fs/btrfs/ctree.c

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