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 <linux/slab.h>
  21 #include "ctree.h"
  22 #include "disk-io.h"
  23 #include "transaction.h"
  24 #include "print-tree.h"
  25 #include "locking.h"
  26
  27 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
  28                       *root, struct btrfs_path *path, int level);
  29 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
  30                       *root, struct btrfs_key *ins_key,
  31                       struct btrfs_path *path, int data_size, int extend);
  32 static int push_node_left(struct btrfs_trans_handle *trans,
  33                           struct btrfs_root *root, struct extent_buffer *dst,
  34                           struct extent_buffer *src, int empty);
  35 static int balance_node_right(struct btrfs_trans_handle *trans,
  36                               struct btrfs_root *root,
  37                               struct extent_buffer *dst_buf,
  38                               struct extent_buffer *src_buf);
  39 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  40                    struct btrfs_path *path, int level, int slot);
  41 static int setup_items_for_insert(struct btrfs_trans_handle *trans,
  42                         struct btrfs_root *root, struct btrfs_path *path,
  43                         struct btrfs_key *cpu_key, u32 *data_size,
  44                         u32 total_data, u32 total_size, int nr);
  45
  46
  47 struct btrfs_path *btrfs_alloc_path(void)
  48 {
  49         struct btrfs_path *path;
  50         path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
  51         if (path)
  52                 path->reada = 1;
  53         return path;
  54 }
  55
  56 /*
  57  * set all locked nodes in the path to blocking locks.  This should
  58  * be done before scheduling
  59  */
  60 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
  61 {
  62         int i;
  63         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
  64                 if (p->nodes[i] && p->locks[i])
  65                         btrfs_set_lock_blocking(p->nodes[i]);
  66         }
  67 }
  68
  69 /*
  70  * reset all the locked nodes in the patch to spinning locks.
  71  *
  72  * held is used to keep lockdep happy, when lockdep is enabled
  73  * we set held to a blocking lock before we go around and
  74  * retake all the spinlocks in the path.  You can safely use NULL
  75  * for held
  76  */
  77 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
  78                                         struct extent_buffer *held)
  79 {
  80         int i;
  81
  82 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  83         /* lockdep really cares that we take all of these spinlocks
  84          * in the right order.  If any of the locks in the path are not
  85          * currently blocking, it is going to complain.  So, make really
  86          * really sure by forcing the path to blocking before we clear
  87          * the path blocking.
  88          */
  89         if (held)
  90                 btrfs_set_lock_blocking(held);
  91         btrfs_set_path_blocking(p);
  92 #endif
  93
  94         for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
  95                 if (p->nodes[i] && p->locks[i])
  96                         btrfs_clear_lock_blocking(p->nodes[i]);
  97         }
  98
  99 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 100         if (held)
 101                 btrfs_clear_lock_blocking(held);
 102 #endif
 103 }
 104
 105 /* this also releases the path */
 106 void btrfs_free_path(struct btrfs_path *p)
 107 {
 108         btrfs_release_path(NULL, p);
 109         kmem_cache_free(btrfs_path_cachep, p);
 110 }
 111
 112 /*
 113  * path release drops references on the extent buffers in the path
 114  * and it drops any locks held by this path
 115  *
 116  * It is safe to call this on paths that no locks or extent buffers held.
 117  */
 118 noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
 119 {
 120         int i;
 121
 122         for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
 123                 p->slots[i] = 0;
 124                 if (!p->nodes[i])
 125                         continue;
 126                 if (p->locks[i]) {
 127                         btrfs_tree_unlock(p->nodes[i]);
 128                         p->locks[i] = 0;
 129                 }
 130                 free_extent_buffer(p->nodes[i]);
 131                 p->nodes[i] = NULL;
 132         }
 133 }
 134
 135 /*
 136  * safely gets a reference on the root node of a tree.  A lock
 137  * is not taken, so a concurrent writer may put a different node
 138  * at the root of the tree.  See btrfs_lock_root_node for the
 139  * looping required.
 140  *
 141  * The extent buffer returned by this has a reference taken, so
 142  * it won't disappear.  It may stop being the root of the tree
 143  * at any time because there are no locks held.
 144  */
 145 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
 146 {
 147         struct extent_buffer *eb;
 148         spin_lock(&root->node_lock);
 149         eb = root->node;
 150         extent_buffer_get(eb);
 151         spin_unlock(&root->node_lock);
 152         return eb;
 153 }
 154
 155 /* loop around taking references on and locking the root node of the
 156  * tree until you end up with a lock on the root.  A locked buffer
 157  * is returned, with a reference held.
 158  */
 159 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
 160 {
 161         struct extent_buffer *eb;
 162
 163         while (1) {
 164                 eb = btrfs_root_node(root);
 165                 btrfs_tree_lock(eb);
 166
 167                 spin_lock(&root->node_lock);
 168                 if (eb == root->node) {
 169                         spin_unlock(&root->node_lock);
 170                         break;
 171                 }
 172                 spin_unlock(&root->node_lock);
 173
 174                 btrfs_tree_unlock(eb);
 175                 free_extent_buffer(eb);
 176         }
 177         return eb;
 178 }
 179
 180 /* cowonly root (everything not a reference counted cow subvolume), just get
 181  * put onto a simple dirty list.  transaction.c walks this to make sure they
 182  * get properly updated on disk.
 183  */
 184 static void add_root_to_dirty_list(struct btrfs_root *root)
 185 {
 186         if (root->track_dirty && list_empty(&root->dirty_list)) {
 187                 list_add(&root->dirty_list,
 188                          &root->fs_info->dirty_cowonly_roots);
 189         }
 190 }
 191
 192 /*
 193  * used by snapshot creation to make a copy of a root for a tree with
 194  * a given objectid.  The buffer with the new root node is returned in
 195  * cow_ret, and this func returns zero on success or a negative error code.
 196  */
 197 int btrfs_copy_root(struct btrfs_trans_handle *trans,
 198                       struct btrfs_root *root,
 199                       struct extent_buffer *buf,
 200                       struct extent_buffer **cow_ret, u64 new_root_objectid)
 201 {
 202         struct extent_buffer *cow;
 203         u32 nritems;
 204         int ret = 0;
 205         int level;
 206         struct btrfs_disk_key disk_key;
 207
 208         WARN_ON(root->ref_cows && trans->transid !=
 209                 root->fs_info->running_transaction->transid);
 210         WARN_ON(root->ref_cows && trans->transid != root->last_trans);
 211
 212         level = btrfs_header_level(buf);
 213         nritems = btrfs_header_nritems(buf);
 214         if (level == 0)
 215                 btrfs_item_key(buf, &disk_key, 0);
 216         else
 217                 btrfs_node_key(buf, &disk_key, 0);
 218
 219         cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
 220                                      new_root_objectid, &disk_key, level,
 221                                      buf->start, 0);
 222         if (IS_ERR(cow))
 223                 return PTR_ERR(cow);
 224
 225         copy_extent_buffer(cow, buf, 0, 0, cow->len);
 226         btrfs_set_header_bytenr(cow, cow->start);
 227         btrfs_set_header_generation(cow, trans->transid);
 228         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
 229         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
 230                                      BTRFS_HEADER_FLAG_RELOC);
 231         if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
 232                 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
 233         else
 234                 btrfs_set_header_owner(cow, new_root_objectid);
 235
 236         write_extent_buffer(cow, root->fs_info->fsid,
 237                             (unsigned long)btrfs_header_fsid(cow),
 238                             BTRFS_FSID_SIZE);
 239
 240         WARN_ON(btrfs_header_generation(buf) > trans->transid);
 241         if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
 242                 ret = btrfs_inc_ref(trans, root, cow, 1);
 243         else
 244                 ret = btrfs_inc_ref(trans, root, cow, 0);
 245
 246         if (ret)
 247                 return ret;
 248
 249         btrfs_mark_buffer_dirty(cow);
 250         *cow_ret = cow;
 251         return 0;
 252 }
 253
 254 /*
 255  * check if the tree block can be shared by multiple trees
 256  */
 257 int btrfs_block_can_be_shared(struct btrfs_root *root,
 258                               struct extent_buffer *buf)
 259 {
 260         /*
 261          * Tree blocks not in refernece counted trees and tree roots
 262          * are never shared. If a block was allocated after the last
 263          * snapshot and the block was not allocated by tree relocation,
 264          * we know the block is not shared.
 265          */
 266         if (root->ref_cows &&
 267             buf != root->node && buf != root->commit_root &&
 268             (btrfs_header_generation(buf) <=
 269              btrfs_root_last_snapshot(&root->root_item) ||
 270              btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
 271                 return 1;
 272 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 273         if (root->ref_cows &&
 274             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
 275                 return 1;
 276 #endif
 277         return 0;
 278 }
 279
 280 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
 281                                        struct btrfs_root *root,
 282                                        struct extent_buffer *buf,
 283                                        struct extent_buffer *cow,
 284                                        int *last_ref)
 285 {
 286         u64 refs;
 287         u64 owner;
 288         u64 flags;
 289         u64 new_flags = 0;
 290         int ret;
 291
 292         /*
 293          * Backrefs update rules:
 294          *
 295          * Always use full backrefs for extent pointers in tree block
 296          * allocated by tree relocation.
 297          *
 298          * If a shared tree block is no longer referenced by its owner
 299          * tree (btrfs_header_owner(buf) == root->root_key.objectid),
 300          * use full backrefs for extent pointers in tree block.
 301          *
 302          * If a tree block is been relocating
 303          * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
 304          * use full backrefs for extent pointers in tree block.
 305          * The reason for this is some operations (such as drop tree)
 306          * are only allowed for blocks use full backrefs.
 307          */
 308
 309         if (btrfs_block_can_be_shared(root, buf)) {
 310                 ret = btrfs_lookup_extent_info(trans, root, buf->start,
 311                                                buf->len, &refs, &flags);
 312                 BUG_ON(ret);
 313                 BUG_ON(refs == 0);
 314         } else {
 315                 refs = 1;
 316                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
 317                     btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
 318                         flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 319                 else
 320                         flags = 0;
 321         }
 322
 323         owner = btrfs_header_owner(buf);
 324         BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
 325                !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
 326
 327         if (refs > 1) {
 328                 if ((owner == root->root_key.objectid ||
 329                      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
 330                     !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
 331                         ret = btrfs_inc_ref(trans, root, buf, 1);
 332                         BUG_ON(ret);
 333
 334                         if (root->root_key.objectid ==
 335                             BTRFS_TREE_RELOC_OBJECTID) {
 336                                 ret = btrfs_dec_ref(trans, root, buf, 0);
 337                                 BUG_ON(ret);
 338                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 339                                 BUG_ON(ret);
 340                         }
 341                         new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
 342                 } else {
 343
 344                         if (root->root_key.objectid ==
 345                             BTRFS_TREE_RELOC_OBJECTID)
 346                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 347                         else
 348                                 ret = btrfs_inc_ref(trans, root, cow, 0);
 349                         BUG_ON(ret);
 350                 }
 351                 if (new_flags != 0) {
 352                         ret = btrfs_set_disk_extent_flags(trans, root,
 353                                                           buf->start,
 354                                                           buf->len,
 355                                                           new_flags, 0);
 356                         BUG_ON(ret);
 357                 }
 358         } else {
 359                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
 360                         if (root->root_key.objectid ==
 361                             BTRFS_TREE_RELOC_OBJECTID)
 362                                 ret = btrfs_inc_ref(trans, root, cow, 1);
 363                         else
 364                                 ret = btrfs_inc_ref(trans, root, cow, 0);
 365                         BUG_ON(ret);
 366                         ret = btrfs_dec_ref(trans, root, buf, 1);
 367                         BUG_ON(ret);
 368                 }
 369                 clean_tree_block(trans, root, buf);
 370                 *last_ref = 1;
 371         }
 372         return 0;
 373 }
 374
 375 /*
 376  * does the dirty work in cow of a single block.  The parent block (if
 377  * supplied) is updated to point to the new cow copy.  The new buffer is marked
 378  * dirty and returned locked.  If you modify the block it needs to be marked
 379  * dirty again.
 380  *
 381  * search_start -- an allocation hint for the new block
 382  *
 383  * empty_size -- a hint that you plan on doing more cow.  This is the size in
 384  * bytes the allocator should try to find free next to the block it returns.
 385  * This is just a hint and may be ignored by the allocator.
 386  */
 387 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 388                              struct btrfs_root *root,
 389                              struct extent_buffer *buf,
 390                              struct extent_buffer *parent, int parent_slot,
 391                              struct extent_buffer **cow_ret,
 392                              u64 search_start, u64 empty_size)
 393 {
 394         struct btrfs_disk_key disk_key;
 395         struct extent_buffer *cow;
 396         int level;
 397         int last_ref = 0;
 398         int unlock_orig = 0;
 399         u64 parent_start;
 400
 401         if (*cow_ret == buf)
 402                 unlock_orig = 1;
 403
 404         btrfs_assert_tree_locked(buf);
 405
 406         WARN_ON(root->ref_cows && trans->transid !=
 407                 root->fs_info->running_transaction->transid);
 408         WARN_ON(root->ref_cows && trans->transid != root->last_trans);
 409
 410         level = btrfs_header_level(buf);
 411
 412         if (level == 0)
 413                 btrfs_item_key(buf, &disk_key, 0);
 414         else
 415                 btrfs_node_key(buf, &disk_key, 0);
 416
 417         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
 418                 if (parent)
 419                         parent_start = parent->start;
 420                 else
 421                         parent_start = 0;
 422         } else
 423                 parent_start = 0;
 424
 425         cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
 426                                      root->root_key.objectid, &disk_key,
 427                                      level, search_start, empty_size);
 428         if (IS_ERR(cow))
 429                 return PTR_ERR(cow);
 430
 431         /* cow is set to blocking by btrfs_init_new_buffer */
 432
 433         copy_extent_buffer(cow, buf, 0, 0, cow->len);
 434         btrfs_set_header_bytenr(cow, cow->start);
 435         btrfs_set_header_generation(cow, trans->transid);
 436         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
 437         btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
 438                                      BTRFS_HEADER_FLAG_RELOC);
 439         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
 440                 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
 441         else
 442                 btrfs_set_header_owner(cow, root->root_key.objectid);
 443
 444         write_extent_buffer(cow, root->fs_info->fsid,
 445                             (unsigned long)btrfs_header_fsid(cow),
 446                             BTRFS_FSID_SIZE);
 447
 448         update_ref_for_cow(trans, root, buf, cow, &last_ref);
 449
 450         if (buf == root->node) {
 451                 WARN_ON(parent && parent != buf);
 452                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
 453                     btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
 454                         parent_start = buf->start;
 455                 else
 456                         parent_start = 0;
 457
 458                 spin_lock(&root->node_lock);
 459                 root->node = cow;
 460                 extent_buffer_get(cow);
 461                 spin_unlock(&root->node_lock);
 462
 463                 btrfs_free_tree_block(trans, root, buf, parent_start,
 464                                       last_ref);
 465                 free_extent_buffer(buf);
 466                 add_root_to_dirty_list(root);
 467         } else {
 468                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
 469                         parent_start = parent->start;
 470                 else
 471                         parent_start = 0;
 472
 473                 WARN_ON(trans->transid != btrfs_header_generation(parent));
 474                 btrfs_set_node_blockptr(parent, parent_slot,
 475                                         cow->start);
 476                 btrfs_set_node_ptr_generation(parent, parent_slot,
 477                                               trans->transid);
 478                 btrfs_mark_buffer_dirty(parent);
 479                 btrfs_free_tree_block(trans, root, buf, parent_start,
 480                                       last_ref);
 481         }
 482         if (unlock_orig)
 483                 btrfs_tree_unlock(buf);
 484         free_extent_buffer(buf);
 485         btrfs_mark_buffer_dirty(cow);
 486         *cow_ret = cow;
 487         return 0;
 488 }
 489
 490 static inline int should_cow_block(struct btrfs_trans_handle *trans,
 491                                    struct btrfs_root *root,
 492                                    struct extent_buffer *buf)
 493 {
 494         if (btrfs_header_generation(buf) == trans->transid &&
 495             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
 496             !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
 497               btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
 498                 return 0;
 499         return 1;
 500 }
 501
 502 /*
 503  * cows a single block, see __btrfs_cow_block for the real work.
 504  * This version of it has extra checks so that a block isn't cow'd more than
 505  * once per transaction, as long as it hasn't been written yet
 506  */
 507 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
 508                     struct btrfs_root *root, struct extent_buffer *buf,
 509                     struct extent_buffer *parent, int parent_slot,
 510                     struct extent_buffer **cow_ret)
 511 {
 512         u64 search_start;
 513         int ret;
 514
 515         if (trans->transaction != root->fs_info->running_transaction) {
 516                 printk(KERN_CRIT "trans %llu running %llu\n",
 517                        (unsigned long long)trans->transid,
 518                        (unsigned long long)
 519                        root->fs_info->running_transaction->transid);
 520                 WARN_ON(1);
 521         }
 522         if (trans->transid != root->fs_info->generation) {
 523                 printk(KERN_CRIT "trans %llu running %llu\n",
 524                        (unsigned long long)trans->transid,
 525                        (unsigned long long)root->fs_info->generation);
 526                 WARN_ON(1);
 527         }
 528
 529         if (!should_cow_block(trans, root, buf)) {
 530                 *cow_ret = buf;
 531                 return 0;
 532         }
 533
 534         search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
 535
 536         if (parent)
 537                 btrfs_set_lock_blocking(parent);
 538         btrfs_set_lock_blocking(buf);
 539
 540         ret = __btrfs_cow_block(trans, root, buf, parent,
 541                                  parent_slot, cow_ret, search_start, 0);
 542         return ret;
 543 }
 544
 545 /*
 546  * helper function for defrag to decide if two blocks pointed to by a
 547  * node are actually close by
 548  */
 549 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
 550 {
 551         if (blocknr < other && other - (blocknr + blocksize) < 32768)
 552                 return 1;
 553         if (blocknr > other && blocknr - (other + blocksize) < 32768)
 554                 return 1;
 555         return 0;
 556 }
 557
 558 /*
 559  * compare two keys in a memcmp fashion
 560  */
 561 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
 562 {
 563         struct btrfs_key k1;
 564
 565         btrfs_disk_key_to_cpu(&k1, disk);
 566
 567         return btrfs_comp_cpu_keys(&k1, k2);
 568 }
 569
 570 /*
 571  * same as comp_keys only with two btrfs_key's
 572  */
 573 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
 574 {
 575         if (k1->objectid > k2->objectid)
 576                 return 1;
 577         if (k1->objectid < k2->objectid)
 578                 return -1;
 579         if (k1->type > k2->type)
 580                 return 1;
 581         if (k1->type < k2->type)
 582                 return -1;
 583         if (k1->offset > k2->offset)
 584                 return 1;
 585         if (k1->offset < k2->offset)
 586                 return -1;
 587         return 0;
 588 }
 589
 590 /*
 591  * this is used by the defrag code to go through all the
 592  * leaves pointed to by a node and reallocate them so that
 593  * disk order is close to key order
 594  */
 595 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
 596                        struct btrfs_root *root, struct extent_buffer *parent,
 597                        int start_slot, int cache_only, u64 *last_ret,
 598                        struct btrfs_key *progress)
 599 {
 600         struct extent_buffer *cur;
 601         u64 blocknr;
 602         u64 gen;
 603         u64 search_start = *last_ret;
 604         u64 last_block = 0;
 605         u64 other;
 606         u32 parent_nritems;
 607         int end_slot;
 608         int i;
 609         int err = 0;
 610         int parent_level;
 611         int uptodate;
 612         u32 blocksize;
 613         int progress_passed = 0;
 614         struct btrfs_disk_key disk_key;
 615
 616         parent_level = btrfs_header_level(parent);
 617         if (cache_only && parent_level != 1)
 618                 return 0;
 619
 620         if (trans->transaction != root->fs_info->running_transaction)
 621                 WARN_ON(1);
 622         if (trans->transid != root->fs_info->generation)
 623                 WARN_ON(1);
 624
 625         parent_nritems = btrfs_header_nritems(parent);
 626         blocksize = btrfs_level_size(root, parent_level - 1);
 627         end_slot = parent_nritems;
 628
 629         if (parent_nritems == 1)
 630                 return 0;
 631
 632         btrfs_set_lock_blocking(parent);
 633
 634         for (i = start_slot; i < end_slot; i++) {
 635                 int close = 1;
 636
 637                 if (!parent->map_token) {
 638                         map_extent_buffer(parent,
 639                                         btrfs_node_key_ptr_offset(i),
 640                                         sizeof(struct btrfs_key_ptr),
 641                                         &parent->map_token, &parent->kaddr,
 642                                         &parent->map_start, &parent->map_len,
 643                                         KM_USER1);
 644                 }
 645                 btrfs_node_key(parent, &disk_key, i);
 646                 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
 647                         continue;
 648
 649                 progress_passed = 1;
 650                 blocknr = btrfs_node_blockptr(parent, i);
 651                 gen = btrfs_node_ptr_generation(parent, i);
 652                 if (last_block == 0)
 653                         last_block = blocknr;
 654
 655                 if (i > 0) {
 656                         other = btrfs_node_blockptr(parent, i - 1);
 657                         close = close_blocks(blocknr, other, blocksize);
 658                 }
 659                 if (!close && i < end_slot - 2) {
 660                         other = btrfs_node_blockptr(parent, i + 1);
 661                         close = close_blocks(blocknr, other, blocksize);
 662                 }
 663                 if (close) {
 664                         last_block = blocknr;
 665                         continue;
 666                 }
 667                 if (parent->map_token) {
 668                         unmap_extent_buffer(parent, parent->map_token,
 669                                             KM_USER1);
 670                         parent->map_token = NULL;
 671                 }
 672
 673                 cur = btrfs_find_tree_block(root, blocknr, blocksize);
 674                 if (cur)
 675                         uptodate = btrfs_buffer_uptodate(cur, gen);
 676                 else
 677                         uptodate = 0;
 678                 if (!cur || !uptodate) {
 679                         if (cache_only) {
 680                                 free_extent_buffer(cur);
 681                                 continue;
 682                         }
 683                         if (!cur) {
 684                                 cur = read_tree_block(root, blocknr,
 685                                                          blocksize, gen);
 686                         } else if (!uptodate) {
 687                                 btrfs_read_buffer(cur, gen);
 688                         }
 689                 }
 690                 if (search_start == 0)
 691                         search_start = last_block;
 692
 693                 btrfs_tree_lock(cur);
 694                 btrfs_set_lock_blocking(cur);
 695                 err = __btrfs_cow_block(trans, root, cur, parent, i,
 696                                         &cur, search_start,
 697                                         min(16 * blocksize,
 698                                             (end_slot - i) * blocksize));
 699                 if (err) {
 700                         btrfs_tree_unlock(cur);
 701                         free_extent_buffer(cur);
 702                         break;
 703                 }
 704                 search_start = cur->start;
 705                 last_block = cur->start;
 706                 *last_ret = search_start;
 707                 btrfs_tree_unlock(cur);
 708                 free_extent_buffer(cur);
 709         }
 710         if (parent->map_token) {
 711                 unmap_extent_buffer(parent, parent->map_token,
 712                                     KM_USER1);
 713                 parent->map_token = NULL;
 714         }
 715         return err;
 716 }
 717
 718 /*
 719  * The leaf data grows from end-to-front in the node.
 720  * this returns the address of the start of the last item,
 721  * which is the stop of the leaf data stack
 722  */
 723 static inline unsigned int leaf_data_end(struct btrfs_root *root,
 724                                          struct extent_buffer *leaf)
 725 {
 726         u32 nr = btrfs_header_nritems(leaf);
 727         if (nr == 0)
 728                 return BTRFS_LEAF_DATA_SIZE(root);
 729         return btrfs_item_offset_nr(leaf, nr - 1);
 730 }
 731
 732 /*
 733  * extra debugging checks to make sure all the items in a key are
 734  * well formed and in the proper order
 735  */
 736 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
 737                       int level)
 738 {
 739         struct extent_buffer *parent = NULL;
 740         struct extent_buffer *node = path->nodes[level];
 741         struct btrfs_disk_key parent_key;
 742         struct btrfs_disk_key node_key;
 743         int parent_slot;
 744         int slot;
 745         struct btrfs_key cpukey;
 746         u32 nritems = btrfs_header_nritems(node);
 747
 748         if (path->nodes[level + 1])
 749                 parent = path->nodes[level + 1];
 750
 751         slot = path->slots[level];
 752         BUG_ON(nritems == 0);
 753         if (parent) {
 754                 parent_slot = path->slots[level + 1];
 755                 btrfs_node_key(parent, &parent_key, parent_slot);
 756                 btrfs_node_key(node, &node_key, 0);
 757                 BUG_ON(memcmp(&parent_key, &node_key,
 758                               sizeof(struct btrfs_disk_key)));
 759                 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
 760                        btrfs_header_bytenr(node));
 761         }
 762         BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
 763         if (slot != 0) {
 764                 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
 765                 btrfs_node_key(node, &node_key, slot);
 766                 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
 767         }
 768         if (slot < nritems - 1) {
 769                 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
 770                 btrfs_node_key(node, &node_key, slot);
 771                 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
 772         }
 773         return 0;
 774 }
 775
 776 /*
 777  * extra checking to make sure all the items in a leaf are
 778  * well formed and in the proper order
 779  */
 780 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
 781                       int level)
 782 {
 783         struct extent_buffer *leaf = path->nodes[level];
 784         struct extent_buffer *parent = NULL;
 785         int parent_slot;
 786         struct btrfs_key cpukey;
 787         struct btrfs_disk_key parent_key;
 788         struct btrfs_disk_key leaf_key;
 789         int slot = path->slots[0];
 790
 791         u32 nritems = btrfs_header_nritems(leaf);
 792
 793         if (path->nodes[level + 1])
 794                 parent = path->nodes[level + 1];
 795
 796         if (nritems == 0)
 797                 return 0;
 798
 799         if (parent) {
 800                 parent_slot = path->slots[level + 1];
 801                 btrfs_node_key(parent, &parent_key, parent_slot);
 802                 btrfs_item_key(leaf, &leaf_key, 0);
 803
 804                 BUG_ON(memcmp(&parent_key, &leaf_key,
 805                        sizeof(struct btrfs_disk_key)));
 806                 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
 807                        btrfs_header_bytenr(leaf));
 808         }
 809         if (slot != 0 && slot < nritems - 1) {
 810                 btrfs_item_key(leaf, &leaf_key, slot);
 811                 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
 812                 if (comp_keys(&leaf_key, &cpukey) <= 0) {
 813                         btrfs_print_leaf(root, leaf);
 814                         printk(KERN_CRIT "slot %d offset bad key\n", slot);
 815                         BUG_ON(1);
 816                 }
 817                 if (btrfs_item_offset_nr(leaf, slot - 1) !=
 818                        btrfs_item_end_nr(leaf, slot)) {
 819                         btrfs_print_leaf(root, leaf);
 820                         printk(KERN_CRIT "slot %d offset bad\n", slot);
 821                         BUG_ON(1);
 822                 }
 823         }
 824         if (slot < nritems - 1) {
 825                 btrfs_item_key(leaf, &leaf_key, slot);
 826                 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
 827                 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
 828                 if (btrfs_item_offset_nr(leaf, slot) !=
 829                         btrfs_item_end_nr(leaf, slot + 1)) {
 830                         btrfs_print_leaf(root, leaf);
 831                         printk(KERN_CRIT "slot %d offset bad\n", slot);
 832                         BUG_ON(1);
 833                 }
 834         }
 835         BUG_ON(btrfs_item_offset_nr(leaf, 0) +
 836                btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
 837         return 0;
 838 }
 839
 840 static noinline int check_block(struct btrfs_root *root,
 841                                 struct btrfs_path *path, int level)
 842 {
 843         return 0;
 844         if (level == 0)
 845                 return check_leaf(root, path, level);
 846         return check_node(root, path, level);
 847 }
 848
 849 /*
 850  * search for key in the extent_buffer.  The items start at offset p,
 851  * and they are item_size apart.  There are 'max' items in p.
 852  *
 853  * the slot in the array is returned via slot, and it points to
 854  * the place where you would insert key if it is not found in
 855  * the array.
 856  *
 857  * slot may point to max if the key is bigger than all of the keys
 858  */
 859 static noinline int generic_bin_search(struct extent_buffer *eb,
 860                                        unsigned long p,
 861                                        int item_size, struct btrfs_key *key,
 862                                        int max, int *slot)
 863 {
 864         int low = 0;
 865         int high = max;
 866         int mid;
 867         int ret;
 868         struct btrfs_disk_key *tmp = NULL;
 869         struct btrfs_disk_key unaligned;
 870         unsigned long offset;
 871         char *map_token = NULL;
 872         char *kaddr = NULL;
 873         unsigned long map_start = 0;
 874         unsigned long map_len = 0;
 875         int err;
 876
 877         while (low < high) {
 878                 mid = (low + high) / 2;
 879                 offset = p + mid * item_size;
 880
 881                 if (!map_token || offset < map_start ||
 882                     (offset + sizeof(struct btrfs_disk_key)) >
 883                     map_start + map_len) {
 884                         if (map_token) {
 885                                 unmap_extent_buffer(eb, map_token, KM_USER0);
 886                                 map_token = NULL;
 887                         }
 888
 889                         err = map_private_extent_buffer(eb, offset,
 890                                                 sizeof(struct btrfs_disk_key),
 891                                                 &map_token, &kaddr,
 892                                                 &map_start, &map_len, KM_USER0);
 893
 894                         if (!err) {
 895                                 tmp = (struct btrfs_disk_key *)(kaddr + offset -
 896                                                         map_start);
 897                         } else {
 898                                 read_extent_buffer(eb, &unaligned,
 899                                                    offset, sizeof(unaligned));
 900                                 tmp = &unaligned;
 901                         }
 902
 903                 } else {
 904                         tmp = (struct btrfs_disk_key *)(kaddr + offset -
 905                                                         map_start);
 906                 }
 907                 ret = comp_keys(tmp, key);
 908
 909                 if (ret < 0)
 910                         low = mid + 1;
 911                 else if (ret > 0)
 912                         high = mid;
 913                 else {
 914                         *slot = mid;
 915                         if (map_token)
 916                                 unmap_extent_buffer(eb, map_token, KM_USER0);
 917                         return 0;
 918                 }
 919         }
 920         *slot = low;
 921         if (map_token)
 922                 unmap_extent_buffer(eb, map_token, KM_USER0);
 923         return 1;
 924 }
 925
 926 /*
 927  * simple bin_search frontend that does the right thing for
 928  * leaves vs nodes
 929  */
 930 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
 931                       int level, int *slot)
 932 {
 933         if (level == 0) {
 934                 return generic_bin_search(eb,
 935                                           offsetof(struct btrfs_leaf, items),
 936                                           sizeof(struct btrfs_item),
 937                                           key, btrfs_header_nritems(eb),
 938                                           slot);
 939         } else {
 940                 return generic_bin_search(eb,
 941                                           offsetof(struct btrfs_node, ptrs),
 942                                           sizeof(struct btrfs_key_ptr),
 943                                           key, btrfs_header_nritems(eb),
 944                                           slot);
 945         }
 946         return -1;
 947 }
 948
 949 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
 950                      int level, int *slot)
 951 {
 952         return bin_search(eb, key, level, slot);
 953 }
 954
 955 static void root_add_used(struct btrfs_root *root, u32 size)
 956 {
 957         spin_lock(&root->accounting_lock);
 958         btrfs_set_root_used(&root->root_item,
 959                             btrfs_root_used(&root->root_item) + size);
 960         spin_unlock(&root->accounting_lock);
 961 }
 962
 963 static void root_sub_used(struct btrfs_root *root, u32 size)
 964 {
 965         spin_lock(&root->accounting_lock);
 966         btrfs_set_root_used(&root->root_item,
 967                             btrfs_root_used(&root->root_item) - size);
 968         spin_unlock(&root->accounting_lock);
 969 }
 970
 971 /* given a node and slot number, this reads the blocks it points to.  The
 972  * extent buffer is returned with a reference taken (but unlocked).
 973  * NULL is returned on error.
 974  */
 975 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
 976                                    struct extent_buffer *parent, int slot)
 977 {
 978         int level = btrfs_header_level(parent);
 979         if (slot < 0)
 980                 return NULL;
 981         if (slot >= btrfs_header_nritems(parent))
 982                 return NULL;
 983
 984         BUG_ON(level == 0);
 985
 986         return read_tree_block(root, btrfs_node_blockptr(parent, slot),
 987                        btrfs_level_size(root, level - 1),
 988                        btrfs_node_ptr_generation(parent, slot));
 989 }
 990
 991 /*
 992  * node level balancing, used to make sure nodes are in proper order for
 993  * item deletion.  We balance from the top down, so we have to make sure
 994  * that a deletion won't leave an node completely empty later on.
 995  */
 996 static noinline int balance_level(struct btrfs_trans_handle *trans,
 997                          struct btrfs_root *root,
 998                          struct btrfs_path *path, int level)
 999 {
1000         struct extent_buffer *right = NULL;
1001         struct extent_buffer *mid;
1002         struct extent_buffer *left = NULL;
1003         struct extent_buffer *parent = NULL;
1004         int ret = 0;
1005         int wret;
1006         int pslot;
1007         int orig_slot = path->slots[level];
1008         int err_on_enospc = 0;
1009         u64 orig_ptr;
1010
1011         if (level == 0)
1012                 return 0;
1013
1014         mid = path->nodes[level];
1015
1016         WARN_ON(!path->locks[level]);
1017         WARN_ON(btrfs_header_generation(mid) != trans->transid);
1018
1019         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1020
1021         if (level < BTRFS_MAX_LEVEL - 1)
1022                 parent = path->nodes[level + 1];
1023         pslot = path->slots[level + 1];
1024
1025         /*
1026          * deal with the case where there is only one pointer in the root
1027          * by promoting the node below to a root
1028          */
1029         if (!parent) {
1030                 struct extent_buffer *child;
1031
1032                 if (btrfs_header_nritems(mid) != 1)
1033                         return 0;
1034
1035                 /* promote the child to a root */
1036                 child = read_node_slot(root, mid, 0);
1037                 BUG_ON(!child);
1038                 btrfs_tree_lock(child);
1039                 btrfs_set_lock_blocking(child);
1040                 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1041                 if (ret) {
1042                         btrfs_tree_unlock(child);
1043                         free_extent_buffer(child);
1044                         goto enospc;
1045                 }
1046
1047                 spin_lock(&root->node_lock);
1048                 root->node = child;
1049                 spin_unlock(&root->node_lock);
1050
1051                 add_root_to_dirty_list(root);
1052                 btrfs_tree_unlock(child);
1053
1054                 path->locks[level] = 0;
1055                 path->nodes[level] = NULL;
1056                 clean_tree_block(trans, root, mid);
1057                 btrfs_tree_unlock(mid);
1058                 /* once for the path */
1059                 free_extent_buffer(mid);
1060
1061                 root_sub_used(root, mid->len);
1062                 btrfs_free_tree_block(trans, root, mid, 0, 1);
1063                 /* once for the root ptr */
1064                 free_extent_buffer(mid);
1065                 return 0;
1066         }
1067         if (btrfs_header_nritems(mid) >
1068             BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1069                 return 0;
1070
1071         if (btrfs_header_nritems(mid) < 2)
1072                 err_on_enospc = 1;
1073
1074         left = read_node_slot(root, parent, pslot - 1);
1075         if (left) {
1076                 btrfs_tree_lock(left);
1077                 btrfs_set_lock_blocking(left);
1078                 wret = btrfs_cow_block(trans, root, left,
1079                                        parent, pslot - 1, &left);
1080                 if (wret) {
1081                         ret = wret;
1082                         goto enospc;
1083                 }
1084         }
1085         right = read_node_slot(root, parent, pslot + 1);
1086         if (right) {
1087                 btrfs_tree_lock(right);
1088                 btrfs_set_lock_blocking(right);
1089                 wret = btrfs_cow_block(trans, root, right,
1090                                        parent, pslot + 1, &right);
1091                 if (wret) {
1092                         ret = wret;
1093                         goto enospc;
1094                 }
1095         }
1096
1097         /* first, try to make some room in the middle buffer */
1098         if (left) {
1099                 orig_slot += btrfs_header_nritems(left);
1100                 wret = push_node_left(trans, root, left, mid, 1);
1101                 if (wret < 0)
1102                         ret = wret;
1103                 if (btrfs_header_nritems(mid) < 2)
1104                         err_on_enospc = 1;
1105         }
1106
1107         /*
1108          * then try to empty the right most buffer into the middle
1109          */
1110         if (right) {
1111                 wret = push_node_left(trans, root, mid, right, 1);
1112                 if (wret < 0 && wret != -ENOSPC)
1113                         ret = wret;
1114                 if (btrfs_header_nritems(right) == 0) {
1115                         clean_tree_block(trans, root, right);
1116                         btrfs_tree_unlock(right);
1117                         wret = del_ptr(trans, root, path, level + 1, pslot +
1118                                        1);
1119                         if (wret)
1120                                 ret = wret;
1121                         root_sub_used(root, right->len);
1122                         btrfs_free_tree_block(trans, root, right, 0, 1);
1123                         free_extent_buffer(right);
1124                         right = NULL;
1125                 } else {
1126                         struct btrfs_disk_key right_key;
1127                         btrfs_node_key(right, &right_key, 0);
1128                         btrfs_set_node_key(parent, &right_key, pslot + 1);
1129                         btrfs_mark_buffer_dirty(parent);
1130                 }
1131         }
1132         if (btrfs_header_nritems(mid) == 1) {
1133                 /*
1134                  * we're not allowed to leave a node with one item in the
1135                  * tree during a delete.  A deletion from lower in the tree
1136                  * could try to delete the only pointer in this node.
1137                  * So, pull some keys from the left.
1138                  * There has to be a left pointer at this point because
1139                  * otherwise we would have pulled some pointers from the
1140                  * right
1141                  */
1142                 BUG_ON(!left);
1143                 wret = balance_node_right(trans, root, mid, left);
1144                 if (wret < 0) {
1145                         ret = wret;
1146                         goto enospc;
1147                 }
1148                 if (wret == 1) {
1149                         wret = push_node_left(trans, root, left, mid, 1);
1150                         if (wret < 0)
1151                                 ret = wret;
1152                 }
1153                 BUG_ON(wret == 1);
1154         }
1155         if (btrfs_header_nritems(mid) == 0) {
1156                 clean_tree_block(trans, root, mid);
1157                 btrfs_tree_unlock(mid);
1158                 wret = del_ptr(trans, root, path, level + 1, pslot);
1159                 if (wret)
1160                         ret = wret;
1161                 root_sub_used(root, mid->len);
1162                 btrfs_free_tree_block(trans, root, mid, 0, 1);
1163                 free_extent_buffer(mid);
1164                 mid = NULL;
1165         } else {
1166                 /* update the parent key to reflect our changes */
1167                 struct btrfs_disk_key mid_key;
1168                 btrfs_node_key(mid, &mid_key, 0);
1169                 btrfs_set_node_key(parent, &mid_key, pslot);
1170                 btrfs_mark_buffer_dirty(parent);
1171         }
1172
1173         /* update the path */
1174         if (left) {
1175                 if (btrfs_header_nritems(left) > orig_slot) {
1176                         extent_buffer_get(left);
1177                         /* left was locked after cow */
1178                         path->nodes[level] = left;
1179                         path->slots[level + 1] -= 1;
1180                         path->slots[level] = orig_slot;
1181                         if (mid) {
1182                                 btrfs_tree_unlock(mid);
1183                                 free_extent_buffer(mid);
1184                         }
1185                 } else {
1186                         orig_slot -= btrfs_header_nritems(left);
1187                         path->slots[level] = orig_slot;
1188                 }
1189         }
1190         /* double check we haven't messed things up */
1191         check_block(root, path, level);
1192         if (orig_ptr !=
1193             btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1194                 BUG();
1195 enospc:
1196         if (right) {
1197                 btrfs_tree_unlock(right);
1198                 free_extent_buffer(right);
1199         }
1200         if (left) {
1201                 if (path->nodes[level] != left)
1202                         btrfs_tree_unlock(left);
1203                 free_extent_buffer(left);
1204         }
1205         return ret;
1206 }
1207
1208 /* Node balancing for insertion.  Here we only split or push nodes around
1209  * when they are completely full.  This is also done top down, so we
1210  * have to be pessimistic.
1211  */
1212 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1213                                           struct btrfs_root *root,
1214                                           struct btrfs_path *path, int level)
1215 {
1216         struct extent_buffer *right = NULL;
1217         struct extent_buffer *mid;
1218         struct extent_buffer *left = NULL;
1219         struct extent_buffer *parent = NULL;
1220         int ret = 0;
1221         int wret;
1222         int pslot;
1223         int orig_slot = path->slots[level];
1224         u64 orig_ptr;
1225
1226         if (level == 0)
1227                 return 1;
1228
1229         mid = path->nodes[level];
1230         WARN_ON(btrfs_header_generation(mid) != trans->transid);
1231         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1232
1233         if (level < BTRFS_MAX_LEVEL - 1)
1234                 parent = path->nodes[level + 1];
1235         pslot = path->slots[level + 1];
1236
1237         if (!parent)
1238                 return 1;
1239
1240         left = read_node_slot(root, parent, pslot - 1);
1241
1242         /* first, try to make some room in the middle buffer */
1243         if (left) {
1244                 u32 left_nr;
1245
1246                 btrfs_tree_lock(left);
1247                 btrfs_set_lock_blocking(left);
1248
1249                 left_nr = btrfs_header_nritems(left);
1250                 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1251                         wret = 1;
1252                 } else {
1253                         ret = btrfs_cow_block(trans, root, left, parent,
1254                                               pslot - 1, &left);
1255                         if (ret)
1256                                 wret = 1;
1257                         else {
1258                                 wret = push_node_left(trans, root,
1259                                                       left, mid, 0);
1260                         }
1261                 }
1262                 if (wret < 0)
1263                         ret = wret;
1264                 if (wret == 0) {
1265                         struct btrfs_disk_key disk_key;
1266                         orig_slot += left_nr;
1267                         btrfs_node_key(mid, &disk_key, 0);
1268                         btrfs_set_node_key(parent, &disk_key, pslot);
1269                         btrfs_mark_buffer_dirty(parent);
1270                         if (btrfs_header_nritems(left) > orig_slot) {
1271                                 path->nodes[level] = left;
1272                                 path->slots[level + 1] -= 1;
1273                                 path->slots[level] = orig_slot;
1274                                 btrfs_tree_unlock(mid);
1275                                 free_extent_buffer(mid);
1276                         } else {
1277                                 orig_slot -=
1278                                         btrfs_header_nritems(left);
1279                                 path->slots[level] = orig_slot;
1280                                 btrfs_tree_unlock(left);
1281                                 free_extent_buffer(left);
1282                         }
1283                         return 0;
1284                 }
1285                 btrfs_tree_unlock(left);
1286                 free_extent_buffer(left);
1287         }
1288         right = read_node_slot(root, parent, pslot + 1);
1289
1290         /*
1291          * then try to empty the right most buffer into the middle
1292          */
1293         if (right) {
1294                 u32 right_nr;
1295
1296                 btrfs_tree_lock(right);
1297                 btrfs_set_lock_blocking(right);
1298
1299                 right_nr = btrfs_header_nritems(right);
1300                 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1301                         wret = 1;
1302                 } else {
1303                         ret = btrfs_cow_block(trans, root, right,
1304                                               parent, pslot + 1,
1305                                               &right);
1306                         if (ret)
1307                                 wret = 1;
1308                         else {
1309                                 wret = balance_node_right(trans, root,
1310                                                           right, mid);
1311                         }
1312                 }
1313                 if (wret < 0)
1314                         ret = wret;
1315                 if (wret == 0) {
1316                         struct btrfs_disk_key disk_key;
1317
1318                         btrfs_node_key(right, &disk_key, 0);
1319                         btrfs_set_node_key(parent, &disk_key, pslot + 1);
1320                         btrfs_mark_buffer_dirty(parent);
1321
1322                         if (btrfs_header_nritems(mid) <= orig_slot) {
1323                                 path->nodes[level] = right;
1324                                 path->slots[level + 1] += 1;
1325                                 path->slots[level] = orig_slot -
1326                                         btrfs_header_nritems(mid);
1327                                 btrfs_tree_unlock(mid);
1328                                 free_extent_buffer(mid);
1329                         } else {
1330                                 btrfs_tree_unlock(right);
1331                                 free_extent_buffer(right);
1332                         }
1333                         return 0;
1334                 }
1335                 btrfs_tree_unlock(right);
1336                 free_extent_buffer(right);
1337         }
1338         return 1;
1339 }
1340
1341 /*
1342  * readahead one full node of leaves, finding things that are close
1343  * to the block in 'slot', and triggering ra on them.
1344  */
1345 static void reada_for_search(struct btrfs_root *root,
1346                              struct btrfs_path *path,
1347                              int level, int slot, u64 objectid)
1348 {
1349         struct extent_buffer *node;
1350         struct btrfs_disk_key disk_key;
1351         u32 nritems;
1352         u64 search;
1353         u64 target;
1354         u64 nread = 0;
1355         int direction = path->reada;
1356         struct extent_buffer *eb;
1357         u32 nr;
1358         u32 blocksize;
1359         u32 nscan = 0;
1360
1361         if (level != 1)
1362                 return;
1363
1364         if (!path->nodes[level])
1365                 return;
1366
1367         node = path->nodes[level];
1368
1369         search = btrfs_node_blockptr(node, slot);
1370         blocksize = btrfs_level_size(root, level - 1);
1371         eb = btrfs_find_tree_block(root, search, blocksize);
1372         if (eb) {
1373                 free_extent_buffer(eb);
1374                 return;
1375         }
1376
1377         target = search;
1378
1379         nritems = btrfs_header_nritems(node);
1380         nr = slot;
1381         while (1) {
1382                 if (direction < 0) {
1383                         if (nr == 0)
1384                                 break;
1385                         nr--;
1386                 } else if (direction > 0) {
1387                         nr++;
1388                         if (nr >= nritems)
1389                                 break;
1390                 }
1391                 if (path->reada < 0 && objectid) {
1392                         btrfs_node_key(node, &disk_key, nr);
1393                         if (btrfs_disk_key_objectid(&disk_key) != objectid)
1394                                 break;
1395                 }
1396                 search = btrfs_node_blockptr(node, nr);
1397                 if ((search <= target && target - search <= 65536) ||
1398                     (search > target && search - target <= 65536)) {
1399                         readahead_tree_block(root, search, blocksize,
1400                                      btrfs_node_ptr_generation(node, nr));
1401                         nread += blocksize;
1402                 }
1403                 nscan++;
1404                 if ((nread > 65536 || nscan > 32))
1405                         break;
1406         }
1407 }
1408
1409 /*
1410  * returns -EAGAIN if it had to drop the path, or zero if everything was in
1411  * cache
1412  */
1413 static noinline int reada_for_balance(struct btrfs_root *root,
1414                                       struct btrfs_path *path, int level)
1415 {
1416         int slot;
1417         int nritems;
1418         struct extent_buffer *parent;
1419         struct extent_buffer *eb;
1420         u64 gen;
1421         u64 block1 = 0;
1422         u64 block2 = 0;
1423         int ret = 0;
1424         int blocksize;
1425
1426         parent = path->nodes[level + 1];
1427         if (!parent)
1428                 return 0;
1429
1430         nritems = btrfs_header_nritems(parent);
1431         slot = path->slots[level + 1];
1432         blocksize = btrfs_level_size(root, level);
1433
1434         if (slot > 0) {
1435                 block1 = btrfs_node_blockptr(parent, slot - 1);
1436                 gen = btrfs_node_ptr_generation(parent, slot - 1);
1437                 eb = btrfs_find_tree_block(root, block1, blocksize);
1438                 if (eb && btrfs_buffer_uptodate(eb, gen))
1439                         block1 = 0;
1440                 free_extent_buffer(eb);
1441         }
1442         if (slot + 1 < nritems) {
1443                 block2 = btrfs_node_blockptr(parent, slot + 1);
1444                 gen = btrfs_node_ptr_generation(parent, slot + 1);
1445                 eb = btrfs_find_tree_block(root, block2, blocksize);
1446                 if (eb && btrfs_buffer_uptodate(eb, gen))
1447                         block2 = 0;
1448                 free_extent_buffer(eb);
1449         }
1450         if (block1 || block2) {
1451                 ret = -EAGAIN;
1452
1453                 /* release the whole path */
1454                 btrfs_release_path(root, path);
1455
1456                 /* read the blocks */
1457                 if (block1)
1458                         readahead_tree_block(root, block1, blocksize, 0);
1459                 if (block2)
1460                         readahead_tree_block(root, block2, blocksize, 0);
1461
1462                 if (block1) {
1463                         eb = read_tree_block(root, block1, blocksize, 0);
1464                         free_extent_buffer(eb);
1465                 }
1466                 if (block2) {
1467                         eb = read_tree_block(root, block2, blocksize, 0);
1468                         free_extent_buffer(eb);
1469                 }
1470         }
1471         return ret;
1472 }
1473
1474
1475 /*
1476  * when we walk down the tree, it is usually safe to unlock the higher layers
1477  * in the tree.  The exceptions are when our path goes through slot 0, because
1478  * operations on the tree might require changing key pointers higher up in the
1479  * tree.
1480  *
1481  * callers might also have set path->keep_locks, which tells this code to keep
1482  * the lock if the path points to the last slot in the block.  This is part of
1483  * walking through the tree, and selecting the next slot in the higher block.
1484  *
1485  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
1486  * if lowest_unlock is 1, level 0 won't be unlocked
1487  */
1488 static noinline void unlock_up(struct btrfs_path *path, int level,
1489                                int lowest_unlock)
1490 {
1491         int i;
1492         int skip_level = level;
1493         int no_skips = 0;
1494         struct extent_buffer *t;
1495
1496         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1497                 if (!path->nodes[i])
1498                         break;
1499                 if (!path->locks[i])
1500                         break;
1501                 if (!no_skips && path->slots[i] == 0) {
1502                         skip_level = i + 1;
1503                         continue;
1504                 }
1505                 if (!no_skips && path->keep_locks) {
1506                         u32 nritems;
1507                         t = path->nodes[i];
1508                         nritems = btrfs_header_nritems(t);
1509                         if (nritems < 1 || path->slots[i] >= nritems - 1) {
1510                                 skip_level = i + 1;
1511                                 continue;
1512                         }
1513                 }
1514                 if (skip_level < i && i >= lowest_unlock)
1515                         no_skips = 1;
1516
1517                 t = path->nodes[i];
1518                 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1519                         btrfs_tree_unlock(t);
1520                         path->locks[i] = 0;
1521                 }
1522         }
1523 }
1524
1525 /*
1526  * This releases any locks held in the path starting at level and
1527  * going all the way up to the root.
1528  *
1529  * btrfs_search_slot will keep the lock held on higher nodes in a few
1530  * corner cases, such as COW of the block at slot zero in the node.  This
1531  * ignores those rules, and it should only be called when there are no
1532  * more updates to be done higher up in the tree.
1533  */
1534 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1535 {
1536         int i;
1537
1538         if (path->keep_locks)
1539                 return;
1540
1541         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1542                 if (!path->nodes[i])
1543                         continue;
1544                 if (!path->locks[i])
1545                         continue;
1546                 btrfs_tree_unlock(path->nodes[i]);
1547                 path->locks[i] = 0;
1548         }
1549 }
1550
1551 /*
1552  * helper function for btrfs_search_slot.  The goal is to find a block
1553  * in cache without setting the path to blocking.  If we find the block
1554  * we return zero and the path is unchanged.
1555  *
1556  * If we can't find the block, we set the path blocking and do some
1557  * reada.  -EAGAIN is returned and the search must be repeated.
1558  */
1559 static int
1560 read_block_for_search(struct btrfs_trans_handle *trans,
1561                        struct btrfs_root *root, struct btrfs_path *p,
1562                        struct extent_buffer **eb_ret, int level, int slot,
1563                        struct btrfs_key *key)
1564 {
1565         u64 blocknr;
1566         u64 gen;
1567         u32 blocksize;
1568         struct extent_buffer *b = *eb_ret;
1569         struct extent_buffer *tmp;
1570         int ret;
1571
1572         blocknr = btrfs_node_blockptr(b, slot);
1573         gen = btrfs_node_ptr_generation(b, slot);
1574         blocksize = btrfs_level_size(root, level - 1);
1575
1576         tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1577         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1578                 /*
1579                  * we found an up to date block without sleeping, return
1580                  * right away
1581                  */
1582                 *eb_ret = tmp;
1583                 return 0;
1584         }
1585
1586         /*
1587          * reduce lock contention at high levels
1588          * of the btree by dropping locks before
1589          * we read.  Don't release the lock on the current
1590          * level because we need to walk this node to figure
1591          * out which blocks to read.
1592          */
1593         btrfs_unlock_up_safe(p, level + 1);
1594         btrfs_set_path_blocking(p);
1595
1596         if (tmp)
1597                 free_extent_buffer(tmp);
1598         if (p->reada)
1599                 reada_for_search(root, p, level, slot, key->objectid);
1600
1601         btrfs_release_path(NULL, p);
1602
1603         ret = -EAGAIN;
1604         tmp = read_tree_block(root, blocknr, blocksize, gen);
1605         if (tmp) {
1606                 /*
1607                  * If the read above didn't mark this buffer up to date,
1608                  * it will never end up being up to date.  Set ret to EIO now
1609                  * and give up so that our caller doesn't loop forever
1610                  * on our EAGAINs.
1611                  */
1612                 if (!btrfs_buffer_uptodate(tmp, 0))
1613                         ret = -EIO;
1614                 free_extent_buffer(tmp);
1615         }
1616         return ret;
1617 }
1618
1619 /*
1620  * helper function for btrfs_search_slot.  This does all of the checks
1621  * for node-level blocks and does any balancing required based on
1622  * the ins_len.
1623  *
1624  * If no extra work was required, zero is returned.  If we had to
1625  * drop the path, -EAGAIN is returned and btrfs_search_slot must
1626  * start over
1627  */
1628 static int
1629 setup_nodes_for_search(struct btrfs_trans_handle *trans,
1630                        struct btrfs_root *root, struct btrfs_path *p,
1631                        struct extent_buffer *b, int level, int ins_len)
1632 {
1633         int ret;
1634         if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1635             BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1636                 int sret;
1637
1638                 sret = reada_for_balance(root, p, level);
1639                 if (sret)
1640                         goto again;
1641
1642                 btrfs_set_path_blocking(p);
1643                 sret = split_node(trans, root, p, level);
1644                 btrfs_clear_path_blocking(p, NULL);
1645
1646                 BUG_ON(sret > 0);
1647                 if (sret) {
1648                         ret = sret;
1649                         goto done;
1650                 }
1651                 b = p->nodes[level];
1652         } else if (ins_len < 0 && btrfs_header_nritems(b) <
1653                    BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
1654                 int sret;
1655
1656                 sret = reada_for_balance(root, p, level);
1657                 if (sret)
1658                         goto again;
1659
1660                 btrfs_set_path_blocking(p);
1661                 sret = balance_level(trans, root, p, level);
1662                 btrfs_clear_path_blocking(p, NULL);
1663
1664                 if (sret) {
1665                         ret = sret;
1666                         goto done;
1667                 }
1668                 b = p->nodes[level];
1669                 if (!b) {
1670                         btrfs_release_path(NULL, p);
1671                         goto again;
1672                 }
1673                 BUG_ON(btrfs_header_nritems(b) == 1);
1674         }
1675         return 0;
1676
1677 again:
1678         ret = -EAGAIN;
1679 done:
1680         return ret;
1681 }
1682
1683 /*
1684  * look for key in the tree.  path is filled in with nodes along the way
1685  * if key is found, we return zero and you can find the item in the leaf
1686  * level of the path (level 0)
1687  *
1688  * If the key isn't found, the path points to the slot where it should
1689  * be inserted, and 1 is returned.  If there are other errors during the
1690  * search a negative error number is returned.
1691  *
1692  * if ins_len > 0, nodes and leaves will be split as we walk down the
1693  * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
1694  * possible)
1695  */
1696 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1697                       *root, struct btrfs_key *key, struct btrfs_path *p, int
1698                       ins_len, int cow)
1699 {
1700         struct extent_buffer *b;
1701         int slot;
1702         int ret;
1703         int err;
1704         int level;
1705         int lowest_unlock = 1;
1706         u8 lowest_level = 0;
1707
1708         lowest_level = p->lowest_level;
1709         WARN_ON(lowest_level && ins_len > 0);
1710         WARN_ON(p->nodes[0] != NULL);
1711
1712         if (ins_len < 0)
1713                 lowest_unlock = 2;
1714
1715 again:
1716         if (p->search_commit_root) {
1717                 b = root->commit_root;
1718                 extent_buffer_get(b);
1719                 if (!p->skip_locking)
1720                         btrfs_tree_lock(b);
1721         } else {
1722                 if (p->skip_locking)
1723                         b = btrfs_root_node(root);
1724                 else
1725                         b = btrfs_lock_root_node(root);
1726         }
1727
1728         while (b) {
1729                 level = btrfs_header_level(b);
1730
1731                 /*
1732                  * setup the path here so we can release it under lock
1733                  * contention with the cow code
1734                  */
1735                 p->nodes[level] = b;
1736                 if (!p->skip_locking)
1737                         p->locks[level] = 1;
1738
1739                 if (cow) {
1740                         /*
1741                          * if we don't really need to cow this block
1742                          * then we don't want to set the path blocking,
1743                          * so we test it here
1744                          */
1745                         if (!should_cow_block(trans, root, b))
1746                                 goto cow_done;
1747
1748                         btrfs_set_path_blocking(p);
1749
1750                         err = btrfs_cow_block(trans, root, b,
1751                                               p->nodes[level + 1],
1752                                               p->slots[level + 1], &b);
1753                         if (err) {
1754                                 ret = err;
1755                                 goto done;
1756                         }
1757                 }
1758 cow_done:
1759                 BUG_ON(!cow && ins_len);
1760                 if (level != btrfs_header_level(b))
1761                         WARN_ON(1);
1762                 level = btrfs_header_level(b);
1763
1764                 p->nodes[level] = b;
1765                 if (!p->skip_locking)
1766                         p->locks[level] = 1;
1767
1768                 btrfs_clear_path_blocking(p, NULL);
1769
1770                 /*
1771                  * we have a lock on b and as long as we aren't changing
1772                  * the tree, there is no way to for the items in b to change.
1773                  * It is safe to drop the lock on our parent before we
1774                  * go through the expensive btree search on b.
1775                  *
1776                  * If cow is true, then we might be changing slot zero,
1777                  * which may require changing the parent.  So, we can't
1778                  * drop the lock until after we know which slot we're
1779                  * operating on.
1780                  */
1781                 if (!cow)
1782                         btrfs_unlock_up_safe(p, level + 1);
1783
1784                 ret = check_block(root, p, level);
1785                 if (ret) {
1786                         ret = -1;
1787                         goto done;
1788                 }
1789
1790                 ret = bin_search(b, key, level, &slot);
1791
1792                 if (level != 0) {
1793                         int dec = 0;
1794                         if (ret && slot > 0) {
1795                                 dec = 1;
1796                                 slot -= 1;
1797                         }
1798                         p->slots[level] = slot;
1799                         err = setup_nodes_for_search(trans, root, p, b, level,
1800                                                      ins_len);
1801                         if (err == -EAGAIN)
1802                                 goto again;
1803                         if (err) {
1804                                 ret = err;
1805                                 goto done;
1806                         }
1807                         b = p->nodes[level];
1808                         slot = p->slots[level];
1809
1810                         unlock_up(p, level, lowest_unlock);
1811
1812                         if (level == lowest_level) {
1813                                 if (dec)
1814                                         p->slots[level]++;
1815                                 goto done;
1816                         }
1817
1818                         err = read_block_for_search(trans, root, p,
1819                                                     &b, level, slot, key);
1820                         if (err == -EAGAIN)
1821                                 goto again;
1822                         if (err) {
1823                                 ret = err;
1824                                 goto done;
1825                         }
1826
1827                         if (!p->skip_locking) {
1828                                 btrfs_clear_path_blocking(p, NULL);
1829                                 err = btrfs_try_spin_lock(b);
1830
1831                                 if (!err) {
1832                                         btrfs_set_path_blocking(p);
1833                                         btrfs_tree_lock(b);
1834                                         btrfs_clear_path_blocking(p, b);
1835                                 }
1836                         }
1837                 } else {
1838                         p->slots[level] = slot;
1839                         if (ins_len > 0 &&
1840                             btrfs_leaf_free_space(root, b) < ins_len) {
1841                                 btrfs_set_path_blocking(p);
1842                                 err = split_leaf(trans, root, key,
1843                                                  p, ins_len, ret == 0);
1844                                 btrfs_clear_path_blocking(p, NULL);
1845
1846                                 BUG_ON(err > 0);
1847                                 if (err) {
1848                                         ret = err;
1849                                         goto done;
1850                                 }
1851                         }
1852                         if (!p->search_for_split)
1853                                 unlock_up(p, level, lowest_unlock);
1854                         goto done;
1855                 }
1856         }
1857         ret = 1;
1858 done:
1859         /*
1860          * we don't really know what they plan on doing with the path
1861          * from here on, so for now just mark it as blocking
1862          */
1863         if (!p->leave_spinning)
1864                 btrfs_set_path_blocking(p);
1865         if (ret < 0)
1866                 btrfs_release_path(root, p);
1867         return ret;
1868 }
1869
1870 /*
1871  * adjust the pointers going up the tree, starting at level
1872  * making sure the right key of each node is points to 'key'.
1873  * This is used after shifting pointers to the left, so it stops
1874  * fixing up pointers when a given leaf/node is not in slot 0 of the
1875  * higher levels
1876  *
1877  * If this fails to write a tree block, it returns -1, but continues
1878  * fixing up the blocks in ram so the tree is consistent.
1879  */
1880 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1881                           struct btrfs_root *root, struct btrfs_path *path,
1882                           struct btrfs_disk_key *key, int level)
1883 {
1884         int i;
1885         int ret = 0;
1886         struct extent_buffer *t;
1887
1888         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1889                 int tslot = path->slots[i];
1890                 if (!path->nodes[i])
1891                         break;
1892                 t = path->nodes[i];
1893                 btrfs_set_node_key(t, key, tslot);
1894                 btrfs_mark_buffer_dirty(path->nodes[i]);
1895                 if (tslot != 0)
1896                         break;
1897         }
1898         return ret;
1899 }
1900
1901 /*
1902  * update item key.
1903  *
1904  * This function isn't completely safe. It's the caller's responsibility
1905  * that the new key won't break the order
1906  */
1907 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1908                             struct btrfs_root *root, struct btrfs_path *path,
1909                             struct btrfs_key *new_key)
1910 {
1911         struct btrfs_disk_key disk_key;
1912         struct extent_buffer *eb;
1913         int slot;
1914
1915         eb = path->nodes[0];
1916         slot = path->slots[0];
1917         if (slot > 0) {
1918                 btrfs_item_key(eb, &disk_key, slot - 1);
1919                 if (comp_keys(&disk_key, new_key) >= 0)
1920                         return -1;
1921         }
1922         if (slot < btrfs_header_nritems(eb) - 1) {
1923                 btrfs_item_key(eb, &disk_key, slot + 1);
1924                 if (comp_keys(&disk_key, new_key) <= 0)
1925                         return -1;
1926         }
1927
1928         btrfs_cpu_key_to_disk(&disk_key, new_key);
1929         btrfs_set_item_key(eb, &disk_key, slot);
1930         btrfs_mark_buffer_dirty(eb);
1931         if (slot == 0)
1932                 fixup_low_keys(trans, root, path, &disk_key, 1);
1933         return 0;
1934 }
1935
1936 /*
1937  * try to push data from one node into the next node left in the
1938  * tree.
1939  *
1940  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1941  * error, and > 0 if there was no room in the left hand block.
1942  */
1943 static int push_node_left(struct btrfs_trans_handle *trans,
1944                           struct btrfs_root *root, struct extent_buffer *dst,
1945                           struct extent_buffer *src, int empty)
1946 {
1947         int push_items = 0;
1948         int src_nritems;
1949         int dst_nritems;
1950         int ret = 0;
1951
1952         src_nritems = btrfs_header_nritems(src);
1953         dst_nritems = btrfs_header_nritems(dst);
1954         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1955         WARN_ON(btrfs_header_generation(src) != trans->transid);
1956         WARN_ON(btrfs_header_generation(dst) != trans->transid);
1957
1958         if (!empty && src_nritems <= 8)
1959                 return 1;
1960
1961         if (push_items <= 0)
1962                 return 1;
1963
1964         if (empty) {
1965                 push_items = min(src_nritems, push_items);
1966                 if (push_items < src_nritems) {
1967                         /* leave at least 8 pointers in the node if
1968                          * we aren't going to empty it
1969                          */
1970                         if (src_nritems - push_items < 8) {
1971                                 if (push_items <= 8)
1972                                         return 1;
1973                                 push_items -= 8;
1974                         }
1975                 }
1976         } else
1977                 push_items = min(src_nritems - 8, push_items);
1978
1979         copy_extent_buffer(dst, src,
1980                            btrfs_node_key_ptr_offset(dst_nritems),
1981                            btrfs_node_key_ptr_offset(0),
1982                            push_items * sizeof(struct btrfs_key_ptr));
1983
1984         if (push_items < src_nritems) {
1985                 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1986                                       btrfs_node_key_ptr_offset(push_items),
1987                                       (src_nritems - push_items) *
1988                                       sizeof(struct btrfs_key_ptr));
1989         }
1990         btrfs_set_header_nritems(src, src_nritems - push_items);
1991         btrfs_set_header_nritems(dst, dst_nritems + push_items);
1992         btrfs_mark_buffer_dirty(src);
1993         btrfs_mark_buffer_dirty(dst);
1994
1995         return ret;
1996 }
1997
1998 /*
1999  * try to push data from one node into the next node right in the
2000  * tree.
2001  *
2002  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
2003  * error, and > 0 if there was no room in the right hand block.
2004  *
2005  * this will  only push up to 1/2 the contents of the left node over
2006  */
2007 static int balance_node_right(struct btrfs_trans_handle *trans,
2008                               struct btrfs_root *root,
2009                               struct extent_buffer *dst,
2010                               struct extent_buffer *src)
2011 {
2012         int push_items = 0;
2013         int max_push;
2014         int src_nritems;
2015         int dst_nritems;
2016         int ret = 0;
2017
2018         WARN_ON(btrfs_header_generation(src) != trans->transid);
2019         WARN_ON(btrfs_header_generation(dst) != trans->transid);
2020
2021         src_nritems = btrfs_header_nritems(src);
2022         dst_nritems = btrfs_header_nritems(dst);
2023         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2024         if (push_items <= 0)
2025                 return 1;
2026
2027         if (src_nritems < 4)
2028                 return 1;
2029
2030         max_push = src_nritems / 2 + 1;
2031         /* don't try to empty the node */
2032         if (max_push >= src_nritems)
2033                 return 1;
2034
2035         if (max_push < push_items)
2036                 push_items = max_push;
2037
2038         memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2039                                       btrfs_node_key_ptr_offset(0),
2040                                       (dst_nritems) *
2041                                       sizeof(struct btrfs_key_ptr));
2042
2043         copy_extent_buffer(dst, src,
2044                            btrfs_node_key_ptr_offset(0),
2045                            btrfs_node_key_ptr_offset(src_nritems - push_items),
2046                            push_items * sizeof(struct btrfs_key_ptr));
2047
2048         btrfs_set_header_nritems(src, src_nritems - push_items);
2049         btrfs_set_header_nritems(dst, dst_nritems + push_items);
2050
2051         btrfs_mark_buffer_dirty(src);
2052         btrfs_mark_buffer_dirty(dst);
2053
2054         return ret;
2055 }
2056
2057 /*
2058  * helper function to insert a new root level in the tree.
2059  * A new node is allocated, and a single item is inserted to
2060  * point to the existing root
2061  *
2062  * returns zero on success or < 0 on failure.
2063  */
2064 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2065                            struct btrfs_root *root,
2066                            struct btrfs_path *path, int level)
2067 {
2068         u64 lower_gen;
2069         struct extent_buffer *lower;
2070         struct extent_buffer *c;
2071         struct extent_buffer *old;
2072         struct btrfs_disk_key lower_key;
2073
2074         BUG_ON(path->nodes[level]);
2075         BUG_ON(path->nodes[level-1] != root->node);
2076
2077         lower = path->nodes[level-1];
2078         if (level == 1)
2079                 btrfs_item_key(lower, &lower_key, 0);
2080         else
2081                 btrfs_node_key(lower, &lower_key, 0);
2082
2083         c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2084                                    root->root_key.objectid, &lower_key,
2085                                    level, root->node->start, 0);
2086         if (IS_ERR(c))
2087                 return PTR_ERR(c);
2088
2089         root_add_used(root, root->nodesize);
2090
2091         memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
2092         btrfs_set_header_nritems(c, 1);
2093         btrfs_set_header_level(c, level);
2094         btrfs_set_header_bytenr(c, c->start);
2095         btrfs_set_header_generation(c, trans->transid);
2096         btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
2097         btrfs_set_header_owner(c, root->root_key.objectid);
2098
2099         write_extent_buffer(c, root->fs_info->fsid,
2100                             (unsigned long)btrfs_header_fsid(c),
2101                             BTRFS_FSID_SIZE);
2102
2103         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2104                             (unsigned long)btrfs_header_chunk_tree_uuid(c),
2105                             BTRFS_UUID_SIZE);
2106
2107         btrfs_set_node_key(c, &lower_key, 0);
2108         btrfs_set_node_blockptr(c, 0, lower->start);
2109         lower_gen = btrfs_header_generation(lower);
2110         WARN_ON(lower_gen != trans->transid);
2111
2112         btrfs_set_node_ptr_generation(c, 0, lower_gen);
2113
2114         btrfs_mark_buffer_dirty(c);
2115
2116         spin_lock(&root->node_lock);
2117         old = root->node;
2118         root->node = c;
2119         spin_unlock(&root->node_lock);
2120
2121         /* the super has an extra ref to root->node */
2122         free_extent_buffer(old);
2123
2124         add_root_to_dirty_list(root);
2125         extent_buffer_get(c);
2126         path->nodes[level] = c;
2127         path->locks[level] = 1;
2128         path->slots[level] = 0;
2129         return 0;
2130 }
2131
2132 /*
2133  * worker function to insert a single pointer in a node.
2134  * the node should have enough room for the pointer already
2135  *
2136  * slot and level indicate where you want the key to go, and
2137  * blocknr is the block the key points to.
2138  *
2139  * returns zero on success and < 0 on any error
2140  */
2141 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2142                       *root, struct btrfs_path *path, struct btrfs_disk_key
2143                       *key, u64 bytenr, int slot, int level)
2144 {
2145         struct extent_buffer *lower;
2146         int nritems;
2147
2148         BUG_ON(!path->nodes[level]);
2149         btrfs_assert_tree_locked(path->nodes[level]);
2150         lower = path->nodes[level];
2151         nritems = btrfs_header_nritems(lower);
2152         BUG_ON(slot > nritems);
2153         if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2154                 BUG();
2155         if (slot != nritems) {
2156                 memmove_extent_buffer(lower,
2157                               btrfs_node_key_ptr_offset(slot + 1),
2158                               btrfs_node_key_ptr_offset(slot),
2159                               (nritems - slot) * sizeof(struct btrfs_key_ptr));
2160         }
2161         btrfs_set_node_key(lower, key, slot);
2162         btrfs_set_node_blockptr(lower, slot, bytenr);
2163         WARN_ON(trans->transid == 0);
2164         btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2165         btrfs_set_header_nritems(lower, nritems + 1);
2166         btrfs_mark_buffer_dirty(lower);
2167         return 0;
2168 }
2169
2170 /*
2171  * split the node at the specified level in path in two.
2172  * The path is corrected to point to the appropriate node after the split
2173  *
2174  * Before splitting this tries to make some room in the node by pushing
2175  * left and right, if either one works, it returns right away.
2176  *
2177  * returns 0 on success and < 0 on failure
2178  */
2179 static noinline int split_node(struct btrfs_trans_handle *trans,
2180                                struct btrfs_root *root,
2181                                struct btrfs_path *path, int level)
2182 {
2183         struct extent_buffer *c;
2184         struct extent_buffer *split;
2185         struct btrfs_disk_key disk_key;
2186         int mid;
2187         int ret;
2188         int wret;
2189         u32 c_nritems;
2190
2191         c = path->nodes[level];
2192         WARN_ON(btrfs_header_generation(c) != trans->transid);
2193         if (c == root->node) {
2194                 /* trying to split the root, lets make a new one */
2195                 ret = insert_new_root(trans, root, path, level + 1);
2196                 if (ret)
2197                         return ret;
2198         } else {
2199                 ret = push_nodes_for_insert(trans, root, path, level);
2200                 c = path->nodes[level];
2201                 if (!ret && btrfs_header_nritems(c) <
2202                     BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2203                         return 0;
2204                 if (ret < 0)
2205                         return ret;
2206         }
2207
2208         c_nritems = btrfs_header_nritems(c);
2209         mid = (c_nritems + 1) / 2;
2210         btrfs_node_key(c, &disk_key, mid);
2211
2212         split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2213                                         root->root_key.objectid,
2214                                         &disk_key, level, c->start, 0);
2215         if (IS_ERR(split))
2216                 return PTR_ERR(split);
2217
2218         root_add_used(root, root->nodesize);
2219
2220         memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
2221         btrfs_set_header_level(split, btrfs_header_level(c));
2222         btrfs_set_header_bytenr(split, split->start);
2223         btrfs_set_header_generation(split, trans->transid);
2224         btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
2225         btrfs_set_header_owner(split, root->root_key.objectid);
2226         write_extent_buffer(split, root->fs_info->fsid,
2227                             (unsigned long)btrfs_header_fsid(split),
2228                             BTRFS_FSID_SIZE);
2229         write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2230                             (unsigned long)btrfs_header_chunk_tree_uuid(split),
2231                             BTRFS_UUID_SIZE);
2232
2233
2234         copy_extent_buffer(split, c,
2235                            btrfs_node_key_ptr_offset(0),
2236                            btrfs_node_key_ptr_offset(mid),
2237                            (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2238         btrfs_set_header_nritems(split, c_nritems - mid);
2239         btrfs_set_header_nritems(c, mid);
2240         ret = 0;
2241
2242         btrfs_mark_buffer_dirty(c);
2243         btrfs_mark_buffer_dirty(split);
2244
2245         wret = insert_ptr(trans, root, path, &disk_key, split->start,
2246                           path->slots[level + 1] + 1,
2247                           level + 1);
2248         if (wret)
2249                 ret = wret;
2250
2251         if (path->slots[level] >= mid) {
2252                 path->slots[level] -= mid;
2253                 btrfs_tree_unlock(c);
2254                 free_extent_buffer(c);
2255                 path->nodes[level] = split;
2256                 path->slots[level + 1] += 1;
2257         } else {
2258                 btrfs_tree_unlock(split);
2259                 free_extent_buffer(split);
2260         }
2261         return ret;
2262 }
2263
2264 /*
2265  * how many bytes are required to store the items in a leaf.  start
2266  * and nr indicate which items in the leaf to check.  This totals up the
2267  * space used both by the item structs and the item data
2268  */
2269 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2270 {
2271         int data_len;
2272         int nritems = btrfs_header_nritems(l);
2273         int end = min(nritems, start + nr) - 1;
2274
2275         if (!nr)
2276                 return 0;
2277         data_len = btrfs_item_end_nr(l, start);
2278         data_len = data_len - btrfs_item_offset_nr(l, end);
2279         data_len += sizeof(struct btrfs_item) * nr;
2280         WARN_ON(data_len < 0);
2281         return data_len;
2282 }
2283
2284 /*
2285  * The space between the end of the leaf items and
2286  * the start of the leaf data.  IOW, how much room
2287  * the leaf has left for both items and data
2288  */
2289 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2290                                    struct extent_buffer *leaf)
2291 {
2292         int nritems = btrfs_header_nritems(leaf);
2293         int ret;
2294         ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2295         if (ret < 0) {
2296                 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2297                        "used %d nritems %d\n",
2298                        ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2299                        leaf_space_used(leaf, 0, nritems), nritems);
2300         }
2301         return ret;
2302 }
2303
2304 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2305                                       struct btrfs_root *root,
2306                                       struct btrfs_path *path,
2307                                       int data_size, int empty,
2308                                       struct extent_buffer *right,
2309                                       int free_space, u32 left_nritems)
2310 {
2311         struct extent_buffer *left = path->nodes[0];
2312         struct extent_buffer *upper = path->nodes[1];
2313         struct btrfs_disk_key disk_key;
2314         int slot;
2315         u32 i;
2316         int push_space = 0;
2317         int push_items = 0;
2318         struct btrfs_item *item;
2319         u32 nr;
2320         u32 right_nritems;
2321         u32 data_end;
2322         u32 this_item_size;
2323
2324         if (empty)
2325                 nr = 0;
2326         else
2327                 nr = 1;
2328
2329         if (path->slots[0] >= left_nritems)
2330                 push_space += data_size;
2331
2332         slot = path->slots[1];
2333         i = left_nritems - 1;
2334         while (i >= nr) {
2335                 item = btrfs_item_nr(left, i);
2336
2337                 if (!empty && push_items > 0) {
2338                         if (path->slots[0] > i)
2339                                 break;
2340                         if (path->slots[0] == i) {
2341                                 int space = btrfs_leaf_free_space(root, left);
2342                                 if (space + push_space * 2 > free_space)
2343                                         break;
2344                         }
2345                 }
2346
2347                 if (path->slots[0] == i)
2348                         push_space += data_size;
2349
2350                 if (!left->map_token) {
2351                         map_extent_buffer(left, (unsigned long)item,
2352                                         sizeof(struct btrfs_item),
2353                                         &left->map_token, &left->kaddr,
2354                                         &left->map_start, &left->map_len,
2355                                         KM_USER1);
2356                 }
2357
2358                 this_item_size = btrfs_item_size(left, item);
2359                 if (this_item_size + sizeof(*item) + push_space > free_space)
2360                         break;
2361
2362                 push_items++;
2363                 push_space += this_item_size + sizeof(*item);
2364                 if (i == 0)
2365                         break;
2366                 i--;
2367         }
2368         if (left->map_token) {
2369                 unmap_extent_buffer(left, left->map_token, KM_USER1);
2370                 left->map_token = NULL;
2371         }
2372
2373         if (push_items == 0)
2374                 goto out_unlock;
2375
2376         if (!empty && push_items == left_nritems)
2377                 WARN_ON(1);
2378
2379         /* push left to right */
2380         right_nritems = btrfs_header_nritems(right);
2381
2382         push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2383         push_space -= leaf_data_end(root, left);
2384
2385         /* make room in the right data area */
2386         data_end = leaf_data_end(root, right);
2387         memmove_extent_buffer(right,
2388                               btrfs_leaf_data(right) + data_end - push_space,
2389                               btrfs_leaf_data(right) + data_end,
2390                               BTRFS_LEAF_DATA_SIZE(root) - data_end);
2391
2392         /* copy from the left data area */
2393         copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2394                      BTRFS_LEAF_DATA_SIZE(root) - push_space,
2395                      btrfs_leaf_data(left) + leaf_data_end(root, left),
2396                      push_space);
2397
2398         memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2399                               btrfs_item_nr_offset(0),
2400                               right_nritems * sizeof(struct btrfs_item));
2401
2402         /* copy the items from left to right */
2403         copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2404                    btrfs_item_nr_offset(left_nritems - push_items),
2405                    push_items * sizeof(struct btrfs_item));
2406
2407         /* update the item pointers */
2408         right_nritems += push_items;
2409         btrfs_set_header_nritems(right, right_nritems);
2410         push_space = BTRFS_LEAF_DATA_SIZE(root);
2411         for (i = 0; i < right_nritems; i++) {
2412                 item = btrfs_item_nr(right, i);
2413                 if (!right->map_token) {
2414                         map_extent_buffer(right, (unsigned long)item,
2415                                         sizeof(struct btrfs_item),
2416                                         &right->map_token, &right->kaddr,
2417                                         &right->map_start, &right->map_len,
2418                                         KM_USER1);
2419                 }
2420                 push_space -= btrfs_item_size(right, item);
2421                 btrfs_set_item_offset(right, item, push_space);
2422         }
2423
2424         if (right->map_token) {
2425                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2426                 right->map_token = NULL;
2427         }
2428         left_nritems -= push_items;
2429         btrfs_set_header_nritems(left, left_nritems);
2430
2431         if (left_nritems)
2432                 btrfs_mark_buffer_dirty(left);
2433         else
2434                 clean_tree_block(trans, root, left);
2435
2436         btrfs_mark_buffer_dirty(right);
2437
2438         btrfs_item_key(right, &disk_key, 0);
2439         btrfs_set_node_key(upper, &disk_key, slot + 1);
2440         btrfs_mark_buffer_dirty(upper);
2441
2442         /* then fixup the leaf pointer in the path */
2443         if (path->slots[0] >= left_nritems) {
2444                 path->slots[0] -= left_nritems;
2445                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2446                         clean_tree_block(trans, root, path->nodes[0]);
2447                 btrfs_tree_unlock(path->nodes[0]);
2448                 free_extent_buffer(path->nodes[0]);
2449                 path->nodes[0] = right;
2450                 path->slots[1] += 1;
2451         } else {
2452                 btrfs_tree_unlock(right);
2453                 free_extent_buffer(right);
2454         }
2455         return 0;
2456
2457 out_unlock:
2458         btrfs_tree_unlock(right);
2459         free_extent_buffer(right);
2460         return 1;
2461 }
2462
2463 /*
2464  * push some data in the path leaf to the right, trying to free up at
2465  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2466  *
2467  * returns 1 if the push failed because the other node didn't have enough
2468  * room, 0 if everything worked out and < 0 if there were major errors.
2469  */
2470 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2471                            *root, struct btrfs_path *path, int data_size,
2472                            int empty)
2473 {
2474         struct extent_buffer *left = path->nodes[0];
2475         struct extent_buffer *right;
2476         struct extent_buffer *upper;
2477         int slot;
2478         int free_space;
2479         u32 left_nritems;
2480         int ret;
2481
2482         if (!path->nodes[1])
2483                 return 1;
2484
2485         slot = path->slots[1];
2486         upper = path->nodes[1];
2487         if (slot >= btrfs_header_nritems(upper) - 1)
2488                 return 1;
2489
2490         btrfs_assert_tree_locked(path->nodes[1]);
2491
2492         right = read_node_slot(root, upper, slot + 1);
2493         btrfs_tree_lock(right);
2494         btrfs_set_lock_blocking(right);
2495
2496         free_space = btrfs_leaf_free_space(root, right);
2497         if (free_space < data_size)
2498                 goto out_unlock;
2499
2500         /* cow and double check */
2501         ret = btrfs_cow_block(trans, root, right, upper,
2502                               slot + 1, &right);
2503         if (ret)
2504                 goto out_unlock;
2505
2506         free_space = btrfs_leaf_free_space(root, right);
2507         if (free_space < data_size)
2508                 goto out_unlock;
2509
2510         left_nritems = btrfs_header_nritems(left);
2511         if (left_nritems == 0)
2512                 goto out_unlock;
2513
2514         return __push_leaf_right(trans, root, path, data_size, empty,
2515                                 right, free_space, left_nritems);
2516 out_unlock:
2517         btrfs_tree_unlock(right);
2518         free_extent_buffer(right);
2519         return 1;
2520 }
2521
2522 /*
2523  * push some data in the path leaf to the left, trying to free up at
2524  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2525  */
2526 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2527                                      struct btrfs_root *root,
2528                                      struct btrfs_path *path, int data_size,
2529                                      int empty, struct extent_buffer *left,
2530                                      int free_space, int right_nritems)
2531 {
2532         struct btrfs_disk_key disk_key;
2533         struct extent_buffer *right = path->nodes[0];
2534         int slot;
2535         int i;
2536         int push_space = 0;
2537         int push_items = 0;
2538         struct btrfs_item *item;
2539         u32 old_left_nritems;
2540         u32 nr;
2541         int ret = 0;
2542         int wret;
2543         u32 this_item_size;
2544         u32 old_left_item_size;
2545
2546         slot = path->slots[1];
2547
2548         if (empty)
2549                 nr = right_nritems;
2550         else
2551                 nr = right_nritems - 1;
2552
2553         for (i = 0; i < nr; i++) {
2554                 item = btrfs_item_nr(right, i);
2555                 if (!right->map_token) {
2556                         map_extent_buffer(right, (unsigned long)item,
2557                                         sizeof(struct btrfs_item),
2558                                         &right->map_token, &right->kaddr,
2559                                         &right->map_start, &right->map_len,
2560                                         KM_USER1);
2561                 }
2562
2563                 if (!empty && push_items > 0) {
2564                         if (path->slots[0] < i)
2565                                 break;
2566                         if (path->slots[0] == i) {
2567                                 int space = btrfs_leaf_free_space(root, right);
2568                                 if (space + push_space * 2 > free_space)
2569                                         break;
2570                         }
2571                 }
2572
2573                 if (path->slots[0] == i)
2574                         push_space += data_size;
2575
2576                 this_item_size = btrfs_item_size(right, item);
2577                 if (this_item_size + sizeof(*item) + push_space > free_space)
2578                         break;
2579
2580                 push_items++;
2581                 push_space += this_item_size + sizeof(*item);
2582         }
2583
2584         if (right->map_token) {
2585                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2586                 right->map_token = NULL;
2587         }
2588
2589         if (push_items == 0) {
2590                 ret = 1;
2591                 goto out;
2592         }
2593         if (!empty && push_items == btrfs_header_nritems(right))
2594                 WARN_ON(1);
2595
2596         /* push data from right to left */
2597         copy_extent_buffer(left, right,
2598                            btrfs_item_nr_offset(btrfs_header_nritems(left)),
2599                            btrfs_item_nr_offset(0),
2600                            push_items * sizeof(struct btrfs_item));
2601
2602         push_space = BTRFS_LEAF_DATA_SIZE(root) -
2603                      btrfs_item_offset_nr(right, push_items - 1);
2604
2605         copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2606                      leaf_data_end(root, left) - push_space,
2607                      btrfs_leaf_data(right) +
2608                      btrfs_item_offset_nr(right, push_items - 1),
2609                      push_space);
2610         old_left_nritems = btrfs_header_nritems(left);
2611         BUG_ON(old_left_nritems <= 0);
2612
2613         old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2614         for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2615                 u32 ioff;
2616
2617                 item = btrfs_item_nr(left, i);
2618                 if (!left->map_token) {
2619                         map_extent_buffer(left, (unsigned long)item,
2620                                         sizeof(struct btrfs_item),
2621                                         &left->map_token, &left->kaddr,
2622                                         &left->map_start, &left->map_len,
2623                                         KM_USER1);
2624                 }
2625
2626                 ioff = btrfs_item_offset(left, item);
2627                 btrfs_set_item_offset(left, item,
2628                       ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2629         }
2630         btrfs_set_header_nritems(left, old_left_nritems + push_items);
2631         if (left->map_token) {
2632                 unmap_extent_buffer(left, left->map_token, KM_USER1);
2633                 left->map_token = NULL;
2634         }
2635
2636         /* fixup right node */
2637         if (push_items > right_nritems) {
2638                 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2639                        right_nritems);
2640                 WARN_ON(1);
2641         }
2642
2643         if (push_items < right_nritems) {
2644                 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2645                                                   leaf_data_end(root, right);
2646                 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2647                                       BTRFS_LEAF_DATA_SIZE(root) - push_space,
2648                                       btrfs_leaf_data(right) +
2649                                       leaf_data_end(root, right), push_space);
2650
2651                 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2652                               btrfs_item_nr_offset(push_items),
2653                              (btrfs_header_nritems(right) - push_items) *
2654                              sizeof(struct btrfs_item));
2655         }
2656         right_nritems -= push_items;
2657         btrfs_set_header_nritems(right, right_nritems);
2658         push_space = BTRFS_LEAF_DATA_SIZE(root);
2659         for (i = 0; i < right_nritems; i++) {
2660                 item = btrfs_item_nr(right, i);
2661
2662                 if (!right->map_token) {
2663                         map_extent_buffer(right, (unsigned long)item,
2664                                         sizeof(struct btrfs_item),
2665                                         &right->map_token, &right->kaddr,
2666                                         &right->map_start, &right->map_len,
2667                                         KM_USER1);
2668                 }
2669
2670                 push_space = push_space - btrfs_item_size(right, item);
2671                 btrfs_set_item_offset(right, item, push_space);
2672         }
2673         if (right->map_token) {
2674                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2675                 right->map_token = NULL;
2676         }
2677
2678         btrfs_mark_buffer_dirty(left);
2679         if (right_nritems)
2680                 btrfs_mark_buffer_dirty(right);
2681         else
2682                 clean_tree_block(trans, root, right);
2683
2684         btrfs_item_key(right, &disk_key, 0);
2685         wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2686         if (wret)
2687                 ret = wret;
2688
2689         /* then fixup the leaf pointer in the path */
2690         if (path->slots[0] < push_items) {
2691                 path->slots[0] += old_left_nritems;
2692                 btrfs_tree_unlock(path->nodes[0]);
2693                 free_extent_buffer(path->nodes[0]);
2694                 path->nodes[0] = left;
2695                 path->slots[1] -= 1;
2696         } else {
2697                 btrfs_tree_unlock(left);
2698                 free_extent_buffer(left);
2699                 path->slots[0] -= push_items;
2700         }
2701         BUG_ON(path->slots[0] < 0);
2702         return ret;
2703 out:
2704         btrfs_tree_unlock(left);
2705         free_extent_buffer(left);
2706         return ret;
2707 }
2708
2709 /*
2710  * push some data in the path leaf to the left, trying to free up at
2711  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2712  */
2713 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2714                           *root, struct btrfs_path *path, int data_size,
2715                           int empty)
2716 {
2717         struct extent_buffer *right = path->nodes[0];
2718         struct extent_buffer *left;
2719         int slot;
2720         int free_space;
2721         u32 right_nritems;
2722         int ret = 0;
2723
2724         slot = path->slots[1];
2725         if (slot == 0)
2726                 return 1;
2727         if (!path->nodes[1])
2728                 return 1;
2729
2730         right_nritems = btrfs_header_nritems(right);
2731         if (right_nritems == 0)
2732                 return 1;
2733
2734         btrfs_assert_tree_locked(path->nodes[1]);
2735
2736         left = read_node_slot(root, path->nodes[1], slot - 1);
2737         btrfs_tree_lock(left);
2738         btrfs_set_lock_blocking(left);
2739
2740         free_space = btrfs_leaf_free_space(root, left);
2741         if (free_space < data_size) {
2742                 ret = 1;
2743                 goto out;
2744         }
2745
2746         /* cow and double check */
2747         ret = btrfs_cow_block(trans, root, left,
2748                               path->nodes[1], slot - 1, &left);
2749         if (ret) {
2750                 /* we hit -ENOSPC, but it isn't fatal here */
2751                 ret = 1;
2752                 goto out;
2753         }
2754
2755         free_space = btrfs_leaf_free_space(root, left);
2756         if (free_space < data_size) {
2757                 ret = 1;
2758                 goto out;
2759         }
2760
2761         return __push_leaf_left(trans, root, path, data_size,
2762                                empty, left, free_space, right_nritems);
2763 out:
2764         btrfs_tree_unlock(left);
2765         free_extent_buffer(left);
2766         return ret;
2767 }
2768
2769 /*
2770  * split the path's leaf in two, making sure there is at least data_size
2771  * available for the resulting leaf level of the path.
2772  *
2773  * returns 0 if all went well and < 0 on failure.
2774  */
2775 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2776                                struct btrfs_root *root,
2777                                struct btrfs_path *path,
2778                                struct extent_buffer *l,
2779                                struct extent_buffer *right,
2780                                int slot, int mid, int nritems)
2781 {
2782         int data_copy_size;
2783         int rt_data_off;
2784         int i;
2785         int ret = 0;
2786         int wret;
2787         struct btrfs_disk_key disk_key;
2788
2789         nritems = nritems - mid;
2790         btrfs_set_header_nritems(right, nritems);
2791         data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2792
2793         copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2794                            btrfs_item_nr_offset(mid),
2795                            nritems * sizeof(struct btrfs_item));
2796
2797         copy_extent_buffer(right, l,
2798                      btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2799                      data_copy_size, btrfs_leaf_data(l) +
2800                      leaf_data_end(root, l), data_copy_size);
2801
2802         rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2803                       btrfs_item_end_nr(l, mid);
2804
2805         for (i = 0; i < nritems; i++) {
2806                 struct btrfs_item *item = btrfs_item_nr(right, i);
2807                 u32 ioff;
2808
2809                 if (!right->map_token) {
2810                         map_extent_buffer(right, (unsigned long)item,
2811                                         sizeof(struct btrfs_item),
2812                                         &right->map_token, &right->kaddr,
2813                                         &right->map_start, &right->map_len,
2814                                         KM_USER1);
2815                 }
2816
2817                 ioff = btrfs_item_offset(right, item);
2818                 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2819         }
2820
2821         if (right->map_token) {
2822                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2823                 right->map_token = NULL;
2824         }
2825
2826         btrfs_set_header_nritems(l, mid);
2827         ret = 0;
2828         btrfs_item_key(right, &disk_key, 0);
2829         wret = insert_ptr(trans, root, path, &disk_key, right->start,
2830                           path->slots[1] + 1, 1);
2831         if (wret)
2832                 ret = wret;
2833
2834         btrfs_mark_buffer_dirty(right);
2835         btrfs_mark_buffer_dirty(l);
2836         BUG_ON(path->slots[0] != slot);
2837
2838         if (mid <= slot) {
2839                 btrfs_tree_unlock(path->nodes[0]);
2840                 free_extent_buffer(path->nodes[0]);
2841                 path->nodes[0] = right;
2842                 path->slots[0] -= mid;
2843                 path->slots[1] += 1;
2844         } else {
2845                 btrfs_tree_unlock(right);
2846                 free_extent_buffer(right);
2847         }
2848
2849         BUG_ON(path->slots[0] < 0);
2850
2851         return ret;
2852 }
2853
2854 /*
2855  * split the path's leaf in two, making sure there is at least data_size
2856  * available for the resulting leaf level of the path.
2857  *
2858  * returns 0 if all went well and < 0 on failure.
2859  */
2860 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2861                                struct btrfs_root *root,
2862                                struct btrfs_key *ins_key,
2863                                struct btrfs_path *path, int data_size,
2864                                int extend)
2865 {
2866         struct btrfs_disk_key disk_key;
2867         struct extent_buffer *l;
2868         u32 nritems;
2869         int mid;
2870         int slot;
2871         struct extent_buffer *right;
2872         int ret = 0;
2873         int wret;
2874         int split;
2875         int num_doubles = 0;
2876
2877         l = path->nodes[0];
2878         slot = path->slots[0];
2879         if (extend && data_size + btrfs_item_size_nr(l, slot) +
2880             sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2881                 return -EOVERFLOW;
2882
2883         /* first try to make some room by pushing left and right */
2884         if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2885                 wret = push_leaf_right(trans, root, path, data_size, 0);
2886                 if (wret < 0)
2887                         return wret;
2888                 if (wret) {
2889                         wret = push_leaf_left(trans, root, path, data_size, 0);
2890                         if (wret < 0)
2891                                 return wret;
2892                 }
2893                 l = path->nodes[0];
2894
2895                 /* did the pushes work? */
2896                 if (btrfs_leaf_free_space(root, l) >= data_size)
2897                         return 0;
2898         }
2899
2900         if (!path->nodes[1]) {
2901                 ret = insert_new_root(trans, root, path, 1);
2902                 if (ret)
2903                         return ret;
2904         }
2905 again:
2906         split = 1;
2907         l = path->nodes[0];
2908         slot = path->slots[0];
2909         nritems = btrfs_header_nritems(l);
2910         mid = (nritems + 1) / 2;
2911
2912         if (mid <= slot) {
2913                 if (nritems == 1 ||
2914                     leaf_space_used(l, mid, nritems - mid) + data_size >
2915                         BTRFS_LEAF_DATA_SIZE(root)) {
2916                         if (slot >= nritems) {
2917                                 split = 0;
2918                         } else {
2919                                 mid = slot;
2920                                 if (mid != nritems &&
2921                                     leaf_space_used(l, mid, nritems - mid) +
2922                                     data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2923                                         split = 2;
2924                                 }
2925                         }
2926                 }
2927         } else {
2928                 if (leaf_space_used(l, 0, mid) + data_size >
2929                         BTRFS_LEAF_DATA_SIZE(root)) {
2930                         if (!extend && data_size && slot == 0) {
2931                                 split = 0;
2932                         } else if ((extend || !data_size) && slot == 0) {
2933                                 mid = 1;
2934                         } else {
2935                                 mid = slot;
2936                                 if (mid != nritems &&
2937                                     leaf_space_used(l, mid, nritems - mid) +
2938                                     data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2939                                         split = 2 ;
2940                                 }
2941                         }
2942                 }
2943         }
2944
2945         if (split == 0)
2946                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2947         else
2948                 btrfs_item_key(l, &disk_key, mid);
2949
2950         right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
2951                                         root->root_key.objectid,
2952                                         &disk_key, 0, l->start, 0);
2953         if (IS_ERR(right))
2954                 return PTR_ERR(right);
2955
2956         root_add_used(root, root->leafsize);
2957
2958         memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2959         btrfs_set_header_bytenr(right, right->start);
2960         btrfs_set_header_generation(right, trans->transid);
2961         btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2962         btrfs_set_header_owner(right, root->root_key.objectid);
2963         btrfs_set_header_level(right, 0);
2964         write_extent_buffer(right, root->fs_info->fsid,
2965                             (unsigned long)btrfs_header_fsid(right),
2966                             BTRFS_FSID_SIZE);
2967
2968         write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2969                             (unsigned long)btrfs_header_chunk_tree_uuid(right),
2970                             BTRFS_UUID_SIZE);
2971
2972         if (split == 0) {
2973                 if (mid <= slot) {
2974                         btrfs_set_header_nritems(right, 0);
2975                         wret = insert_ptr(trans, root, path,
2976                                           &disk_key, right->start,
2977                                           path->slots[1] + 1, 1);
2978                         if (wret)
2979                                 ret = wret;
2980
2981                         btrfs_tree_unlock(path->nodes[0]);
2982                         free_extent_buffer(path->nodes[0]);
2983                         path->nodes[0] = right;
2984                         path->slots[0] = 0;
2985                         path->slots[1] += 1;
2986                 } else {
2987                         btrfs_set_header_nritems(right, 0);
2988                         wret = insert_ptr(trans, root, path,
2989                                           &disk_key,
2990                                           right->start,
2991                                           path->slots[1], 1);
2992                         if (wret)
2993                                 ret = wret;
2994                         btrfs_tree_unlock(path->nodes[0]);
2995                         free_extent_buffer(path->nodes[0]);
2996                         path->nodes[0] = right;
2997                         path->slots[0] = 0;
2998                         if (path->slots[1] == 0) {
2999                                 wret = fixup_low_keys(trans, root,
3000                                                 path, &disk_key, 1);
3001                                 if (wret)
3002                                         ret = wret;
3003                         }
3004                 }
3005                 btrfs_mark_buffer_dirty(right);
3006                 return ret;
3007         }
3008
3009         ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
3010         BUG_ON(ret);
3011
3012         if (split == 2) {
3013                 BUG_ON(num_doubles != 0);
3014                 num_doubles++;
3015                 goto again;
3016         }
3017
3018         return ret;
3019 }
3020
3021 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3022                                          struct btrfs_root *root,
3023                                          struct btrfs_path *path, int ins_len)
3024 {
3025         struct btrfs_key key;
3026         struct extent_buffer *leaf;
3027         struct btrfs_file_extent_item *fi;
3028         u64 extent_len = 0;
3029         u32 item_size;
3030         int ret;
3031
3032         leaf = path->nodes[0];
3033         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3034
3035         BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3036                key.type != BTRFS_EXTENT_CSUM_KEY);
3037
3038         if (btrfs_leaf_free_space(root, leaf) >= ins_len)
3039                 return 0;
3040
3041         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3042         if (key.type == BTRFS_EXTENT_DATA_KEY) {
3043                 fi = btrfs_item_ptr(leaf, path->slots[0],
3044                                     struct btrfs_file_extent_item);
3045                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3046         }
3047         btrfs_release_path(root, path);
3048
3049         path->keep_locks = 1;
3050         path->search_for_split = 1;
3051         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3052         path->search_for_split = 0;
3053         if (ret < 0)
3054                 goto err;
3055
3056         ret = -EAGAIN;
3057         leaf = path->nodes[0];
3058         /* if our item isn't there or got smaller, return now */
3059         if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3060                 goto err;
3061
3062         /* the leaf has  changed, it now has room.  return now */
3063         if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
3064                 goto err;
3065
3066         if (key.type == BTRFS_EXTENT_DATA_KEY) {
3067                 fi = btrfs_item_ptr(leaf, path->slots[0],
3068                                     struct btrfs_file_extent_item);
3069                 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3070                         goto err;
3071         }
3072
3073         btrfs_set_path_blocking(path);
3074         ret = split_leaf(trans, root, &key, path, ins_len, 1);
3075         if (ret)
3076                 goto err;
3077
3078         path->keep_locks = 0;
3079         btrfs_unlock_up_safe(path, 1);
3080         return 0;
3081 err:
3082         path->keep_locks = 0;
3083         return ret;
3084 }
3085
3086 static noinline int split_item(struct btrfs_trans_handle *trans,
3087                                struct btrfs_root *root,
3088                                struct btrfs_path *path,
3089                                struct btrfs_key *new_key,
3090                                unsigned long split_offset)
3091 {
3092         struct extent_buffer *leaf;
3093         struct btrfs_item *item;
3094         struct btrfs_item *new_item;
3095         int slot;
3096         char *buf;
3097         u32 nritems;
3098         u32 item_size;
3099         u32 orig_offset;
3100         struct btrfs_disk_key disk_key;
3101
3102         leaf = path->nodes[0];
3103         BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3104
3105         btrfs_set_path_blocking(path);
3106
3107         item = btrfs_item_nr(leaf, path->slots[0]);
3108         orig_offset = btrfs_item_offset(leaf, item);
3109         item_size = btrfs_item_size(leaf, item);
3110
3111         buf = kmalloc(item_size, GFP_NOFS);
3112         if (!buf)
3113                 return -ENOMEM;
3114
3115         read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3116                             path->slots[0]), item_size);
3117
3118         slot = path->slots[0] + 1;
3119         nritems = btrfs_header_nritems(leaf);
3120         if (slot != nritems) {
3121                 /* shift the items */
3122                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3123                                 btrfs_item_nr_offset(slot),
3124                                 (nritems - slot) * sizeof(struct btrfs_item));
3125         }
3126
3127         btrfs_cpu_key_to_disk(&disk_key, new_key);
3128         btrfs_set_item_key(leaf, &disk_key, slot);
3129
3130         new_item = btrfs_item_nr(leaf, slot);
3131
3132         btrfs_set_item_offset(leaf, new_item, orig_offset);
3133         btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3134
3135         btrfs_set_item_offset(leaf, item,
3136                               orig_offset + item_size - split_offset);
3137         btrfs_set_item_size(leaf, item, split_offset);
3138
3139         btrfs_set_header_nritems(leaf, nritems + 1);
3140
3141         /* write the data for the start of the original item */
3142         write_extent_buffer(leaf, buf,
3143                             btrfs_item_ptr_offset(leaf, path->slots[0]),
3144                             split_offset);
3145
3146         /* write the data for the new item */
3147         write_extent_buffer(leaf, buf + split_offset,
3148                             btrfs_item_ptr_offset(leaf, slot),
3149                             item_size - split_offset);
3150         btrfs_mark_buffer_dirty(leaf);
3151
3152         BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
3153         kfree(buf);
3154         return 0;
3155 }
3156
3157 /*
3158  * This function splits a single item into two items,
3159  * giving 'new_key' to the new item and splitting the
3160  * old one at split_offset (from the start of the item).
3161  *
3162  * The path may be released by this operation.  After
3163  * the split, the path is pointing to the old item.  The
3164  * new item is going to be in the same node as the old one.
3165  *
3166  * Note, the item being split must be smaller enough to live alone on
3167  * a tree block with room for one extra struct btrfs_item
3168  *
3169  * This allows us to split the item in place, keeping a lock on the
3170  * leaf the entire time.
3171  */
3172 int btrfs_split_item(struct btrfs_trans_handle *trans,
3173                      struct btrfs_root *root,
3174                      struct btrfs_path *path,
3175                      struct btrfs_key *new_key,
3176                      unsigned long split_offset)
3177 {
3178         int ret;
3179         ret = setup_leaf_for_split(trans, root, path,
3180                                    sizeof(struct btrfs_item));
3181         if (ret)
3182                 return ret;
3183
3184         ret = split_item(trans, root, path, new_key, split_offset);
3185         return ret;
3186 }
3187
3188 /*
3189  * This function duplicate a item, giving 'new_key' to the new item.
3190  * It guarantees both items live in the same tree leaf and the new item
3191  * is contiguous with the original item.
3192  *
3193  * This allows us to split file extent in place, keeping a lock on the
3194  * leaf the entire time.
3195  */
3196 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
3197                          struct btrfs_root *root,
3198                          struct btrfs_path *path,
3199                          struct btrfs_key *new_key)
3200 {
3201         struct extent_buffer *leaf;
3202         int ret;
3203         u32 item_size;
3204
3205         leaf = path->nodes[0];
3206         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3207         ret = setup_leaf_for_split(trans, root, path,
3208                                    item_size + sizeof(struct btrfs_item));
3209         if (ret)
3210                 return ret;
3211
3212         path->slots[0]++;
3213         ret = setup_items_for_insert(trans, root, path, new_key, &item_size,
3214                                      item_size, item_size +
3215                                      sizeof(struct btrfs_item), 1);
3216         BUG_ON(ret);
3217
3218         leaf = path->nodes[0];
3219         memcpy_extent_buffer(leaf,
3220                              btrfs_item_ptr_offset(leaf, path->slots[0]),
3221                              btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
3222                              item_size);
3223         return 0;
3224 }
3225
3226 /*
3227  * make the item pointed to by the path smaller.  new_size indicates
3228  * how small to make it, and from_end tells us if we just chop bytes
3229  * off the end of the item or if we shift the item to chop bytes off
3230  * the front.
3231  */
3232 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3233                         struct btrfs_root *root,
3234                         struct btrfs_path *path,
3235                         u32 new_size, int from_end)
3236 {
3237         int ret = 0;
3238         int slot;
3239         int slot_orig;
3240         struct extent_buffer *leaf;
3241         struct btrfs_item *item;
3242         u32 nritems;
3243         unsigned int data_end;
3244         unsigned int old_data_start;
3245         unsigned int old_size;
3246         unsigned int size_diff;
3247         int i;
3248
3249         slot_orig = path->slots[0];
3250         leaf = path->nodes[0];
3251         slot = path->slots[0];
3252
3253         old_size = btrfs_item_size_nr(leaf, slot);
3254         if (old_size == new_size)
3255                 return 0;
3256
3257         nritems = btrfs_header_nritems(leaf);
3258         data_end = leaf_data_end(root, leaf);
3259
3260         old_data_start = btrfs_item_offset_nr(leaf, slot);
3261
3262         size_diff = old_size - new_size;
3263
3264         BUG_ON(slot < 0);
3265         BUG_ON(slot >= nritems);
3266
3267         /*
3268          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3269          */
3270         /* first correct the data pointers */
3271         for (i = slot; i < nritems; i++) {
3272                 u32 ioff;
3273                 item = btrfs_item_nr(leaf, i);
3274
3275                 if (!leaf->map_token) {
3276                         map_extent_buffer(leaf, (unsigned long)item,
3277                                         sizeof(struct btrfs_item),
3278                                         &leaf->map_token, &leaf->kaddr,
3279                                         &leaf->map_start, &leaf->map_len,
3280                                         KM_USER1);
3281                 }
3282
3283                 ioff = btrfs_item_offset(leaf, item);
3284                 btrfs_set_item_offset(leaf, item, ioff + size_diff);
3285         }
3286
3287         if (leaf->map_token) {
3288                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3289                 leaf->map_token = NULL;
3290         }
3291
3292         /* shift the data */
3293         if (from_end) {
3294                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3295                               data_end + size_diff, btrfs_leaf_data(leaf) +
3296                               data_end, old_data_start + new_size - data_end);
3297         } else {
3298                 struct btrfs_disk_key disk_key;
3299                 u64 offset;
3300
3301                 btrfs_item_key(leaf, &disk_key, slot);
3302
3303                 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3304                         unsigned long ptr;
3305                         struct btrfs_file_extent_item *fi;
3306
3307                         fi = btrfs_item_ptr(leaf, slot,
3308                                             struct btrfs_file_extent_item);
3309                         fi = (struct btrfs_file_extent_item *)(
3310                              (unsigned long)fi - size_diff);
3311
3312                         if (btrfs_file_extent_type(leaf, fi) ==
3313                             BTRFS_FILE_EXTENT_INLINE) {
3314                                 ptr = btrfs_item_ptr_offset(leaf, slot);
3315                                 memmove_extent_buffer(leaf, ptr,
3316                                       (unsigned long)fi,
3317                                       offsetof(struct btrfs_file_extent_item,
3318                                                  disk_bytenr));
3319                         }
3320                 }
3321
3322                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3323                               data_end + size_diff, btrfs_leaf_data(leaf) +
3324                               data_end, old_data_start - data_end);
3325
3326                 offset = btrfs_disk_key_offset(&disk_key);
3327                 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3328                 btrfs_set_item_key(leaf, &disk_key, slot);
3329                 if (slot == 0)
3330                         fixup_low_keys(trans, root, path, &disk_key, 1);
3331         }
3332
3333         item = btrfs_item_nr(leaf, slot);
3334         btrfs_set_item_size(leaf, item, new_size);
3335         btrfs_mark_buffer_dirty(leaf);
3336
3337         ret = 0;
3338         if (btrfs_leaf_free_space(root, leaf) < 0) {
3339                 btrfs_print_leaf(root, leaf);
3340                 BUG();
3341         }
3342         return ret;
3343 }
3344
3345 /*
3346  * make the item pointed to by the path bigger, data_size is the new size.
3347  */
3348 int btrfs_extend_item(struct btrfs_trans_handle *trans,
3349                       struct btrfs_root *root, struct btrfs_path *path,
3350                       u32 data_size)
3351 {
3352         int ret = 0;
3353         int slot;
3354         int slot_orig;
3355         struct extent_buffer *leaf;
3356         struct btrfs_item *item;
3357         u32 nritems;
3358         unsigned int data_end;
3359         unsigned int old_data;
3360         unsigned int old_size;
3361         int i;
3362
3363         slot_orig = path->slots[0];
3364         leaf = path->nodes[0];
3365
3366         nritems = btrfs_header_nritems(leaf);
3367         data_end = leaf_data_end(root, leaf);
3368
3369         if (btrfs_leaf_free_space(root, leaf) < data_size) {
3370                 btrfs_print_leaf(root, leaf);
3371                 BUG();
3372         }
3373         slot = path->slots[0];
3374         old_data = btrfs_item_end_nr(leaf, slot);
3375
3376         BUG_ON(slot < 0);
3377         if (slot >= nritems) {
3378                 btrfs_print_leaf(root, leaf);
3379                 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3380                        slot, nritems);
3381                 BUG_ON(1);
3382         }
3383
3384         /*
3385          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3386          */
3387         /* first correct the data pointers */
3388         for (i = slot; i < nritems; i++) {
3389                 u32 ioff;
3390                 item = btrfs_item_nr(leaf, i);
3391
3392                 if (!leaf->map_token) {
3393                         map_extent_buffer(leaf, (unsigned long)item,
3394                                         sizeof(struct btrfs_item),
3395                                         &leaf->map_token, &leaf->kaddr,
3396                                         &leaf->map_start, &leaf->map_len,
3397                                         KM_USER1);
3398                 }
3399                 ioff = btrfs_item_offset(leaf, item);
3400                 btrfs_set_item_offset(leaf, item, ioff - data_size);
3401         }
3402
3403         if (leaf->map_token) {
3404                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3405                 leaf->map_token = NULL;
3406         }
3407
3408         /* shift the data */
3409         memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3410                       data_end - data_size, btrfs_leaf_data(leaf) +
3411                       data_end, old_data - data_end);
3412
3413         data_end = old_data;
3414         old_size = btrfs_item_size_nr(leaf, slot);
3415         item = btrfs_item_nr(leaf, slot);
3416         btrfs_set_item_size(leaf, item, old_size + data_size);
3417         btrfs_mark_buffer_dirty(leaf);
3418
3419         ret = 0;
3420         if (btrfs_leaf_free_space(root, leaf) < 0) {
3421                 btrfs_print_leaf(root, leaf);
3422                 BUG();
3423         }
3424         return ret;
3425 }
3426
3427 /*
3428  * Given a key and some data, insert items into the tree.
3429  * This does all the path init required, making room in the tree if needed.
3430  * Returns the number of keys that were inserted.
3431  */
3432 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3433                             struct btrfs_root *root,
3434                             struct btrfs_path *path,
3435                             struct btrfs_key *cpu_key, u32 *data_size,
3436                             int nr)
3437 {
3438         struct extent_buffer *leaf;
3439         struct btrfs_item *item;
3440         int ret = 0;
3441         int slot;
3442         int i;
3443         u32 nritems;
3444         u32 total_data = 0;
3445         u32 total_size = 0;
3446         unsigned int data_end;
3447         struct btrfs_disk_key disk_key;
3448         struct btrfs_key found_key;
3449
3450         for (i = 0; i < nr; i++) {
3451                 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3452                     BTRFS_LEAF_DATA_SIZE(root)) {
3453                         break;
3454                         nr = i;
3455                 }
3456                 total_data += data_size[i];
3457                 total_size += data_size[i] + sizeof(struct btrfs_item);
3458         }
3459         BUG_ON(nr == 0);
3460
3461         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3462         if (ret == 0)
3463                 return -EEXIST;
3464         if (ret < 0)
3465                 goto out;
3466
3467         leaf = path->nodes[0];
3468
3469         nritems = btrfs_header_nritems(leaf);
3470         data_end = leaf_data_end(root, leaf);
3471
3472         if (btrfs_leaf_free_space(root, leaf) < total_size) {
3473                 for (i = nr; i >= 0; i--) {
3474                         total_data -= data_size[i];
3475                         total_size -= data_size[i] + sizeof(struct btrfs_item);
3476                         if (total_size < btrfs_leaf_free_space(root, leaf))
3477                                 break;
3478                 }
3479                 nr = i;
3480         }
3481
3482         slot = path->slots[0];
3483         BUG_ON(slot < 0);
3484
3485         if (slot != nritems) {
3486                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3487
3488                 item = btrfs_item_nr(leaf, slot);
3489                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3490
3491                 /* figure out how many keys we can insert in here */
3492                 total_data = data_size[0];
3493                 for (i = 1; i < nr; i++) {
3494                         if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3495                                 break;
3496                         total_data += data_size[i];
3497                 }
3498                 nr = i;
3499
3500                 if (old_data < data_end) {
3501                         btrfs_print_leaf(root, leaf);
3502                         printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3503                                slot, old_data, data_end);
3504                         BUG_ON(1);
3505                 }
3506                 /*
3507                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3508                  */
3509                 /* first correct the data pointers */
3510                 WARN_ON(leaf->map_token);
3511                 for (i = slot; i < nritems; i++) {
3512                         u32 ioff;
3513
3514                         item = btrfs_item_nr(leaf, i);
3515                         if (!leaf->map_token) {
3516                                 map_extent_buffer(leaf, (unsigned long)item,
3517                                         sizeof(struct btrfs_item),
3518                                         &leaf->map_token, &leaf->kaddr,
3519                                         &leaf->map_start, &leaf->map_len,
3520                                         KM_USER1);
3521                         }
3522
3523                         ioff = btrfs_item_offset(leaf, item);
3524                         btrfs_set_item_offset(leaf, item, ioff - total_data);
3525                 }
3526                 if (leaf->map_token) {
3527                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3528                         leaf->map_token = NULL;
3529                 }
3530
3531                 /* shift the items */
3532                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3533                               btrfs_item_nr_offset(slot),
3534                               (nritems - slot) * sizeof(struct btrfs_item));
3535
3536                 /* shift the data */
3537                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3538                               data_end - total_data, btrfs_leaf_data(leaf) +
3539                               data_end, old_data - data_end);
3540                 data_end = old_data;
3541         } else {
3542                 /*
3543                  * this sucks but it has to be done, if we are inserting at
3544                  * the end of the leaf only insert 1 of the items, since we
3545                  * have no way of knowing whats on the next leaf and we'd have
3546                  * to drop our current locks to figure it out
3547                  */
3548                 nr = 1;
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         btrfs_set_header_nritems(leaf, nritems + nr);
3561         btrfs_mark_buffer_dirty(leaf);
3562
3563         ret = 0;
3564         if (slot == 0) {
3565                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3566                 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3567         }
3568
3569         if (btrfs_leaf_free_space(root, leaf) < 0) {
3570                 btrfs_print_leaf(root, leaf);
3571                 BUG();
3572         }
3573 out:
3574         if (!ret)
3575                 ret = nr;
3576         return ret;
3577 }
3578
3579 /*
3580  * this is a helper for btrfs_insert_empty_items, the main goal here is
3581  * to save stack depth by doing the bulk of the work in a function
3582  * that doesn't call btrfs_search_slot
3583  */
3584 static noinline_for_stack int
3585 setup_items_for_insert(struct btrfs_trans_handle *trans,
3586                       struct btrfs_root *root, struct btrfs_path *path,
3587                       struct btrfs_key *cpu_key, u32 *data_size,
3588                       u32 total_data, u32 total_size, int nr)
3589 {
3590         struct btrfs_item *item;
3591         int i;
3592         u32 nritems;
3593         unsigned int data_end;
3594         struct btrfs_disk_key disk_key;
3595         int ret;
3596         struct extent_buffer *leaf;
3597         int slot;
3598
3599         leaf = path->nodes[0];
3600         slot = path->slots[0];
3601
3602         nritems = btrfs_header_nritems(leaf);
3603         data_end = leaf_data_end(root, leaf);
3604
3605         if (btrfs_leaf_free_space(root, leaf) < total_size) {
3606                 btrfs_print_leaf(root, leaf);
3607                 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3608                        total_size, btrfs_leaf_free_space(root, leaf));
3609                 BUG();
3610         }
3611
3612         if (slot != nritems) {
3613                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3614
3615                 if (old_data < data_end) {
3616                         btrfs_print_leaf(root, leaf);
3617                         printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3618                                slot, old_data, data_end);
3619                         BUG_ON(1);
3620                 }
3621                 /*
3622                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3623                  */
3624                 /* first correct the data pointers */
3625                 WARN_ON(leaf->map_token);
3626                 for (i = slot; i < nritems; i++) {
3627                         u32 ioff;
3628
3629                         item = btrfs_item_nr(leaf, i);
3630                         if (!leaf->map_token) {
3631                                 map_extent_buffer(leaf, (unsigned long)item,
3632                                         sizeof(struct btrfs_item),
3633                                         &leaf->map_token, &leaf->kaddr,
3634                                         &leaf->map_start, &leaf->map_len,
3635                                         KM_USER1);
3636                         }
3637
3638                         ioff = btrfs_item_offset(leaf, item);
3639                         btrfs_set_item_offset(leaf, item, ioff - total_data);
3640                 }
3641                 if (leaf->map_token) {
3642                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3643                         leaf->map_token = NULL;
3644                 }
3645
3646                 /* shift the items */
3647                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3648                               btrfs_item_nr_offset(slot),
3649                               (nritems - slot) * sizeof(struct btrfs_item));
3650
3651                 /* shift the data */
3652                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3653                               data_end - total_data, btrfs_leaf_data(leaf) +
3654                               data_end, old_data - data_end);
3655                 data_end = old_data;
3656         }
3657
3658         /* setup the item for the new data */
3659         for (i = 0; i < nr; i++) {
3660                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3661                 btrfs_set_item_key(leaf, &disk_key, slot + i);
3662                 item = btrfs_item_nr(leaf, slot + i);
3663                 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3664                 data_end -= data_size[i];
3665                 btrfs_set_item_size(leaf, item, data_size[i]);
3666         }
3667
3668         btrfs_set_header_nritems(leaf, nritems + nr);
3669
3670         ret = 0;
3671         if (slot == 0) {
3672                 struct btrfs_disk_key disk_key;
3673                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3674                 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3675         }
3676         btrfs_unlock_up_safe(path, 1);
3677         btrfs_mark_buffer_dirty(leaf);
3678
3679         if (btrfs_leaf_free_space(root, leaf) < 0) {
3680                 btrfs_print_leaf(root, leaf);
3681                 BUG();
3682         }
3683         return ret;
3684 }
3685
3686 /*
3687  * Given a key and some data, insert items into the tree.
3688  * This does all the path init required, making room in the tree if needed.
3689  */
3690 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3691                             struct btrfs_root *root,
3692                             struct btrfs_path *path,
3693                             struct btrfs_key *cpu_key, u32 *data_size,
3694                             int nr)
3695 {
3696         struct extent_buffer *leaf;
3697         int ret = 0;
3698         int slot;
3699         int i;
3700         u32 total_size = 0;
3701         u32 total_data = 0;
3702
3703         for (i = 0; i < nr; i++)
3704                 total_data += data_size[i];
3705
3706         total_size = total_data + (nr * sizeof(struct btrfs_item));
3707         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3708         if (ret == 0)
3709                 return -EEXIST;
3710         if (ret < 0)
3711                 goto out;
3712
3713         leaf = path->nodes[0];
3714         slot = path->slots[0];
3715         BUG_ON(slot < 0);
3716
3717         ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3718                                total_data, total_size, nr);
3719
3720 out:
3721         return ret;
3722 }
3723
3724 /*
3725  * Given a key and some data, insert an item into the tree.
3726  * This does all the path init required, making room in the tree if needed.
3727  */
3728 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3729                       *root, struct btrfs_key *cpu_key, void *data, u32
3730                       data_size)
3731 {
3732         int ret = 0;
3733         struct btrfs_path *path;
3734         struct extent_buffer *leaf;
3735         unsigned long ptr;
3736
3737         path = btrfs_alloc_path();
3738         BUG_ON(!path);
3739         ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3740         if (!ret) {
3741                 leaf = path->nodes[0];
3742                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3743                 write_extent_buffer(leaf, data, ptr, data_size);
3744                 btrfs_mark_buffer_dirty(leaf);
3745         }
3746         btrfs_free_path(path);
3747         return ret;
3748 }
3749
3750 /*
3751  * delete the pointer from a given node.
3752  *
3753  * the tree should have been previously balanced so the deletion does not
3754  * empty a node.
3755  */
3756 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3757                    struct btrfs_path *path, int level, int slot)
3758 {
3759         struct extent_buffer *parent = path->nodes[level];
3760         u32 nritems;
3761         int ret = 0;
3762         int wret;
3763
3764         nritems = btrfs_header_nritems(parent);
3765         if (slot != nritems - 1) {
3766                 memmove_extent_buffer(parent,
3767                               btrfs_node_key_ptr_offset(slot),
3768                               btrfs_node_key_ptr_offset(slot + 1),
3769                               sizeof(struct btrfs_key_ptr) *
3770                               (nritems - slot - 1));
3771         }
3772         nritems--;
3773         btrfs_set_header_nritems(parent, nritems);
3774         if (nritems == 0 && parent == root->node) {
3775                 BUG_ON(btrfs_header_level(root->node) != 1);
3776                 /* just turn the root into a leaf and break */
3777                 btrfs_set_header_level(root->node, 0);
3778         } else if (slot == 0) {
3779                 struct btrfs_disk_key disk_key;
3780
3781                 btrfs_node_key(parent, &disk_key, 0);
3782                 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3783                 if (wret)
3784                         ret = wret;
3785         }
3786         btrfs_mark_buffer_dirty(parent);
3787         return ret;
3788 }
3789
3790 /*
3791  * a helper function to delete the leaf pointed to by path->slots[1] and
3792  * path->nodes[1].
3793  *
3794  * This deletes the pointer in path->nodes[1] and frees the leaf
3795  * block extent.  zero is returned if it all worked out, < 0 otherwise.
3796  *
3797  * The path must have already been setup for deleting the leaf, including
3798  * all the proper balancing.  path->nodes[1] must be locked.
3799  */
3800 static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3801                                    struct btrfs_root *root,
3802                                    struct btrfs_path *path,
3803                                    struct extent_buffer *leaf)
3804 {
3805         int ret;
3806
3807         WARN_ON(btrfs_header_generation(leaf) != trans->transid);
3808         ret = del_ptr(trans, root, path, 1, path->slots[1]);
3809         if (ret)
3810                 return ret;
3811
3812         /*
3813          * btrfs_free_extent is expensive, we want to make sure we
3814          * aren't holding any locks when we call it
3815          */
3816         btrfs_unlock_up_safe(path, 0);
3817
3818         root_sub_used(root, leaf->len);
3819
3820         btrfs_free_tree_block(trans, root, leaf, 0, 1);
3821         return 0;
3822 }
3823 /*
3824  * delete the item at the leaf level in path.  If that empties
3825  * the leaf, remove it from the tree
3826  */
3827 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3828                     struct btrfs_path *path, int slot, int nr)
3829 {
3830         struct extent_buffer *leaf;
3831         struct btrfs_item *item;
3832         int last_off;
3833         int dsize = 0;
3834         int ret = 0;
3835         int wret;
3836         int i;
3837         u32 nritems;
3838
3839         leaf = path->nodes[0];
3840         last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3841
3842         for (i = 0; i < nr; i++)
3843                 dsize += btrfs_item_size_nr(leaf, slot + i);
3844
3845         nritems = btrfs_header_nritems(leaf);
3846
3847         if (slot + nr != nritems) {
3848                 int data_end = leaf_data_end(root, leaf);
3849
3850                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3851                               data_end + dsize,
3852                               btrfs_leaf_data(leaf) + data_end,
3853                               last_off - data_end);
3854
3855                 for (i = slot + nr; i < nritems; i++) {
3856                         u32 ioff;
3857
3858                         item = btrfs_item_nr(leaf, i);
3859                         if (!leaf->map_token) {
3860                                 map_extent_buffer(leaf, (unsigned long)item,
3861                                         sizeof(struct btrfs_item),
3862                                         &leaf->map_token, &leaf->kaddr,
3863                                         &leaf->map_start, &leaf->map_len,
3864                                         KM_USER1);
3865                         }
3866                         ioff = btrfs_item_offset(leaf, item);
3867                         btrfs_set_item_offset(leaf, item, ioff + dsize);
3868                 }
3869
3870                 if (leaf->map_token) {
3871                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3872                         leaf->map_token = NULL;
3873                 }
3874
3875                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3876                               btrfs_item_nr_offset(slot + nr),
3877                               sizeof(struct btrfs_item) *
3878                               (nritems - slot - nr));
3879         }
3880         btrfs_set_header_nritems(leaf, nritems - nr);
3881         nritems -= nr;
3882
3883         /* delete the leaf if we've emptied it */
3884         if (nritems == 0) {
3885                 if (leaf == root->node) {
3886                         btrfs_set_header_level(leaf, 0);
3887                 } else {
3888                         btrfs_set_path_blocking(path);
3889                         clean_tree_block(trans, root, leaf);
3890                         ret = btrfs_del_leaf(trans, root, path, leaf);
3891                         BUG_ON(ret);
3892                 }
3893         } else {
3894                 int used = leaf_space_used(leaf, 0, nritems);
3895                 if (slot == 0) {
3896                         struct btrfs_disk_key disk_key;
3897
3898                         btrfs_item_key(leaf, &disk_key, 0);
3899                         wret = fixup_low_keys(trans, root, path,
3900                                               &disk_key, 1);
3901                         if (wret)
3902                                 ret = wret;
3903                 }
3904
3905                 /* delete the leaf if it is mostly empty */
3906                 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
3907                         /* push_leaf_left fixes the path.
3908                          * make sure the path still points to our leaf
3909                          * for possible call to del_ptr below
3910                          */
3911                         slot = path->slots[1];
3912                         extent_buffer_get(leaf);
3913
3914                         btrfs_set_path_blocking(path);
3915                         wret = push_leaf_left(trans, root, path, 1, 1);
3916                         if (wret < 0 && wret != -ENOSPC)
3917                                 ret = wret;
3918
3919                         if (path->nodes[0] == leaf &&
3920                             btrfs_header_nritems(leaf)) {
3921                                 wret = push_leaf_right(trans, root, path, 1, 1);
3922                                 if (wret < 0 && wret != -ENOSPC)
3923                                         ret = wret;
3924                         }
3925
3926                         if (btrfs_header_nritems(leaf) == 0) {
3927                                 path->slots[1] = slot;
3928                                 ret = btrfs_del_leaf(trans, root, path, leaf);
3929                                 BUG_ON(ret);
3930                                 free_extent_buffer(leaf);
3931                         } else {
3932                                 /* if we're still in the path, make sure
3933                                  * we're dirty.  Otherwise, one of the
3934                                  * push_leaf functions must have already
3935                                  * dirtied this buffer
3936                                  */
3937                                 if (path->nodes[0] == leaf)
3938                                         btrfs_mark_buffer_dirty(leaf);
3939                                 free_extent_buffer(leaf);
3940                         }
3941                 } else {
3942                         btrfs_mark_buffer_dirty(leaf);
3943                 }
3944         }
3945         return ret;
3946 }
3947
3948 /*
3949  * search the tree again to find a leaf with lesser keys
3950  * returns 0 if it found something or 1 if there are no lesser leaves.
3951  * returns < 0 on io errors.
3952  *
3953  * This may release the path, and so you may lose any locks held at the
3954  * time you call it.
3955  */
3956 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3957 {
3958         struct btrfs_key key;
3959         struct btrfs_disk_key found_key;
3960         int ret;
3961
3962         btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3963
3964         if (key.offset > 0)
3965                 key.offset--;
3966         else if (key.type > 0)
3967                 key.type--;
3968         else if (key.objectid > 0)
3969                 key.objectid--;
3970         else
3971                 return 1;
3972
3973         btrfs_release_path(root, path);
3974         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3975         if (ret < 0)
3976                 return ret;
3977         btrfs_item_key(path->nodes[0], &found_key, 0);
3978         ret = comp_keys(&found_key, &key);
3979         if (ret < 0)
3980                 return 0;
3981         return 1;
3982 }
3983
3984 /*
3985  * A helper function to walk down the tree starting at min_key, and looking
3986  * for nodes or leaves that are either in cache or have a minimum
3987  * transaction id.  This is used by the btree defrag code, and tree logging
3988  *
3989  * This does not cow, but it does stuff the starting key it finds back
3990  * into min_key, so you can call btrfs_search_slot with cow=1 on the
3991  * key and get a writable path.
3992  *
3993  * This does lock as it descends, and path->keep_locks should be set
3994  * to 1 by the caller.
3995  *
3996  * This honors path->lowest_level to prevent descent past a given level
3997  * of the tree.
3998  *
3999  * min_trans indicates the oldest transaction that you are interested
4000  * in walking through.  Any nodes or leaves older than min_trans are
4001  * skipped over (without reading them).
4002  *
4003  * returns zero if something useful was found, < 0 on error and 1 if there
4004  * was nothing in the tree that matched the search criteria.
4005  */
4006 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
4007                          struct btrfs_key *max_key,
4008                          struct btrfs_path *path, int cache_only,
4009                          u64 min_trans)
4010 {
4011         struct extent_buffer *cur;
4012         struct btrfs_key found_key;
4013         int slot;
4014         int sret;
4015         u32 nritems;
4016         int level;
4017         int ret = 1;
4018
4019         WARN_ON(!path->keep_locks);
4020 again:
4021         cur = btrfs_lock_root_node(root);
4022         level = btrfs_header_level(cur);
4023         WARN_ON(path->nodes[level]);
4024         path->nodes[level] = cur;
4025         path->locks[level] = 1;
4026
4027         if (btrfs_header_generation(cur) < min_trans) {
4028                 ret = 1;
4029                 goto out;
4030         }
4031         while (1) {
4032                 nritems = btrfs_header_nritems(cur);
4033                 level = btrfs_header_level(cur);
4034                 sret = bin_search(cur, min_key, level, &slot);
4035
4036                 /* at the lowest level, we're done, setup the path and exit */
4037                 if (level == path->lowest_level) {
4038                         if (slot >= nritems)
4039                                 goto find_next_key;
4040                         ret = 0;
4041                         path->slots[level] = slot;
4042                         btrfs_item_key_to_cpu(cur, &found_key, slot);
4043                         goto out;
4044                 }
4045                 if (sret && slot > 0)
4046                         slot--;
4047                 /*
4048                  * check this node pointer against the cache_only and
4049                  * min_trans parameters.  If it isn't in cache or is too
4050                  * old, skip to the next one.
4051                  */
4052                 while (slot < nritems) {
4053                         u64 blockptr;
4054                         u64 gen;
4055                         struct extent_buffer *tmp;
4056                         struct btrfs_disk_key disk_key;
4057
4058                         blockptr = btrfs_node_blockptr(cur, slot);
4059                         gen = btrfs_node_ptr_generation(cur, slot);
4060                         if (gen < min_trans) {
4061                                 slot++;
4062                                 continue;
4063                         }
4064                         if (!cache_only)
4065                                 break;
4066
4067                         if (max_key) {
4068                                 btrfs_node_key(cur, &disk_key, slot);
4069                                 if (comp_keys(&disk_key, max_key) >= 0) {
4070                                         ret = 1;
4071                                         goto out;
4072                                 }
4073                         }
4074
4075                         tmp = btrfs_find_tree_block(root, blockptr,
4076                                             btrfs_level_size(root, level - 1));
4077
4078                         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
4079                                 free_extent_buffer(tmp);
4080                                 break;
4081                         }
4082                         if (tmp)
4083                                 free_extent_buffer(tmp);
4084                         slot++;
4085                 }
4086 find_next_key:
4087                 /*
4088                  * we didn't find a candidate key in this node, walk forward
4089                  * and find another one
4090                  */
4091                 if (slot >= nritems) {
4092                         path->slots[level] = slot;
4093                         btrfs_set_path_blocking(path);
4094                         sret = btrfs_find_next_key(root, path, min_key, level,
4095                                                   cache_only, min_trans);
4096                         if (sret == 0) {
4097                                 btrfs_release_path(root, path);
4098                                 goto again;
4099                         } else {
4100                                 goto out;
4101                         }
4102                 }
4103                 /* save our key for returning back */
4104                 btrfs_node_key_to_cpu(cur, &found_key, slot);
4105                 path->slots[level] = slot;
4106                 if (level == path->lowest_level) {
4107                         ret = 0;
4108                         unlock_up(path, level, 1);
4109                         goto out;
4110                 }
4111                 btrfs_set_path_blocking(path);
4112                 cur = read_node_slot(root, cur, slot);
4113
4114                 btrfs_tree_lock(cur);
4115
4116                 path->locks[level - 1] = 1;
4117                 path->nodes[level - 1] = cur;
4118                 unlock_up(path, level, 1);
4119                 btrfs_clear_path_blocking(path, NULL);
4120         }
4121 out:
4122         if (ret == 0)
4123                 memcpy(min_key, &found_key, sizeof(found_key));
4124         btrfs_set_path_blocking(path);
4125         return ret;
4126 }
4127
4128 /*
4129  * this is similar to btrfs_next_leaf, but does not try to preserve
4130  * and fixup the path.  It looks for and returns the next key in the
4131  * tree based on the current path and the cache_only and min_trans
4132  * parameters.
4133  *
4134  * 0 is returned if another key is found, < 0 if there are any errors
4135  * and 1 is returned if there are no higher keys in the tree
4136  *
4137  * path->keep_locks should be set to 1 on the search made before
4138  * calling this function.
4139  */
4140 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4141                         struct btrfs_key *key, int level,
4142                         int cache_only, u64 min_trans)
4143 {
4144         int slot;
4145         struct extent_buffer *c;
4146
4147         WARN_ON(!path->keep_locks);
4148         while (level < BTRFS_MAX_LEVEL) {
4149                 if (!path->nodes[level])
4150                         return 1;
4151
4152                 slot = path->slots[level] + 1;
4153                 c = path->nodes[level];
4154 next:
4155                 if (slot >= btrfs_header_nritems(c)) {
4156                         int ret;
4157                         int orig_lowest;
4158                         struct btrfs_key cur_key;
4159                         if (level + 1 >= BTRFS_MAX_LEVEL ||
4160                             !path->nodes[level + 1])
4161                                 return 1;
4162
4163                         if (path->locks[level + 1]) {
4164                                 level++;
4165                                 continue;
4166                         }
4167
4168                         slot = btrfs_header_nritems(c) - 1;
4169                         if (level == 0)
4170                                 btrfs_item_key_to_cpu(c, &cur_key, slot);
4171                         else
4172                                 btrfs_node_key_to_cpu(c, &cur_key, slot);
4173
4174                         orig_lowest = path->lowest_level;
4175                         btrfs_release_path(root, path);
4176                         path->lowest_level = level;
4177                         ret = btrfs_search_slot(NULL, root, &cur_key, path,
4178                                                 0, 0);
4179                         path->lowest_level = orig_lowest;
4180                         if (ret < 0)
4181                                 return ret;
4182
4183                         c = path->nodes[level];
4184                         slot = path->slots[level];
4185                         if (ret == 0)
4186                                 slot++;
4187                         goto next;
4188                 }
4189
4190                 if (level == 0)
4191                         btrfs_item_key_to_cpu(c, key, slot);
4192                 else {
4193                         u64 blockptr = btrfs_node_blockptr(c, slot);
4194                         u64 gen = btrfs_node_ptr_generation(c, slot);
4195
4196                         if (cache_only) {
4197                                 struct extent_buffer *cur;
4198                                 cur = btrfs_find_tree_block(root, blockptr,
4199                                             btrfs_level_size(root, level - 1));
4200                                 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4201                                         slot++;
4202                                         if (cur)
4203                                                 free_extent_buffer(cur);
4204                                         goto next;
4205                                 }
4206                                 free_extent_buffer(cur);
4207                         }
4208                         if (gen < min_trans) {
4209                                 slot++;
4210                                 goto next;
4211                         }
4212                         btrfs_node_key_to_cpu(c, key, slot);
4213                 }
4214                 return 0;
4215         }
4216         return 1;
4217 }
4218
4219 /*
4220  * search the tree again to find a leaf with greater keys
4221  * returns 0 if it found something or 1 if there are no greater leaves.
4222  * returns < 0 on io errors.
4223  */
4224 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4225 {
4226         int slot;
4227         int level;
4228         struct extent_buffer *c;
4229         struct extent_buffer *next;
4230         struct btrfs_key key;
4231         u32 nritems;
4232         int ret;
4233         int old_spinning = path->leave_spinning;
4234         int force_blocking = 0;
4235
4236         nritems = btrfs_header_nritems(path->nodes[0]);
4237         if (nritems == 0)
4238                 return 1;
4239
4240         /*
4241          * we take the blocks in an order that upsets lockdep.  Using
4242          * blocking mode is the only way around it.
4243          */
4244 #ifdef CONFIG_DEBUG_LOCK_ALLOC
4245         force_blocking = 1;
4246 #endif
4247
4248         btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4249 again:
4250         level = 1;
4251         next = NULL;
4252         btrfs_release_path(root, path);
4253
4254         path->keep_locks = 1;
4255
4256         if (!force_blocking)
4257                 path->leave_spinning = 1;
4258
4259         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4260         path->keep_locks = 0;
4261
4262         if (ret < 0)
4263                 return ret;
4264
4265         nritems = btrfs_header_nritems(path->nodes[0]);
4266         /*
4267          * by releasing the path above we dropped all our locks.  A balance
4268          * could have added more items next to the key that used to be
4269          * at the very end of the block.  So, check again here and
4270          * advance the path if there are now more items available.
4271          */
4272         if (nritems > 0 && path->slots[0] < nritems - 1) {
4273                 if (ret == 0)
4274                         path->slots[0]++;
4275                 ret = 0;
4276                 goto done;
4277         }
4278
4279         while (level < BTRFS_MAX_LEVEL) {
4280                 if (!path->nodes[level]) {
4281                         ret = 1;
4282                         goto done;
4283                 }
4284
4285                 slot = path->slots[level] + 1;
4286                 c = path->nodes[level];
4287                 if (slot >= btrfs_header_nritems(c)) {
4288                         level++;
4289                         if (level == BTRFS_MAX_LEVEL) {
4290                                 ret = 1;
4291                                 goto done;
4292                         }
4293                         continue;
4294                 }
4295
4296                 if (next) {
4297                         btrfs_tree_unlock(next);
4298                         free_extent_buffer(next);
4299                 }
4300
4301                 next = c;
4302                 ret = read_block_for_search(NULL, root, path, &next, level,
4303                                             slot, &key);
4304                 if (ret == -EAGAIN)
4305                         goto again;
4306
4307                 if (ret < 0) {
4308                         btrfs_release_path(root, path);
4309                         goto done;
4310                 }
4311
4312                 if (!path->skip_locking) {
4313                         ret = btrfs_try_spin_lock(next);
4314                         if (!ret) {
4315                                 btrfs_set_path_blocking(path);
4316                                 btrfs_tree_lock(next);
4317                                 if (!force_blocking)
4318                                         btrfs_clear_path_blocking(path, next);
4319                         }
4320                         if (force_blocking)
4321                                 btrfs_set_lock_blocking(next);
4322                 }
4323                 break;
4324         }
4325         path->slots[level] = slot;
4326         while (1) {
4327                 level--;
4328                 c = path->nodes[level];
4329                 if (path->locks[level])
4330                         btrfs_tree_unlock(c);
4331
4332                 free_extent_buffer(c);
4333                 path->nodes[level] = next;
4334                 path->slots[level] = 0;
4335                 if (!path->skip_locking)
4336                         path->locks[level] = 1;
4337
4338                 if (!level)
4339                         break;
4340
4341                 ret = read_block_for_search(NULL, root, path, &next, level,
4342                                             0, &key);
4343                 if (ret == -EAGAIN)
4344                         goto again;
4345
4346                 if (ret < 0) {
4347                         btrfs_release_path(root, path);
4348                         goto done;
4349                 }
4350
4351                 if (!path->skip_locking) {
4352                         btrfs_assert_tree_locked(path->nodes[level]);
4353                         ret = btrfs_try_spin_lock(next);
4354                         if (!ret) {
4355                                 btrfs_set_path_blocking(path);
4356                                 btrfs_tree_lock(next);
4357                                 if (!force_blocking)
4358                                         btrfs_clear_path_blocking(path, next);
4359                         }
4360                         if (force_blocking)
4361                                 btrfs_set_lock_blocking(next);
4362                 }
4363         }
4364         ret = 0;
4365 done:
4366         unlock_up(path, 0, 1);
4367         path->leave_spinning = old_spinning;
4368         if (!old_spinning)
4369                 btrfs_set_path_blocking(path);
4370
4371         return ret;
4372 }
4373
4374 /*
4375  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4376  * searching until it gets past min_objectid or finds an item of 'type'
4377  *
4378  * returns 0 if something is found, 1 if nothing was found and < 0 on error
4379  */
4380 int btrfs_previous_item(struct btrfs_root *root,
4381                         struct btrfs_path *path, u64 min_objectid,
4382                         int type)
4383 {
4384         struct btrfs_key found_key;
4385         struct extent_buffer *leaf;
4386         u32 nritems;
4387         int ret;
4388
4389         while (1) {
4390                 if (path->slots[0] == 0) {
4391                         btrfs_set_path_blocking(path);
4392                         ret = btrfs_prev_leaf(root, path);
4393                         if (ret != 0)
4394                                 return ret;
4395                 } else {
4396                         path->slots[0]--;
4397                 }
4398                 leaf = path->nodes[0];
4399                 nritems = btrfs_header_nritems(leaf);
4400                 if (nritems == 0)
4401                         return 1;
4402                 if (path->slots[0] == nritems)
4403                         path->slots[0]--;
4404
4405                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4406                 if (found_key.objectid < min_objectid)
4407                         break;
4408                 if (found_key.type == type)
4409                         return 0;
4410                 if (found_key.objectid == min_objectid &&
4411                     found_key.type < type)
4412                         break;
4413         }
4414         return 1;
4415 }