(l, mid);
2783
2784         for (i = 0; i < nritems; i++) {
2785                 struct btrfs_item *item = btrfs_item_nr(right, i);
2786                 u32 ioff;
2787
2788                 if (!right->map_token) {
2789                         map_extent_buffer(right, (unsigned long)item,
2790                                         sizeof(struct btrfs_item),
2791                                         &right->map_token, &right->kaddr,
2792                                         &right->map_start, &right->map_len,
2793                                         KM_USER1);
2794                 }
2795
2796                 ioff = btrfs_item_offset(right, item);
2797                 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2798         }
2799
2800         if (right->map_token) {
2801                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2802                 right->map_token = NULL;
2803         }
2804
2805         btrfs_set_header_nritems(l, mid);
2806         ret = 0;
2807         btrfs_item_key(right, &disk_key, 0);
2808         wret = insert_ptr(trans, root, path, &disk_key, right->start,
2809                           path->slots[1] + 1, 1);
2810         if (wret)
2811                 ret = wret;
2812
2813         btrfs_mark_buffer_dirty(right);
2814         btrfs_mark_buffer_dirty(l);
2815         BUG_ON(path->slots[0] != slot);
2816
2817         if (mid <= slot) {
2818                 btrfs_tree_unlock(path->nodes[0]);
2819                 free_extent_buffer(path->nodes[0]);
2820                 path->nodes[0] = right;
2821                 path->slots[0] -= mid;
2822                 path->slots[1] += 1;
2823         } else {
2824                 btrfs_tree_unlock(right);
2825                 free_extent_buffer(right);
2826         }
2827
2828         BUG_ON(path->slots[0] < 0);
2829
2830         return ret;
2831 }
2832
2833 /*
2834  * split the path's leaf in two, making sure there is at least data_size
2835  * available for the resulting leaf level of the path.
2836  *
2837  * returns 0 if all went well and < 0 on failure.
2838  */
2839 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2840                                struct btrfs_root *root,
2841                                struct btrfs_key *ins_key,
2842                                struct btrfs_path *path, int data_size,
2843                                int extend)
2844 {
2845         struct btrfs_disk_key disk_key;
2846         struct extent_buffer *l;
2847         u32 nritems;
2848         int mid;
2849         int slot;
2850         struct extent_buffer *right;
2851         int ret = 0;
2852         int wret;
2853         int split;
2854         int num_doubles = 0;
2855
2856         /* first try to make some room by pushing left and right */
2857         if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2858                 wret = push_leaf_right(trans, root, path, data_size, 0);
2859                 if (wret < 0)
2860                         return wret;
2861                 if (wret) {
2862                         wret = push_leaf_left(trans, root, path, data_size, 0);
2863                         if (wret < 0)
2864                                 return wret;
2865                 }
2866                 l = path->nodes[0];
2867
2868                 /* did the pushes work? */
2869                 if (btrfs_leaf_free_space(root, l) >= data_size)
2870                         return 0;
2871         }
2872
2873         if (!path->nodes[1]) {
2874                 ret = insert_new_root(trans, root, path, 1);
2875                 if (ret)
2876                         return ret;
2877         }
2878 again:
2879         split = 1;
2880         l = path->nodes[0];
2881         slot = path->slots[0];
2882         nritems = btrfs_header_nritems(l);
2883         mid = (nritems + 1) / 2;
2884
2885         if (mid <= slot) {
2886                 if (nritems == 1 ||
2887                     leaf_space_used(l, mid, nritems - mid) + data_size >
2888                         BTRFS_LEAF_DATA_SIZE(root)) {
2889                         if (slot >= nritems) {
2890                                 split = 0;
2891                         } else {
2892                                 mid = slot;
2893                                 if (mid != nritems &&
2894                                     leaf_space_used(l, mid, nritems - mid) +
2895                                     data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2896                                         split = 2;
2897                                 }
2898                         }
2899                 }
2900         } else {
2901                 if (leaf_space_used(l, 0, mid) + data_size >
2902                         BTRFS_LEAF_DATA_SIZE(root)) {
2903                         if (!extend && data_size && slot == 0) {
2904                                 split = 0;
2905                         } else if ((extend || !data_size) && slot == 0) {
2906                                 mid = 1;
2907                         } else {
2908                                 mid = slot;
2909                                 if (mid != nritems &&
2910                                     leaf_space_used(l, mid, nritems - mid) +
2911                                     data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2912                                         split = 2 ;
2913                                 }
2914                         }
2915                 }
2916         }
2917
2918         if (split == 0)
2919                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2920         else
2921                 btrfs_item_key(l, &disk_key, mid);
2922
2923         right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
2924                                         root->root_key.objectid,
2925                                         &disk_key, 0, l->start, 0);
2926         if (IS_ERR(right)) {
2927                 BUG_ON(1);
2928                 return PTR_ERR(right);
2929         }
2930
2931         memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2932         btrfs_set_header_bytenr(right, right->start);
2933         btrfs_set_header_generation(right, trans->transid);
2934         btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2935         btrfs_set_header_owner(right, root->root_key.objectid);
2936         btrfs_set_header_level(right, 0);
2937         write_extent_buffer(right, root->fs_info->fsid,
2938                             (unsigned long)btrfs_header_fsid(right),
2939                             BTRFS_FSID_SIZE);
2940
2941         write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2942                             (unsigned long)btrfs_header_chunk_tree_uuid(right),
2943                             BTRFS_UUID_SIZE);
2944
2945         if (split == 0) {
2946                 if (mid <= slot) {
2947                         btrfs_set_header_nritems(right, 0);
2948                         wret = insert_ptr(trans, root, path,
2949                                           &disk_key, right->start,
2950                                           path->slots[1] + 1, 1);
2951                         if (wret)
2952                                 ret = wret;
2953
2954                         btrfs_tree_unlock(path->nodes[0]);
2955                         free_extent_buffer(path->nodes[0]);
2956                         path->nodes[0] = right;
2957                         path->slots[0] = 0;
2958                         path->slots[1] += 1;
2959                 } else {
2960                         btrfs_set_header_nritems(right, 0);
2961                         wret = insert_ptr(trans, root, path,
2962                                           &disk_key,
2963                                           right->start,
2964                                           path->slots[1], 1);
2965                         if (wret)
2966                                 ret = wret;
2967                         btrfs_tree_unlock(path->nodes[0]);
2968                         free_extent_buffer(path->nodes[0]);
2969                         path->nodes[0] = right;
2970                         path->slots[0] = 0;
2971                         if (path->slots[1] == 0) {
2972                                 wret = fixup_low_keys(trans, root,
2973                                                 path, &disk_key, 1);
2974                                 if (wret)
2975                                         ret = wret;
2976                         }
2977                 }
2978                 btrfs_mark_buffer_dirty(right);
2979                 return ret;
2980         }
2981
2982         ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2983         BUG_ON(ret);
2984
2985         if (split == 2) {
2986                 BUG_ON(num_doubles != 0);
2987                 num_doubles++;
2988                 goto again;
2989         }
2990
2991         return ret;
2992 }
2993
2994 /*
2995  * This function splits a single item into two items,
2996  * giving 'new_key' to the new item and splitting the
2997  * old one at split_offset (from the start of the item).
2998  *
2999  * The path may be released by this operation.  After
3000  * the split, the path is pointing to the old item.  The
3001  * new item is going to be in the same node as the old one.
3002  *
3003  * Note, the item being split must be smaller enough to live alone on
3004  * a tree block with room for one extra struct btrfs_item
3005  *
3006  * This allows us to split the item in place, keeping a lock on the
3007  * leaf the entire time.
3008  */
3009 int btrfs_split_item(struct btrfs_trans_handle *trans,
3010                      struct btrfs_root *root,
3011                      struct btrfs_path *path,
3012                      struct btrfs_key *new_key,
3013                      unsigned long split_offset)
3014 {
3015         u32 item_size;
3016         struct extent_buffer *leaf;
3017         struct btrfs_key orig_key;
3018         struct btrfs_item *item;
3019         struct btrfs_item *new_item;
3020         int ret = 0;
3021         int slot;
3022         u32 nritems;
3023         u32 orig_offset;
3024         struct btrfs_disk_key disk_key;
3025         char *buf;
3026
3027         leaf = path->nodes[0];
3028         btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
3029         if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
3030                 goto split;
3031
3032         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3033         btrfs_release_path(root, path);
3034
3035         path->search_for_split = 1;
3036         path->keep_locks = 1;
3037
3038         ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
3039         path->search_for_split = 0;
3040
3041         /* if our item isn't there or got smaller, return now */
3042         if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
3043                                                         path->slots[0])) {
3044                 path->keep_locks = 0;
3045                 return -EAGAIN;
3046         }
3047
3048         btrfs_set_path_blocking(path);
3049         ret = split_leaf(trans, root, &orig_key, path,
3050                          sizeof(struct btrfs_item), 1);
3051         path->keep_locks = 0;
3052         BUG_ON(ret);
3053
3054         btrfs_unlock_up_safe(path, 1);
3055         leaf = path->nodes[0];
3056         BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3057
3058 split:
3059         /*
3060          * make sure any changes to the path from split_leaf leave it
3061          * in a blocking state
3062          */
3063         btrfs_set_path_blocking(path);
3064
3065         item = btrfs_item_nr(leaf, path->slots[0]);
3066         orig_offset = btrfs_item_offset(leaf, item);
3067         item_size = btrfs_item_size(leaf, item);
3068
3069         buf = kmalloc(item_size, GFP_NOFS);
3070         read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3071                             path->slots[0]), item_size);
3072         slot = path->slots[0] + 1;
3073         leaf = path->nodes[0];
3074
3075         nritems = btrfs_header_nritems(leaf);
3076
3077         if (slot != nritems) {
3078                 /* shift the items */
3079                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3080                               btrfs_item_nr_offset(slot),
3081                               (nritems - slot) * sizeof(struct btrfs_item));
3082
3083         }
3084
3085         btrfs_cpu_key_to_disk(&disk_key, new_key);
3086         btrfs_set_item_key(leaf, &disk_key, slot);
3087
3088         new_item = btrfs_item_nr(leaf, slot);
3089
3090         btrfs_set_item_offset(leaf, new_item, orig_offset);
3091         btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3092
3093         btrfs_set_item_offset(leaf, item,
3094                               orig_offset + item_size - split_offset);
3095         btrfs_set_item_size(leaf, item, split_offset);
3096
3097         btrfs_set_header_nritems(leaf, nritems + 1);
3098
3099         /* write the data for the start of the original item */
3100         write_extent_buffer(leaf, buf,
3101                             btrfs_item_ptr_offset(leaf, path->slots[0]),
3102                             split_offset);
3103
3104         /* write the data for the new item */
3105         write_extent_buffer(leaf, buf + split_offset,
3106                             btrfs_item_ptr_offset(leaf, slot),
3107                             item_size - split_offset);
3108         btrfs_mark_buffer_dirty(leaf);
3109
3110         ret = 0;
3111         if (btrfs_leaf_free_space(root, leaf) < 0) {
3112                 btrfs_print_leaf(root, leaf);
3113                 BUG();
3114         }
3115         kfree(buf);
3116         return ret;
3117 }
3118
3119 /*
3120  * make the item pointed to by the path smaller.  new_size indicates
3121  * how small to make it, and from_end tells us if we just chop bytes
3122  * off the end of the item or if we shift the item to chop bytes off
3123  * the front.
3124  */
3125 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3126                         struct btrfs_root *root,
3127                         struct btrfs_path *path,
3128                         u32 new_size, int from_end)
3129 {
3130         int ret = 0;
3131         int slot;
3132         int slot_orig;
3133         struct extent_buffer *leaf;
3134         struct btrfs_item *item;
3135         u32 nritems;
3136         unsigned int data_end;
3137         unsigned int old_data_start;
3138         unsigned int old_size;
3139         unsigned int size_diff;
3140         int i;
3141
3142         slot_orig = path->slots[0];
3143         leaf = path->nodes[0];
3144         slot = path->slots[0];
3145
3146         old_size = btrfs_item_size_nr(leaf, slot);
3147         if (old_size == new_size)
3148                 return 0;
3149
3150         nritems = btrfs_header_nritems(leaf);
3151         data_end = leaf_data_end(root, leaf);
3152
3153         old_data_start = btrfs_item_offset_nr(leaf, slot);
3154
3155         size_diff = old_size - new_size;
3156
3157         BUG_ON(slot < 0);
3158         BUG_ON(slot >= nritems);
3159
3160         /*
3161          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3162          */
3163         /* first correct the data pointers */
3164         for (i = slot; i < nritems; i++) {
3165                 u32 ioff;
3166                 item = btrfs_item_nr(leaf, i);
3167
3168                 if (!leaf->map_token) {
3169                         map_extent_buffer(leaf, (unsigned long)item,
3170                                         sizeof(struct btrfs_item),
3171                                         &leaf->map_token, &leaf->kaddr,
3172                                         &leaf->map_start, &leaf->map_len,
3173                                         KM_USER1);
3174                 }
3175
3176                 ioff = btrfs_item_offset(leaf, item);
3177                 btrfs_set_item_offset(leaf, item, ioff + size_diff);
3178         }
3179
3180         if (leaf->map_token) {
3181                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3182                 leaf->map_token = NULL;
3183         }
3184
3185         /* shift the data */
3186         if (from_end) {
3187                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3188                               data_end + size_diff, btrfs_leaf_data(leaf) +
3189                               data_end, old_data_start + new_size - data_end);
3190         } else {
3191                 struct btrfs_disk_key disk_key;
3192                 u64 offset;
3193
3194                 btrfs_item_key(leaf, &disk_key, slot);
3195
3196                 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3197                         unsigned long ptr;
3198                         struct btrfs_file_extent_item *fi;
3199
3200                         fi = btrfs_item_ptr(leaf, slot,
3201                                             struct btrfs_file_extent_item);
3202                         fi = (struct btrfs_file_extent_item *)(
3203                              (unsigned long)fi - size_diff);
3204
3205                         if (btrfs_file_extent_type(leaf, fi) ==
3206                             BTRFS_FILE_EXTENT_INLINE) {
3207                                 ptr = btrfs_item_ptr_offset(leaf, slot);
3208                                 memmove_extent_buffer(leaf, ptr,
3209                                       (unsigned long)fi,
3210                                       offsetof(struct btrfs_file_extent_item,
3211                                                  disk_bytenr));
3212                         }
3213                 }
3214
3215                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3216                               data_end + size_diff, btrfs_leaf_data(leaf) +
3217                               data_end, old_data_start - data_end);
3218
3219                 offset = btrfs_disk_key_offset(&disk_key);
3220                 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3221                 btrfs_set_item_key(leaf, &disk_key, slot);
3222                 if (slot == 0)
3223                         fixup_low_keys(trans, root, path, &disk_key, 1);
3224         }
3225
3226         item = btrfs_item_nr(leaf, slot);
3227         btrfs_set_item_size(leaf, item, new_size);
3228         btrfs_mark_buffer_dirty(leaf);
3229
3230         ret = 0;
3231         if (btrfs_leaf_free_space(root, leaf) < 0) {
3232                 btrfs_print_leaf(root, leaf);
3233                 BUG();
3234         }
3235         return ret;
3236 }
3237
3238 /*
3239  * make the item pointed to by the path bigger, data_size is the new size.
3240  */
3241 int btrfs_extend_item(struct btrfs_trans_handle *trans,
3242                       struct btrfs_root *root, struct btrfs_path *path,
3243                       u32 data_size)
3244 {
3245         int ret = 0;
3246         int slot;
3247         int slot_orig;
3248         struct extent_buffer *leaf;
3249         struct btrfs_item *item;
3250         u32 nritems;
3251         unsigned int data_end;
3252         unsigned int old_data;
3253         unsigned int old_size;
3254         int i;
3255
3256         slot_orig = path->slots[0];
3257         leaf = path->nodes[0];
3258
3259         nritems = btrfs_header_nritems(leaf);
3260         data_end = leaf_data_end(root, leaf);
3261
3262         if (btrfs_leaf_free_space(root, leaf) < data_size) {
3263                 btrfs_print_leaf(root, leaf);
3264                 BUG();
3265         }
3266         slot = path->slots[0];
3267         old_data = btrfs_item_end_nr(leaf, slot);
3268
3269         BUG_ON(slot < 0);
3270         if (slot >= nritems) {
3271                 btrfs_print_leaf(root, leaf);
3272                 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3273                        slot, nritems);
3274                 BUG_ON(1);
3275         }
3276
3277         /*
3278          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3279          */
3280         /* first correct the data pointers */
3281         for (i = slot; i < nritems; i++) {
3282                 u32 ioff;
3283                 item = btrfs_item_nr(leaf, i);
3284
3285                 if (!leaf->map_token) {
3286                         map_extent_buffer(leaf, (unsigned long)item,
3287                                         sizeof(struct btrfs_item),
3288                                         &leaf->map_token, &leaf->kaddr,
3289                                         &leaf->map_start, &leaf->map_len,
3290                                         KM_USER1);
3291                 }
3292                 ioff = btrfs_item_offset(leaf, item);
3293                 btrfs_set_item_offset(leaf, item, ioff - data_size);
3294         }
3295
3296         if (leaf->map_token) {
3297                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3298                 leaf->map_token = NULL;
3299         }
3300
3301         /* shift the data */
3302         memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3303                       data_end - data_size, btrfs_leaf_data(leaf) +
3304                       data_end, old_data - data_end);
3305
3306         data_end = old_data;
3307         old_size = btrfs_item_size_nr(leaf, slot);
3308         item = btrfs_item_nr(leaf, slot);
3309         btrfs_set_item_size(leaf, item, old_size + data_size);
3310         btrfs_mark_buffer_dirty(leaf);
3311
3312         ret = 0;
3313         if (btrfs_leaf_free_space(root, leaf) < 0) {
3314                 btrfs_print_leaf(root, leaf);
3315                 BUG();
3316         }
3317         return ret;
3318 }
3319
3320 /*
3321  * Given a key and some data, insert items into the tree.
3322  * This does all the path init required, making room in the tree if needed.
3323  * Returns the number of keys that were inserted.
3324  */
3325 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3326                             struct btrfs_root *root,
3327                             struct btrfs_path *path,
3328                             struct btrfs_key *cpu_key, u32 *data_size,
3329                             int nr)
3330 {
3331         struct extent_buffer *leaf;
3332         struct btrfs_item *item;
3333         int ret = 0;
3334         int slot;
3335         int i;
3336         u32 nritems;
3337         u32 total_data = 0;
3338         u32 total_size = 0;
3339         unsigned int data_end;
3340         struct btrfs_disk_key disk_key;
3341         struct btrfs_key found_key;
3342
3343         for (i = 0; i < nr; i++) {
3344                 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3345                     BTRFS_LEAF_DATA_SIZE(root)) {
3346                         break;
3347                         nr = i;
3348                 }
3349                 total_data += data_size[i];
3350                 total_size += data_size[i] + sizeof(struct btrfs_item);
3351         }
3352         BUG_ON(nr == 0);
3353
3354         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3355         if (ret == 0)
3356                 return -EEXIST;
3357         if (ret < 0)
3358                 goto out;
3359
3360         leaf = path->nodes[0];
3361
3362         nritems = btrfs_header_nritems(leaf);
3363         data_end = leaf_data_end(root, leaf);
3364
3365         if (btrfs_leaf_free_space(root, leaf) < total_size) {
3366                 for (i = nr; i >= 0; i--) {
3367                         total_data -= data_size[i];
3368                         total_size -= data_size[i] + sizeof(struct btrfs_item);
3369                         if (total_size < btrfs_leaf_free_space(root, leaf))
3370                                 break;
3371                 }
3372                 nr = i;
3373         }
3374
3375         slot = path->slots[0];
3376         BUG_ON(slot < 0);
3377
3378         if (slot != nritems) {
3379                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3380
3381                 item = btrfs_item_nr(leaf, slot);
3382                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3383
3384                 /* figure out how many keys we can insert in here */
3385                 total_data = data_size[0];
3386                 for (i = 1; i < nr; i++) {
3387                         if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3388                                 break;
3389                         total_data += data_size[i];
3390                 }
3391                 nr = i;
3392
3393                 if (old_data < data_end) {
3394                         btrfs_print_leaf(root, leaf);
3395                         printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3396                                slot, old_data, data_end);
3397                         BUG_ON(1);
3398                 }
3399                 /*
3400                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3401                  */
3402                 /* first correct the data pointers */
3403                 WARN_ON(leaf->map_token);
3404                 for (i = slot; i < nritems; i++) {
3405                         u32 ioff;
3406
3407                         item = btrfs_item_nr(leaf, i);
3408                         if (!leaf->map_token) {
3409                                 map_extent_buffer(leaf, (unsigned long)item,
3410                                         sizeof(struct btrfs_item),
3411                                         &leaf->map_token, &leaf->kaddr,
3412                                         &leaf->map_start, &leaf->map_len,
3413                                         KM_USER1);
3414                         }
3415
3416                         ioff = btrfs_item_offset(leaf, item);
3417                         btrfs_set_item_offset(leaf, item, ioff - total_data);
3418                 }
3419                 if (leaf->map_token) {
3420                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3421                         leaf->map_token = NULL;
3422                 }
3423
3424                 /* shift the items */
3425                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3426                               btrfs_item_nr_offset(slot),
3427                               (nritems - slot) * sizeof(struct btrfs_item));
3428
3429                 /* shift the data */
3430                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3431                               data_end - total_data, btrfs_leaf_data(leaf) +
3432                               data_end, old_data - data_end);
3433                 data_end = old_data;
3434         } else {
3435                 /*
3436                  * this sucks but it has to be done, if we are inserting at
3437                  * the end of the leaf only insert 1 of the items, since we
3438                  * have no way of knowing whats on the next leaf and we'd have
3439                  * to drop our current locks to figure it out
3440                  */
3441                 nr = 1;
3442         }
3443
3444         /* setup the item for the new data */
3445         for (i = 0; i < nr; i++) {
3446                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3447                 btrfs_set_item_key(leaf, &disk_key, slot + i);
3448                 item = btrfs_item_nr(leaf, slot + i);
3449                 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3450                 data_end -= data_size[i];
3451                 btrfs_set_item_size(leaf, item, data_size[i]);
3452         }
3453         btrfs_set_header_nritems(leaf, nritems + nr);
3454         btrfs_mark_buffer_dirty(leaf);
3455
3456         ret = 0;
3457         if (slot == 0) {
3458                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3459                 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3460         }
3461
3462         if (btrfs_leaf_free_space(root, leaf) < 0) {
3463                 btrfs_print_leaf(root, leaf);
3464                 BUG();
3465         }
3466 out:
3467         if (!ret)
3468                 ret = nr;
3469         return ret;
3470 }
3471
3472 /*
3473  * this is a helper for btrfs_insert_empty_items, the main goal here is
3474  * to save stack depth by doing the bulk of the work in a function
3475  * that doesn't call btrfs_search_slot
3476  */
3477 static noinline_for_stack int
3478 setup_items_for_insert(struct btrfs_trans_handle *trans,
3479                       struct btrfs_root *root, struct btrfs_path *path,
3480                       struct btrfs_key *cpu_key, u32 *data_size,
3481                       u32 total_data, u32 total_size, int nr)
3482 {
3483         struct btrfs_item *item;
3484         int i;
3485         u32 nritems;
3486         unsigned int data_end;
3487         struct btrfs_disk_key disk_key;
3488         int ret;
3489         struct extent_buffer *leaf;
3490         int slot;
3491
3492         leaf = path->nodes[0];
3493         slot = path->slots[0];
3494
3495         nritems = btrfs_header_nritems(leaf);
3496         data_end = leaf_data_end(root, leaf);
3497
3498         if (btrfs_leaf_free_space(root, leaf) < total_size) {
3499                 btrfs_print_leaf(root, leaf);
3500                 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3501                        total_size, btrfs_leaf_free_space(root, leaf));
3502                 BUG();
3503         }
3504
3505         if (slot != nritems) {
3506                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3507
3508                 if (old_data < data_end) {
3509                         btrfs_print_leaf(root, leaf);
3510                         printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3511                                slot, old_data, data_end);
3512                         BUG_ON(1);
3513                 }
3514                 /*
3515                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3516                  */
3517                 /* first correct the data pointers */
3518                 WARN_ON(leaf->map_token);
3519                 for (i = slot; i < nritems; i++) {
3520                         u32 ioff;
3521
3522                         item = btrfs_item_nr(leaf, i);
3523                         if (!leaf->map_token) {
3524                                 map_extent_buffer(leaf, (unsigned long)item,
3525                                         sizeof(struct btrfs_item),
3526                                         &leaf->map_token, &leaf->kaddr,
3527                                         &leaf->map_start, &leaf->map_len,
3528                                         KM_USER1);
3529                         }
3530
3531                         ioff = btrfs_item_offset(leaf, item);
3532                         btrfs_set_item_offset(leaf, item, ioff - total_data);
3533                 }
3534                 if (leaf->map_token) {
3535                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3536                         leaf->map_token = NULL;
3537                 }
3538
3539                 /* shift the items */
3540                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3541                               btrfs_item_nr_offset(slot),
3542                               (nritems - slot) * sizeof(struct btrfs_item));
3543
3544                 /* shift the data */
3545                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3546                               data_end - total_data, btrfs_leaf_data(leaf) +
3547                               data_end, old_data - data_end);
3548                 data_end = old_data;
3549         }
3550
3551         /* setup the item for the new data */
3552         for (i = 0; i < nr; i++) {
3553                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3554                 btrfs_set_item_key(leaf, &disk_key, slot + i);
3555                 item = btrfs_item_nr(leaf, slot + i);
3556                 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3557                 data_end -= data_size[i];
3558                 btrfs_set_item_size(leaf, item, data_size[i]);
3559         }
3560
3561         btrfs_set_header_nritems(leaf, nritems + nr);
3562
3563         ret = 0;
3564         if (slot == 0) {
3565                 struct btrfs_disk_key disk_key;
3566                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3567                 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3568         }
3569         btrfs_unlock_up_safe(path, 1);
3570         btrfs_mark_buffer_dirty(leaf);
3571
3572         if (btrfs_leaf_free_space(root, leaf) < 0) {
3573                 btrfs_print_leaf(root, leaf);
3574                 BUG();
3575         }
3576         return ret;
3577 }
3578
3579 /*
3580  * Given a key and some data, insert items into the tree.
3581  * This does all the path init required, making room in the tree if needed.
3582  */
3583 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3584                             struct btrfs_root *root,
3585                             struct btrfs_path *path,
3586                             struct btrfs_key *cpu_key, u32 *data_size,
3587                             int nr)
3588 {
3589         struct extent_buffer *leaf;
3590         int ret = 0;
3591         int slot;
3592         int i;
3593         u32 total_size = 0;
3594         u32 total_data = 0;
3595
3596         for (i = 0; i < nr; i++)
3597                 total_data += data_size[i];
3598
3599         total_size = total_data + (nr * sizeof(struct btrfs_item));
3600         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3601         if (ret == 0)
3602                 return -EEXIST;
3603         if (ret < 0)
3604                 goto out;
3605
3606         leaf = path->nodes[0];
3607         slot = path->slots[0];
3608         BUG_ON(slot < 0);
3609
3610         ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3611                                total_data, total_size, nr);
3612
3613 out:
3614         return ret;
3615 }
3616
3617 /*
3618  * Given a key and some data, insert an item into the tree.
3619  * This does all the path init required, making room in the tree if needed.
3620  */
3621 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3622                       *root, struct btrfs_key *cpu_key, void *data, u32
3623                       data_size)
3624 {
3625         int ret = 0;
3626         struct btrfs_path *path;
3627         struct extent_buffer *leaf;
3628         unsigned long ptr;
3629
3630         path = btrfs_alloc_path();
3631         BUG_ON(!path);
3632         ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3633         if (!ret) {
3634                 leaf = path->nodes[0];
3635                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3636                 write_extent_buffer(leaf, data, ptr, data_size);
3637                 btrfs_mark_buffer_dirty(leaf);
3638         }
3639         btrfs_free_path(path);
3640         return ret;
3641 }
3642
3643 /*
3644  * delete the pointer from a given node.
3645  *
3646  * the tree should have been previously balanced so the deletion does not
3647  * empty a node.
3648  */
3649 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3650                    struct btrfs_path *path, int level, int slot)
3651 {
3652         struct extent_buffer *parent = path->nodes[level];
3653         u32 nritems;
3654         int ret = 0;
3655         int wret;
3656
3657         nritems = btrfs_header_nritems(parent);
3658         if (slot != nritems - 1) {
3659                 memmove_extent_buffer(parent,
3660                               btrfs_node_key_ptr_offset(slot),
3661                               btrfs_node_key_ptr_offset(slot + 1),
3662                               sizeof(struct btrfs_key_ptr) *
3663                               (nritems - slot - 1));
3664         }
3665         nritems--;
3666         btrfs_set_header_nritems(parent, nritems);
3667         if (nritems == 0 && parent == root->node) {
3668                 BUG_ON(btrfs_header_level(root->node) != 1);
3669                 /* just turn the root into a leaf and break */
3670                 btrfs_set_header_level(root->node, 0);
3671         } else if (slot == 0) {
3672                 struct btrfs_disk_key disk_key;
3673
3674                 btrfs_node_key(parent, &disk_key, 0);
3675                 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3676                 if (wret)
3677                         ret = wret;
3678         }
3679         btrfs_mark_buffer_dirty(parent);
3680         return ret;
3681 }
3682
3683 /*
3684  * a helper function to delete the leaf pointed to by path->slots[1] and
3685  * path->nodes[1].
3686  *
3687  * This deletes the pointer in path->nodes[1] and frees the leaf
3688  * block extent.  zero is returned if it all worked out, < 0 otherwise.
3689  *
3690  * The path must have already been setup for deleting the leaf, including
3691  * all the proper balancing.  path->nodes[1] must be locked.
3692  */
3693 static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3694                                    struct btrfs_root *root,
3695                                    struct btrfs_path *path,
3696                                    struct extent_buffer *leaf)
3697 {
3698         int ret;
3699
3700         WARN_ON(btrfs_header_generation(leaf) != trans->transid);
3701         ret = del_ptr(trans, root, path, 1, path->slots[1]);
3702         if (ret)
3703                 return ret;
3704
3705         /*
3706          * btrfs_free_extent is expensive, we want to make sure we
3707          * aren't holding any locks when we call it
3708          */
3709         btrfs_unlock_up_safe(path, 0);
3710
3711         ret = btrfs_free_extent(trans, root, leaf->start, leaf->len,
3712                                 0, root->root_key.objectid, 0, 0);
3713         return ret;
3714 }
3715 /*
3716  * delete the item at the leaf level in path.  If that empties
3717  * the leaf, remove it from the tree
3718  */
3719 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3720                     struct btrfs_path *path, int slot, int nr)
3721 {
3722         struct extent_buffer *leaf;
3723         struct btrfs_item *item;
3724         int last_off;
3725         int dsize = 0;
3726         int ret = 0;
3727         int wret;
3728         int i;
3729         u32 nritems;
3730
3731         leaf = path->nodes[0];
3732         last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3733
3734         for (i = 0; i < nr; i++)
3735                 dsize += btrfs_item_size_nr(leaf, slot + i);
3736
3737         nritems = btrfs_header_nritems(leaf);
3738
3739         if (slot + nr != nritems) {
3740                 int data_end = leaf_data_end(root, leaf);
3741
3742                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3743                               data_end + dsize,
3744                               btrfs_leaf_data(leaf) + data_end,
3745                               last_off - data_end);
3746
3747                 for (i = slot + nr; i < nritems; i++) {
3748                         u32 ioff;
3749
3750                         item = btrfs_item_nr(leaf, i);
3751                         if (!leaf->map_token) {
3752                                 map_extent_buffer(leaf, (unsigned long)item,
3753                                         sizeof(struct btrfs_item),
3754                                         &leaf->map_token, &leaf->kaddr,
3755                                         &leaf->map_start, &leaf->map_len,
3756                                         KM_USER1);
3757                         }
3758                         ioff = btrfs_item_offset(leaf, item);
3759                         btrfs_set_item_offset(leaf, item, ioff + dsize);
3760                 }
3761
3762                 if (leaf->map_token) {
3763                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3764                         leaf->map_token = NULL;
3765                 }
3766
3767                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3768                               btrfs_item_nr_offset(slot + nr),
3769                               sizeof(struct btrfs_item) *
3770                               (nritems - slot - nr));
3771         }
3772         btrfs_set_header_nritems(leaf, nritems - nr);
3773         nritems -= nr;
3774
3775         /* delete the leaf if we've emptied it */
3776         if (nritems == 0) {
3777                 if (leaf == root->node) {
3778                         btrfs_set_header_level(leaf, 0);
3779                 } else {
3780                         ret = btrfs_del_leaf(trans, root, path, leaf);
3781                         BUG_ON(ret);
3782                 }
3783         } else {
3784                 int used = leaf_space_used(leaf, 0, nritems);
3785                 if (slot == 0) {
3786                         struct btrfs_disk_key disk_key;
3787
3788                         btrfs_item_key(leaf, &disk_key, 0);
3789                         wret = fixup_low_keys(trans, root, path,
3790                                               &disk_key, 1);
3791                         if (wret)
3792                                 ret = wret;
3793                 }
3794
3795                 /* delete the leaf if it is mostly empty */
3796                 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
3797                         /* push_leaf_left fixes the path.
3798                          * make sure the path still points to our leaf
3799                          * for possible call to del_ptr below
3800                          */
3801                         slot = path->slots[1];
3802                         extent_buffer_get(leaf);
3803
3804                         btrfs_set_path_blocking(path);
3805                         wret = push_leaf_left(trans, root, path, 1, 1);
3806                         if (wret < 0 && wret != -ENOSPC)
3807                                 ret = wret;
3808
3809                         if (path->nodes[0] == leaf &&
3810                             btrfs_header_nritems(leaf)) {
3811                                 wret = push_leaf_right(trans, root, path, 1, 1);
3812                                 if (wret < 0 && wret != -ENOSPC)
3813                                         ret = wret;
3814                         }
3815
3816                         if (btrfs_header_nritems(leaf) == 0) {
3817                                 path->slots[1] = slot;
3818                                 ret = btrfs_del_leaf(trans, root, path, leaf);
3819                                 BUG_ON(ret);
3820                                 free_extent_buffer(leaf);
3821                         } else {
3822                                 /* if we're still in the path, make sure
3823                                  * we're dirty.  Otherwise, one of the
3824                                  * push_leaf functions must have already
3825                                  * dirtied this buffer
3826                                  */
3827                                 if (path->nodes[0] == leaf)
3828                                         btrfs_mark_buffer_dirty(leaf);
3829                                 free_extent_buffer(leaf);
3830                         }
3831                 } else {
3832                         btrfs_mark_buffer_dirty(leaf);
3833                 }
3834         }
3835         return ret;
3836 }
3837
3838 /*
3839  * search the tree again to find a leaf with lesser keys
3840  * returns 0 if it found something or 1 if there are no lesser leaves.
3841  * returns < 0 on io errors.
3842  *
3843  * This may release the path, and so you may lose any locks held at the
3844  * time you call it.
3845  */
3846 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3847 {
3848         struct btrfs_key key;
3849         struct btrfs_disk_key found_key;
3850         int ret;
3851
3852         btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3853
3854         if (key.offset > 0)
3855                 key.offset--;
3856         else if (key.type > 0)
3857                 key.type--;
3858         else if (key.objectid > 0)
3859                 key.objectid--;
3860         else
3861                 return 1;
3862
3863         btrfs_release_path(root, path);
3864         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3865         if (ret < 0)
3866                 return ret;
3867         btrfs_item_key(path->nodes[0], &found_key, 0);
3868         ret = comp_keys(&found_key, &key);
3869         if (ret < 0)
3870                 return 0;
3871         return 1;
3872 }
3873
3874 /*
3875  * A helper function to walk down the tree starting at min_key, and looking
3876  * for nodes or leaves that are either in cache or have a minimum
3877  * transaction id.  This is used by the btree defrag code, and tree logging
3878  *
3879  * This does not cow, but it does stuff the starting key it finds back
3880  * into min_key, so you can call btrfs_search_slot with cow=1 on the
3881  * key and get a writable path.
3882  *
3883  * This does lock as it descends, and path->keep_locks should be set
3884  * to 1 by the caller.
3885  *
3886  * This honors path->lowest_level to prevent descent past a given level
3887  * of the tree.
3888  *
3889  * min_trans indicates the oldest transaction that you are interested
3890  * in walking through.  Any nodes or leaves older than min_trans are
3891  * skipped over (without reading them).
3892  *
3893  * returns zero if something useful was found, < 0 on error and 1 if there
3894  * was nothing in the tree that matched the search criteria.
3895  */
3896 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3897                          struct btrfs_key *max_key,
3898                          struct btrfs_path *path, int cache_only,
3899                          u64 min_trans)
3900 {
3901         struct extent_buffer *cur;
3902         struct btrfs_key found_key;
3903         int slot;
3904         int sret;
3905         u32 nritems;
3906         int level;
3907         int ret = 1;
3908
3909         WARN_ON(!path->keep_locks);
3910 again:
3911         cur = btrfs_lock_root_node(root);
3912         level = btrfs_header_level(cur);
3913         WARN_ON(path->nodes[level]);
3914         path->nodes[level] = cur;
3915         path->locks[level] = 1;
3916
3917         if (btrfs_header_generation(cur) < min_trans) {
3918                 ret = 1;
3919                 goto out;
3920         }
3921         while (1) {
3922                 nritems = btrfs_header_nritems(cur);
3923                 level = btrfs_header_level(cur);
3924                 sret = bin_search(cur, min_key, level, &slot);
3925
3926                 /* at the lowest level, we're done, setup the path and exit */
3927                 if (level == path->lowest_level) {
3928                         if (slot >= nritems)
3929                                 goto find_next_key;
3930                         ret = 0;
3931                         path->slots[level] = slot;
3932                         btrfs_item_key_to_cpu(cur, &found_key, slot);
3933                         goto out;
3934                 }
3935                 if (sret && slot > 0)
3936                         slot--;
3937                 /*
3938                  * check this node pointer against the cache_only and
3939                  * min_trans parameters.  If it isn't in cache or is too
3940                  * old, skip to the next one.
3941                  */
3942                 while (slot < nritems) {
3943                         u64 blockptr;
3944                         u64 gen;
3945                         struct extent_buffer *tmp;
3946                         struct btrfs_disk_key disk_key;
3947
3948                         blockptr = btrfs_node_blockptr(cur, slot);
3949                         gen = btrfs_node_ptr_generation(cur, slot);
3950                         if (gen < min_trans) {
3951                                 slot++;
3952                                 continue;
3953                         }
3954                         if (!cache_only)
3955                                 break;
3956
3957                         if (max_key) {
3958                                 btrfs_node_key(cur, &disk_key, slot);
3959                                 if (comp_keys(&disk_key, max_key) >= 0) {
3960                                         ret = 1;
3961                                         goto out;
3962                                 }
3963                         }
3964
3965                         tmp = btrfs_find_tree_block(root, blockptr,
3966                                             btrfs_level_size(root, level - 1));
3967
3968                         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3969                                 free_extent_buffer(tmp);
3970                                 break;
3971                         }
3972                         if (tmp)
3973                                 free_extent_buffer(tmp);
3974                         slot++;
3975                 }
3976 find_next_key:
3977                 /*
3978                  * we didn't find a candidate key in this node, walk forward
3979                  * and find another one
3980                  */
3981                 if (slot >= nritems) {
3982                         path->slots[level] = slot;
3983                         btrfs_set_path_blocking(path);
3984                         sret = btrfs_find_next_key(root, path, min_key, level,
3985                                                   cache_only, min_trans);
3986                         if (sret == 0) {
3987                                 btrfs_release_path(root, path);
3988                                 goto again;
3989                         } else {
3990                                 goto out;
3991                         }
3992                 }
3993                 /* save our key for returning back */
3994                 btrfs_node_key_to_cpu(cur, &found_key, slot);
3995                 path->slots[level] = slot;
3996                 if (level == path->lowest_level) {
3997                         ret = 0;
3998                         unlock_up(path, level, 1);
3999                         goto out;
4000                 }
4001                 btrfs_set_path_blocking(path);
4002                 cur = read_node_slot(root, cur, slot);
4003
4004                 btrfs_tree_lock(cur);
4005
4006                 path->locks[level - 1] = 1;
4007                 path->nodes[level - 1] = cur;
4008                 unlock_up(path, level, 1);
4009                 btrfs_clear_path_blocking(path, NULL);
4010         }
4011 out:
4012         if (ret == 0)
4013                 memcpy(min_key, &found_key, sizeof(found_key));
4014         btrfs_set_path_blocking(path);
4015         return ret;
4016 }
4017
4018 /*
4019  * this is similar to btrfs_next_leaf, but does not try to preserve
4020  * and fixup the path.  It looks for and returns the next key in the
4021  * tree based on the current path and the cache_only and min_trans
4022  * parameters.
4023  *
4024  * 0 is returned if another key is found, < 0 if there are any errors
4025  * and 1 is returned if there are no higher keys in the tree
4026  *
4027  * path->keep_locks should be set to 1 on the search made before
4028  * calling this function.
4029  */
4030 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4031                         struct btrfs_key *key, int level,
4032                         int cache_only, u64 min_trans)
4033 {
4034         int slot;
4035         struct extent_buffer *c;
4036
4037         WARN_ON(!path->keep_locks);
4038         while (level < BTRFS_MAX_LEVEL) {
4039                 if (!path->nodes[level])
4040                         return 1;
4041
4042                 slot = path->slots[level] + 1;
4043                 c = path->nodes[level];
4044 next:
4045                 if (slot >= btrfs_header_nritems(c)) {
4046                         int ret;
4047                         int orig_lowest;
4048                         struct btrfs_key cur_key;
4049                         if (level + 1 >= BTRFS_MAX_LEVEL ||
4050                             !path->nodes[level + 1])
4051                                 return 1;
4052
4053                         if (path->locks[level + 1]) {
4054                                 level++;
4055                                 continue;
4056                         }
4057
4058                         slot = btrfs_header_nritems(c) - 1;
4059                         if (level == 0)
4060                                 btrfs_item_key_to_cpu(c, &cur_key, slot);
4061                         else
4062                                 btrfs_node_key_to_cpu(c, &cur_key, slot);
4063
4064                         orig_lowest = path->lowest_level;
4065                         btrfs_release_path(root, path);
4066                         path->lowest_level = level;
4067                         ret = btrfs_search_slot(NULL, root, &cur_key, path,
4068                                                 0, 0);
4069                         path->lowest_level = orig_lowest;
4070                         if (ret < 0)
4071                                 return ret;
4072
4073                         c = path->nodes[level];
4074                         slot = path->slots[level];
4075                         if (ret == 0)
4076                                 slot++;
4077                         goto next;
4078                 }
4079
4080                 if (level == 0)
4081                         btrfs_item_key_to_cpu(c, key, slot);
4082                 else {
4083                         u64 blockptr = btrfs_node_blockptr(c, slot);
4084                         u64 gen = btrfs_node_ptr_generation(c, slot);
4085
4086                         if (cache_only) {
4087                                 struct extent_buffer *cur;
4088                                 cur = btrfs_find_tree_block(root, blockptr,
4089                                             btrfs_level_size(root, level - 1));
4090                                 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4091                                         slot++;
4092                                         if (cur)
4093                                                 free_extent_buffer(cur);
4094                                         goto next;
4095                                 }
4096                                 free_extent_buffer(cur);
4097                         }
4098                         if (gen < min_trans) {
4099                                 slot++;
4100                                 goto next;
4101                         }
4102                         btrfs_node_key_to_cpu(c, key, slot);
4103                 }
4104                 return 0;
4105         }
4106         return 1;
4107 }
4108
4109 /*
4110  * search the tree again to find a leaf with greater keys
4111  * returns 0 if it found something or 1 if there are no greater leaves.
4112  * returns < 0 on io errors.
4113  */
4114 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4115 {
4116         int slot;
4117         int level;
4118         struct extent_buffer *c;
4119         struct extent_buffer *next;
4120         struct btrfs_key key;
4121         u32 nritems;
4122         int ret;
4123         int old_spinning = path->leave_spinning;
4124         int force_blocking = 0;
4125
4126         nritems = btrfs_header_nritems(path->nodes[0]);
4127         if (nritems == 0)
4128                 return 1;
4129
4130         /*
4131          * we take the blocks in an order that upsets lockdep.  Using
4132          * blocking mode is the only way around it.
4133          */
4134 #ifdef CONFIG_DEBUG_LOCK_ALLOC
4135         force_blocking = 1;
4136 #endif
4137
4138         btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4139 again:
4140         level = 1;
4141         next = NULL;
4142         btrfs_release_path(root, path);
4143
4144         path->keep_locks = 1;
4145
4146         if (!force_blocking)
4147                 path->leave_spinning = 1;
4148
4149         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4150         path->keep_locks = 0;
4151
4152         if (ret < 0)
4153                 return ret;
4154
4155         nritems = btrfs_header_nritems(path->nodes[0]);
4156         /*
4157          * by releasing the path above we dropped all our locks.  A balance
4158          * could have added more items next to the key that used to be
4159          * at the very end of the block.  So, check again here and
4160          * advance the path if there are now more items available.
4161          */
4162         if (nritems > 0 && path->slots[0] < nritems - 1) {
4163                 if (ret == 0)
4164                         path->slots[0]++;
4165                 ret = 0;
4166                 goto done;
4167         }
4168
4169         while (level < BTRFS_MAX_LEVEL) {
4170                 if (!path->nodes[level]) {
4171                         ret = 1;
4172                         goto done;
4173                 }
4174
4175                 slot = path->slots[level] + 1;
4176                 c = path->nodes[level];
4177                 if (slot >= btrfs_header_nritems(c)) {
4178                         level++;
4179                         if (level == BTRFS_MAX_LEVEL) {
4180                                 ret = 1;
4181                                 goto done;
4182                         }
4183                         continue;
4184                 }
4185
4186                 if (next) {
4187                         btrfs_tree_unlock(next);
4188                         free_extent_buffer(next);
4189                 }
4190
4191                 next = c;
4192                 ret = read_block_for_search(NULL, root, path, &next, level,
4193                                             slot, &key);
4194                 if (ret == -EAGAIN)
4195                         goto again;
4196
4197                 if (ret < 0) {
4198                         btrfs_release_path(root, path);
4199                         goto done;
4200                 }
4201
4202                 if (!path->skip_locking) {
4203                         ret = btrfs_try_spin_lock(next);
4204                         if (!ret) {
4205                                 btrfs_set_path_blocking(path);
4206                                 btrfs_tree_lock(next);
4207                                 if (!force_blocking)
4208                                         btrfs_clear_path_blocking(path, next);
4209                         }
4210                         if (force_blocking)
4211                                 btrfs_set_lock_blocking(next);
4212                 }
4213                 break;
4214         }
4215         path->slots[level] = slot;
4216         while (1) {
4217                 level--;
4218                 c = path->nodes[level];
4219                 if (path->locks[level])
4220                         btrfs_tree_unlock(c);
4221
4222                 free_extent_buffer(c);
4223                 path->nodes[level] = next;
4224                 path->slots[level] = 0;
4225                 if (!path->skip_locking)
4226                         path->locks[level] = 1;
4227
4228                 if (!level)
4229                         break;
4230
4231                 ret = read_block_for_search(NULL, root, path, &next, level,
4232                                             0, &key);
4233                 if (ret == -EAGAIN)
4234                         goto again;
4235
4236                 if (ret < 0) {
4237                         btrfs_release_path(root, path);
4238                         goto done;
4239                 }
4240
4241                 if (!path->skip_locking) {
4242                         btrfs_assert_tree_locked(path->nodes[level]);
4243                         ret = btrfs_try_spin_lock(next);
4244                         if (!ret) {
4245                                 btrfs_set_path_blocking(path);
4246                                 btrfs_tree_lock(next);
4247                                 if (!force_blocking)
4248                                         btrfs_clear_path_blocking(path, next);
4249                         }
4250                         if (force_blocking)
4251                                 btrfs_set_lock_blocking(next);
4252                 }
4253         }
4254         ret = 0;
4255 done:
4256         unlock_up(path, 0, 1);
4257         path->leave_spinning = old_spinning;
4258         if (!old_spinning)
4259                 btrfs_set_path_blocking(path);
4260
4261         return ret;
4262 }
4263
4264 /*
4265  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4266  * searching until it gets past min_objectid or finds an item of 'type'
4267  *
4268  * returns 0 if something is found, 1 if nothing was found and < 0 on error
4269  */
4270 int btrfs_previous_item(struct btrfs_root *root,
4271                         struct btrfs_path *path, u64 min_objectid,
4272                         int type)
4273 {
4274         struct btrfs_key found_key;
4275         struct extent_buffer *leaf;
4276         u32 nritems;
4277         int ret;
4278
4279         while (1) {
4280                 if (path->slots[0] == 0) {
4281                         btrfs_set_path_blocking(path);
4282                         ret = btrfs_prev_leaf(root, path);
4283                         if (ret != 0)
4284                                 return ret;
4285                 } else {
4286                         path->slots[0]--;
4287                 }
4288                 leaf = path->nodes[0];
4289                 nritems = btrfs_header_nritems(leaf);
4290                 if (nritems == 0)
4291                         return 1;
4292                 if (path->slots[0] == nritems)
4293                         path->slots[0]--;
4294
4295                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4296                 if (found_key.objectid < min_objectid)
4297                         break;
4298                 if (found_key.type == type)
4299                         return 0;
4300                 if (found_key.objectid == min_objectid &&
4301                     found_key.type < type)
4302                         break;
4303         }
4304         return 1;
4305 }