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 (trans->transaction->delayed_refs.flushing &&
 953             btrfs_header_nritems(mid) > 2)
 954                 return 0;
 955
 956         if (btrfs_header_nritems(mid) < 2)
 957                 err_on_enospc = 1;
 958
 959         left = read_node_slot(root, parent, pslot - 1);
 960         if (left) {
 961                 btrfs_tree_lock(left);
 962                 btrfs_set_lock_blocking(left);
 963                 wret = btrfs_cow_block(trans, root, left,
 964                                        parent, pslot - 1, &left);
 965                 if (wret) {
 966                         ret = wret;
 967                         goto enospc;
 968                 }
 969         }
 970         right = read_node_slot(root, parent, pslot + 1);
 971         if (right) {
 972                 btrfs_tree_lock(right);
 973                 btrfs_set_lock_blocking(right);
 974                 wret = btrfs_cow_block(trans, root, right,
 975                                        parent, pslot + 1, &right);
 976                 if (wret) {
 977                         ret = wret;
 978                         goto enospc;
 979                 }
 980         }
 981
 982         /* first, try to make some room in the middle buffer */
 983         if (left) {
 984                 orig_slot += btrfs_header_nritems(left);
 985                 wret = push_node_left(trans, root, left, mid, 1);
 986                 if (wret < 0)
 987                         ret = wret;
 988                 if (btrfs_header_nritems(mid) < 2)
 989                         err_on_enospc = 1;
 990         }
 991
 992         /*
 993          * then try to empty the right most buffer into the middle
 994          */
 995         if (right) {
 996                 wret = push_node_left(trans, root, mid, right, 1);
 997                 if (wret < 0 && wret != -ENOSPC)
 998                         ret = wret;
 999                 if (btrfs_header_nritems(right) == 0) {
1000                         u64 bytenr = right->start;
1001                         u64 generation = btrfs_header_generation(parent);
1002                         u32 blocksize = right->len;
1003
1004                         clean_tree_block(trans, root, right);
1005                         btrfs_tree_unlock(right);
1006                         free_extent_buffer(right);
1007                         right = NULL;
1008                         wret = del_ptr(trans, root, path, level + 1, pslot +
1009                                        1);
1010                         if (wret)
1011                                 ret = wret;
1012                         wret = btrfs_free_extent(trans, root, bytenr,
1013                                                  blocksize, parent->start,
1014                                                  btrfs_header_owner(parent),
1015                                                  generation, level, 1);
1016                         if (wret)
1017                                 ret = wret;
1018                 } else {
1019                         struct btrfs_disk_key right_key;
1020                         btrfs_node_key(right, &right_key, 0);
1021                         btrfs_set_node_key(parent, &right_key, pslot + 1);
1022                         btrfs_mark_buffer_dirty(parent);
1023                 }
1024         }
1025         if (btrfs_header_nritems(mid) == 1) {
1026                 /*
1027                  * we're not allowed to leave a node with one item in the
1028                  * tree during a delete.  A deletion from lower in the tree
1029                  * could try to delete the only pointer in this node.
1030                  * So, pull some keys from the left.
1031                  * There has to be a left pointer at this point because
1032                  * otherwise we would have pulled some pointers from the
1033                  * right
1034                  */
1035                 BUG_ON(!left);
1036                 wret = balance_node_right(trans, root, mid, left);
1037                 if (wret < 0) {
1038                         ret = wret;
1039                         goto enospc;
1040                 }
1041                 if (wret == 1) {
1042                         wret = push_node_left(trans, root, left, mid, 1);
1043                         if (wret < 0)
1044                                 ret = wret;
1045                 }
1046                 BUG_ON(wret == 1);
1047         }
1048         if (btrfs_header_nritems(mid) == 0) {
1049                 /* we've managed to empty the middle node, drop it */
1050                 u64 root_gen = btrfs_header_generation(parent);
1051                 u64 bytenr = mid->start;
1052                 u32 blocksize = mid->len;
1053
1054                 clean_tree_block(trans, root, mid);
1055                 btrfs_tree_unlock(mid);
1056                 free_extent_buffer(mid);
1057                 mid = NULL;
1058                 wret = del_ptr(trans, root, path, level + 1, pslot);
1059                 if (wret)
1060                         ret = wret;
1061                 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
1062                                          parent->start,
1063                                          btrfs_header_owner(parent),
1064                                          root_gen, level, 1);
1065                 if (wret)
1066                         ret = wret;
1067         } else {
1068                 /* update the parent key to reflect our changes */
1069                 struct btrfs_disk_key mid_key;
1070                 btrfs_node_key(mid, &mid_key, 0);
1071                 btrfs_set_node_key(parent, &mid_key, pslot);
1072                 btrfs_mark_buffer_dirty(parent);
1073         }
1074
1075         /* update the path */
1076         if (left) {
1077                 if (btrfs_header_nritems(left) > orig_slot) {
1078                         extent_buffer_get(left);
1079                         /* left was locked after cow */
1080                         path->nodes[level] = left;
1081                         path->slots[level + 1] -= 1;
1082                         path->slots[level] = orig_slot;
1083                         if (mid) {
1084                                 btrfs_tree_unlock(mid);
1085                                 free_extent_buffer(mid);
1086                         }
1087                 } else {
1088                         orig_slot -= btrfs_header_nritems(left);
1089                         path->slots[level] = orig_slot;
1090                 }
1091         }
1092         /* double check we haven't messed things up */
1093         check_block(root, path, level);
1094         if (orig_ptr !=
1095             btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1096                 BUG();
1097 enospc:
1098         if (right) {
1099                 btrfs_tree_unlock(right);
1100                 free_extent_buffer(right);
1101         }
1102         if (left) {
1103                 if (path->nodes[level] != left)
1104                         btrfs_tree_unlock(left);
1105                 free_extent_buffer(left);
1106         }
1107         return ret;
1108 }
1109
1110 /* Node balancing for insertion.  Here we only split or push nodes around
1111  * when they are completely full.  This is also done top down, so we
1112  * have to be pessimistic.
1113  */
1114 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1115                                           struct btrfs_root *root,
1116                                           struct btrfs_path *path, int level)
1117 {
1118         struct extent_buffer *right = NULL;
1119         struct extent_buffer *mid;
1120         struct extent_buffer *left = NULL;
1121         struct extent_buffer *parent = NULL;
1122         int ret = 0;
1123         int wret;
1124         int pslot;
1125         int orig_slot = path->slots[level];
1126         u64 orig_ptr;
1127
1128         if (level == 0)
1129                 return 1;
1130
1131         mid = path->nodes[level];
1132         WARN_ON(btrfs_header_generation(mid) != trans->transid);
1133         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1134
1135         if (level < BTRFS_MAX_LEVEL - 1)
1136                 parent = path->nodes[level + 1];
1137         pslot = path->slots[level + 1];
1138
1139         if (!parent)
1140                 return 1;
1141
1142         left = read_node_slot(root, parent, pslot - 1);
1143
1144         /* first, try to make some room in the middle buffer */
1145         if (left) {
1146                 u32 left_nr;
1147
1148                 btrfs_tree_lock(left);
1149                 btrfs_set_lock_blocking(left);
1150
1151                 left_nr = btrfs_header_nritems(left);
1152                 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1153                         wret = 1;
1154                 } else {
1155                         ret = btrfs_cow_block(trans, root, left, parent,
1156                                               pslot - 1, &left);
1157                         if (ret)
1158                                 wret = 1;
1159                         else {
1160                                 wret = push_node_left(trans, root,
1161                                                       left, mid, 0);
1162                         }
1163                 }
1164                 if (wret < 0)
1165                         ret = wret;
1166                 if (wret == 0) {
1167                         struct btrfs_disk_key disk_key;
1168                         orig_slot += left_nr;
1169                         btrfs_node_key(mid, &disk_key, 0);
1170                         btrfs_set_node_key(parent, &disk_key, pslot);
1171                         btrfs_mark_buffer_dirty(parent);
1172                         if (btrfs_header_nritems(left) > orig_slot) {
1173                                 path->nodes[level] = left;
1174                                 path->slots[level + 1] -= 1;
1175                                 path->slots[level] = orig_slot;
1176                                 btrfs_tree_unlock(mid);
1177                                 free_extent_buffer(mid);
1178                         } else {
1179                                 orig_slot -=
1180                                         btrfs_header_nritems(left);
1181                                 path->slots[level] = orig_slot;
1182                                 btrfs_tree_unlock(left);
1183                                 free_extent_buffer(left);
1184                         }
1185                         return 0;
1186                 }
1187                 btrfs_tree_unlock(left);
1188                 free_extent_buffer(left);
1189         }
1190         right = read_node_slot(root, parent, pslot + 1);
1191
1192         /*
1193          * then try to empty the right most buffer into the middle
1194          */
1195         if (right) {
1196                 u32 right_nr;
1197
1198                 btrfs_tree_lock(right);
1199                 btrfs_set_lock_blocking(right);
1200
1201                 right_nr = btrfs_header_nritems(right);
1202                 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1203                         wret = 1;
1204                 } else {
1205                         ret = btrfs_cow_block(trans, root, right,
1206                                               parent, pslot + 1,
1207                                               &right);
1208                         if (ret)
1209                                 wret = 1;
1210                         else {
1211                                 wret = balance_node_right(trans, root,
1212                                                           right, mid);
1213                         }
1214                 }
1215                 if (wret < 0)
1216                         ret = wret;
1217                 if (wret == 0) {
1218                         struct btrfs_disk_key disk_key;
1219
1220                         btrfs_node_key(right, &disk_key, 0);
1221                         btrfs_set_node_key(parent, &disk_key, pslot + 1);
1222                         btrfs_mark_buffer_dirty(parent);
1223
1224                         if (btrfs_header_nritems(mid) <= orig_slot) {
1225                                 path->nodes[level] = right;
1226                                 path->slots[level + 1] += 1;
1227                                 path->slots[level] = orig_slot -
1228                                         btrfs_header_nritems(mid);
1229                                 btrfs_tree_unlock(mid);
1230                                 free_extent_buffer(mid);
1231                         } else {
1232                                 btrfs_tree_unlock(right);
1233                                 free_extent_buffer(right);
1234                         }
1235                         return 0;
1236                 }
1237                 btrfs_tree_unlock(right);
1238                 free_extent_buffer(right);
1239         }
1240         return 1;
1241 }
1242
1243 /*
1244  * readahead one full node of leaves, finding things that are close
1245  * to the block in 'slot', and triggering ra on them.
1246  */
1247 static noinline void reada_for_search(struct btrfs_root *root,
1248                                       struct btrfs_path *path,
1249                                       int level, int slot, u64 objectid)
1250 {
1251         struct extent_buffer *node;
1252         struct btrfs_disk_key disk_key;
1253         u32 nritems;
1254         u64 search;
1255         u64 target;
1256         u64 nread = 0;
1257         int direction = path->reada;
1258         struct extent_buffer *eb;
1259         u32 nr;
1260         u32 blocksize;
1261         u32 nscan = 0;
1262
1263         if (level != 1)
1264                 return;
1265
1266         if (!path->nodes[level])
1267                 return;
1268
1269         node = path->nodes[level];
1270
1271         search = btrfs_node_blockptr(node, slot);
1272         blocksize = btrfs_level_size(root, level - 1);
1273         eb = btrfs_find_tree_block(root, search, blocksize);
1274         if (eb) {
1275                 free_extent_buffer(eb);
1276                 return;
1277         }
1278
1279         target = search;
1280
1281         nritems = btrfs_header_nritems(node);
1282         nr = slot;
1283         while (1) {
1284                 if (direction < 0) {
1285                         if (nr == 0)
1286                                 break;
1287                         nr--;
1288                 } else if (direction > 0) {
1289                         nr++;
1290                         if (nr >= nritems)
1291                                 break;
1292                 }
1293                 if (path->reada < 0 && objectid) {
1294                         btrfs_node_key(node, &disk_key, nr);
1295                         if (btrfs_disk_key_objectid(&disk_key) != objectid)
1296                                 break;
1297                 }
1298                 search = btrfs_node_blockptr(node, nr);
1299                 if ((search <= target && target - search <= 65536) ||
1300                     (search > target && search - target <= 65536)) {
1301                         readahead_tree_block(root, search, blocksize,
1302                                      btrfs_node_ptr_generation(node, nr));
1303                         nread += blocksize;
1304                 }
1305                 nscan++;
1306                 if ((nread > 65536 || nscan > 32))
1307                         break;
1308         }
1309 }
1310
1311 /*
1312  * returns -EAGAIN if it had to drop the path, or zero if everything was in
1313  * cache
1314  */
1315 static noinline int reada_for_balance(struct btrfs_root *root,
1316                                       struct btrfs_path *path, int level)
1317 {
1318         int slot;
1319         int nritems;
1320         struct extent_buffer *parent;
1321         struct extent_buffer *eb;
1322         u64 gen;
1323         u64 block1 = 0;
1324         u64 block2 = 0;
1325         int ret = 0;
1326         int blocksize;
1327
1328         parent = path->nodes[level - 1];
1329         if (!parent)
1330                 return 0;
1331
1332         nritems = btrfs_header_nritems(parent);
1333         slot = path->slots[level];
1334         blocksize = btrfs_level_size(root, level);
1335
1336         if (slot > 0) {
1337                 block1 = btrfs_node_blockptr(parent, slot - 1);
1338                 gen = btrfs_node_ptr_generation(parent, slot - 1);
1339                 eb = btrfs_find_tree_block(root, block1, blocksize);
1340                 if (eb && btrfs_buffer_uptodate(eb, gen))
1341                         block1 = 0;
1342                 free_extent_buffer(eb);
1343         }
1344         if (slot < nritems) {
1345                 block2 = btrfs_node_blockptr(parent, slot + 1);
1346                 gen = btrfs_node_ptr_generation(parent, slot + 1);
1347                 eb = btrfs_find_tree_block(root, block2, blocksize);
1348                 if (eb && btrfs_buffer_uptodate(eb, gen))
1349                         block2 = 0;
1350                 free_extent_buffer(eb);
1351         }
1352         if (block1 || block2) {
1353                 ret = -EAGAIN;
1354                 btrfs_release_path(root, path);
1355                 if (block1)
1356                         readahead_tree_block(root, block1, blocksize, 0);
1357                 if (block2)
1358                         readahead_tree_block(root, block2, blocksize, 0);
1359
1360                 if (block1) {
1361                         eb = read_tree_block(root, block1, blocksize, 0);
1362                         free_extent_buffer(eb);
1363                 }
1364                 if (block1) {
1365                         eb = read_tree_block(root, block2, blocksize, 0);
1366                         free_extent_buffer(eb);
1367                 }
1368         }
1369         return ret;
1370 }
1371
1372
1373 /*
1374  * when we walk down the tree, it is usually safe to unlock the higher layers
1375  * in the tree.  The exceptions are when our path goes through slot 0, because
1376  * operations on the tree might require changing key pointers higher up in the
1377  * tree.
1378  *
1379  * callers might also have set path->keep_locks, which tells this code to keep
1380  * the lock if the path points to the last slot in the block.  This is part of
1381  * walking through the tree, and selecting the next slot in the higher block.
1382  *
1383  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
1384  * if lowest_unlock is 1, level 0 won't be unlocked
1385  */
1386 static noinline void unlock_up(struct btrfs_path *path, int level,
1387                                int lowest_unlock)
1388 {
1389         int i;
1390         int skip_level = level;
1391         int no_skips = 0;
1392         struct extent_buffer *t;
1393
1394         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1395                 if (!path->nodes[i])
1396                         break;
1397                 if (!path->locks[i])
1398                         break;
1399                 if (!no_skips && path->slots[i] == 0) {
1400                         skip_level = i + 1;
1401                         continue;
1402                 }
1403                 if (!no_skips && path->keep_locks) {
1404                         u32 nritems;
1405                         t = path->nodes[i];
1406                         nritems = btrfs_header_nritems(t);
1407                         if (nritems < 1 || path->slots[i] >= nritems - 1) {
1408                                 skip_level = i + 1;
1409                                 continue;
1410                         }
1411                 }
1412                 if (skip_level < i && i >= lowest_unlock)
1413                         no_skips = 1;
1414
1415                 t = path->nodes[i];
1416                 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1417                         btrfs_tree_unlock(t);
1418                         path->locks[i] = 0;
1419                 }
1420         }
1421 }
1422
1423 /*
1424  * This releases any locks held in the path starting at level and
1425  * going all the way up to the root.
1426  *
1427  * btrfs_search_slot will keep the lock held on higher nodes in a few
1428  * corner cases, such as COW of the block at slot zero in the node.  This
1429  * ignores those rules, and it should only be called when there are no
1430  * more updates to be done higher up in the tree.
1431  */
1432 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1433 {
1434         int i;
1435
1436         if (path->keep_locks || path->lowest_level)
1437                 return;
1438
1439         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1440                 if (!path->nodes[i])
1441                         continue;
1442                 if (!path->locks[i])
1443                         continue;
1444                 btrfs_tree_unlock(path->nodes[i]);
1445                 path->locks[i] = 0;
1446         }
1447 }
1448
1449 /*
1450  * look for key in the tree.  path is filled in with nodes along the way
1451  * if key is found, we return zero and you can find the item in the leaf
1452  * level of the path (level 0)
1453  *
1454  * If the key isn't found, the path points to the slot where it should
1455  * be inserted, and 1 is returned.  If there are other errors during the
1456  * search a negative error number is returned.
1457  *
1458  * if ins_len > 0, nodes and leaves will be split as we walk down the
1459  * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
1460  * possible)
1461  */
1462 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1463                       *root, struct btrfs_key *key, struct btrfs_path *p, int
1464                       ins_len, int cow)
1465 {
1466         struct extent_buffer *b;
1467         struct extent_buffer *tmp;
1468         int slot;
1469         int ret;
1470         int level;
1471         int should_reada = p->reada;
1472         int lowest_unlock = 1;
1473         int blocksize;
1474         u8 lowest_level = 0;
1475         u64 blocknr;
1476         u64 gen;
1477
1478         lowest_level = p->lowest_level;
1479         WARN_ON(lowest_level && ins_len > 0);
1480         WARN_ON(p->nodes[0] != NULL);
1481
1482         if (ins_len < 0)
1483                 lowest_unlock = 2;
1484
1485 again:
1486         if (p->skip_locking)
1487                 b = btrfs_root_node(root);
1488         else
1489                 b = btrfs_lock_root_node(root);
1490
1491         while (b) {
1492                 level = btrfs_header_level(b);
1493
1494                 /*
1495                  * setup the path here so we can release it under lock
1496                  * contention with the cow code
1497                  */
1498                 p->nodes[level] = b;
1499                 if (!p->skip_locking)
1500                         p->locks[level] = 1;
1501
1502                 if (cow) {
1503                         int wret;
1504
1505                         /* is a cow on this block not required */
1506                         if (btrfs_header_generation(b) == trans->transid &&
1507                             btrfs_header_owner(b) == root->root_key.objectid &&
1508                             !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1509                                 goto cow_done;
1510                         }
1511                         btrfs_set_path_blocking(p);
1512
1513                         wret = btrfs_cow_block(trans, root, b,
1514                                                p->nodes[level + 1],
1515                                                p->slots[level + 1], &b);
1516                         if (wret) {
1517                                 free_extent_buffer(b);
1518                                 ret = wret;
1519                                 goto done;
1520                         }
1521                 }
1522 cow_done:
1523                 BUG_ON(!cow && ins_len);
1524                 if (level != btrfs_header_level(b))
1525                         WARN_ON(1);
1526                 level = btrfs_header_level(b);
1527
1528                 p->nodes[level] = b;
1529                 if (!p->skip_locking)
1530                         p->locks[level] = 1;
1531
1532                 btrfs_clear_path_blocking(p, NULL);
1533
1534                 /*
1535                  * we have a lock on b and as long as we aren't changing
1536                  * the tree, there is no way to for the items in b to change.
1537                  * It is safe to drop the lock on our parent before we
1538                  * go through the expensive btree search on b.
1539                  *
1540                  * If cow is true, then we might be changing slot zero,
1541                  * which may require changing the parent.  So, we can't
1542                  * drop the lock until after we know which slot we're
1543                  * operating on.
1544                  */
1545                 if (!cow)
1546                         btrfs_unlock_up_safe(p, level + 1);
1547
1548                 ret = check_block(root, p, level);
1549                 if (ret) {
1550                         ret = -1;
1551                         goto done;
1552                 }
1553
1554                 ret = bin_search(b, key, level, &slot);
1555
1556                 if (level != 0) {
1557                         if (ret && slot > 0)
1558                                 slot -= 1;
1559                         p->slots[level] = slot;
1560                         if ((p->search_for_split || ins_len > 0) &&
1561                             btrfs_header_nritems(b) >=
1562                             BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1563                                 int sret;
1564
1565                                 sret = reada_for_balance(root, p, level);
1566                                 if (sret)
1567                                         goto again;
1568
1569                                 btrfs_set_path_blocking(p);
1570                                 sret = split_node(trans, root, p, level);
1571                                 btrfs_clear_path_blocking(p, NULL);
1572
1573                                 BUG_ON(sret > 0);
1574                                 if (sret) {
1575                                         ret = sret;
1576                                         goto done;
1577                                 }
1578                                 b = p->nodes[level];
1579                                 slot = p->slots[level];
1580                         } else if (ins_len < 0 &&
1581                                    btrfs_header_nritems(b) <
1582                                    BTRFS_NODEPTRS_PER_BLOCK(root) / 4) {
1583                                 int sret;
1584
1585                                 sret = reada_for_balance(root, p, level);
1586                                 if (sret)
1587                                         goto again;
1588
1589                                 btrfs_set_path_blocking(p);
1590                                 sret = balance_level(trans, root, p, level);
1591                                 btrfs_clear_path_blocking(p, NULL);
1592
1593                                 if (sret) {
1594                                         ret = sret;
1595                                         goto done;
1596                                 }
1597                                 b = p->nodes[level];
1598                                 if (!b) {
1599                                         btrfs_release_path(NULL, p);
1600                                         goto again;
1601                                 }
1602                                 slot = p->slots[level];
1603                                 BUG_ON(btrfs_header_nritems(b) == 1);
1604                         }
1605                         unlock_up(p, level, lowest_unlock);
1606
1607                         /* this is only true while dropping a snapshot */
1608                         if (level == lowest_level) {
1609                                 ret = 0;
1610                                 goto done;
1611                         }
1612
1613                         blocknr = btrfs_node_blockptr(b, slot);
1614                         gen = btrfs_node_ptr_generation(b, slot);
1615                         blocksize = btrfs_level_size(root, level - 1);
1616
1617                         tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1618                         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1619                                 b = tmp;
1620                         } else {
1621                                 /*
1622                                  * reduce lock contention at high levels
1623                                  * of the btree by dropping locks before
1624                                  * we read.
1625                                  */
1626                                 if (level > 0) {
1627                                         btrfs_release_path(NULL, p);
1628                                         if (tmp)
1629                                                 free_extent_buffer(tmp);
1630                                         if (should_reada)
1631                                                 reada_for_search(root, p,
1632                                                                  level, slot,
1633                                                                  key->objectid);
1634
1635                                         tmp = read_tree_block(root, blocknr,
1636                                                          blocksize, gen);
1637                                         if (tmp)
1638                                                 free_extent_buffer(tmp);
1639                                         goto again;
1640                                 } else {
1641                                         btrfs_set_path_blocking(p);
1642                                         if (tmp)
1643                                                 free_extent_buffer(tmp);
1644                                         if (should_reada)
1645                                                 reada_for_search(root, p,
1646                                                                  level, slot,
1647                                                                  key->objectid);
1648                                         b = read_node_slot(root, b, slot);
1649                                 }
1650                         }
1651                         if (!p->skip_locking) {
1652                                 int lret;
1653
1654                                 btrfs_clear_path_blocking(p, NULL);
1655                                 lret = btrfs_try_spin_lock(b);
1656
1657                                 if (!lret) {
1658                                         btrfs_set_path_blocking(p);
1659                                         btrfs_tree_lock(b);
1660                                         btrfs_clear_path_blocking(p, b);
1661                                 }
1662                         }
1663                 } else {
1664                         p->slots[level] = slot;
1665                         if (ins_len > 0 &&
1666                             btrfs_leaf_free_space(root, b) < ins_len) {
1667                                 int sret;
1668
1669                                 btrfs_set_path_blocking(p);
1670                                 sret = split_leaf(trans, root, key,
1671                                                       p, ins_len, ret == 0);
1672                                 btrfs_clear_path_blocking(p, NULL);
1673
1674                                 BUG_ON(sret > 0);
1675                                 if (sret) {
1676                                         ret = sret;
1677                                         goto done;
1678                                 }
1679                         }
1680                         if (!p->search_for_split)
1681                                 unlock_up(p, level, lowest_unlock);
1682                         goto done;
1683                 }
1684         }
1685         ret = 1;
1686 done:
1687         /*
1688          * we don't really know what they plan on doing with the path
1689          * from here on, so for now just mark it as blocking
1690          */
1691         if (!p->leave_spinning)
1692                 btrfs_set_path_blocking(p);
1693         return ret;
1694 }
1695
1696 int btrfs_merge_path(struct btrfs_trans_handle *trans,
1697                      struct btrfs_root *root,
1698                      struct btrfs_key *node_keys,
1699                      u64 *nodes, int lowest_level)
1700 {
1701         struct extent_buffer *eb;
1702         struct extent_buffer *parent;
1703         struct btrfs_key key;
1704         u64 bytenr;
1705         u64 generation;
1706         u32 blocksize;
1707         int level;
1708         int slot;
1709         int key_match;
1710         int ret;
1711
1712         eb = btrfs_lock_root_node(root);
1713         ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb);
1714         BUG_ON(ret);
1715
1716         btrfs_set_lock_blocking(eb);
1717
1718         parent = eb;
1719         while (1) {
1720                 level = btrfs_header_level(parent);
1721                 if (level == 0 || level <= lowest_level)
1722                         break;
1723
1724                 ret = bin_search(parent, &node_keys[lowest_level], level,
1725                                  &slot);
1726                 if (ret && slot > 0)
1727                         slot--;
1728
1729                 bytenr = btrfs_node_blockptr(parent, slot);
1730                 if (nodes[level - 1] == bytenr)
1731                         break;
1732
1733                 blocksize = btrfs_level_size(root, level - 1);
1734                 generation = btrfs_node_ptr_generation(parent, slot);
1735                 btrfs_node_key_to_cpu(eb, &key, slot);
1736                 key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
1737
1738                 if (generation == trans->transid) {
1739                         eb = read_tree_block(root, bytenr, blocksize,
1740                                              generation);
1741                         btrfs_tree_lock(eb);
1742                         btrfs_set_lock_blocking(eb);
1743                 }
1744
1745                 /*
1746                  * if node keys match and node pointer hasn't been modified
1747                  * in the running transaction, we can merge the path. for
1748                  * blocks owened by reloc trees, the node pointer check is
1749                  * skipped, this is because these blocks are fully controlled
1750                  * by the space balance code, no one else can modify them.
1751                  */
1752                 if (!nodes[level - 1] || !key_match ||
1753                     (generation == trans->transid &&
1754                      btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
1755                         if (level == 1 || level == lowest_level + 1) {
1756                                 if (generation == trans->transid) {
1757                                         btrfs_tree_unlock(eb);
1758                                         free_extent_buffer(eb);
1759                                 }
1760                                 break;
1761                         }
1762
1763                         if (generation != trans->transid) {
1764                                 eb = read_tree_block(root, bytenr, blocksize,
1765                                                 generation);
1766                                 btrfs_tree_lock(eb);
1767                                 btrfs_set_lock_blocking(eb);
1768                         }
1769
1770                         ret = btrfs_cow_block(trans, root, eb, parent, slot,
1771                                               &eb);
1772                         BUG_ON(ret);
1773
1774                         if (root->root_key.objectid ==
1775                             BTRFS_TREE_RELOC_OBJECTID) {
1776                                 if (!nodes[level - 1]) {
1777                                         nodes[level - 1] = eb->start;
1778                                         memcpy(&node_keys[level - 1], &key,
1779                                                sizeof(node_keys[0]));
1780                                 } else {
1781                                         WARN_ON(1);
1782                                 }
1783                         }
1784
1785                         btrfs_tree_unlock(parent);
1786                         free_extent_buffer(parent);
1787                         parent = eb;
1788                         continue;
1789                 }
1790
1791                 btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
1792                 btrfs_set_node_ptr_generation(parent, slot, trans->transid);
1793                 btrfs_mark_buffer_dirty(parent);
1794
1795                 ret = btrfs_inc_extent_ref(trans, root,
1796                                         nodes[level - 1],
1797                                         blocksize, parent->start,
1798                                         btrfs_header_owner(parent),
1799                                         btrfs_header_generation(parent),
1800                                         level - 1);
1801                 BUG_ON(ret);
1802
1803                 /*
1804                  * If the block was created in the running transaction,
1805                  * it's possible this is the last reference to it, so we
1806                  * should drop the subtree.
1807                  */
1808                 if (generation == trans->transid) {
1809                         ret = btrfs_drop_subtree(trans, root, eb, parent);
1810                         BUG_ON(ret);
1811                         btrfs_tree_unlock(eb);
1812                         free_extent_buffer(eb);
1813                 } else {
1814                         ret = btrfs_free_extent(trans, root, bytenr,
1815                                         blocksize, parent->start,
1816                                         btrfs_header_owner(parent),
1817                                         btrfs_header_generation(parent),
1818                                         level - 1, 1);
1819                         BUG_ON(ret);
1820                 }
1821                 break;
1822         }
1823         btrfs_tree_unlock(parent);
1824         free_extent_buffer(parent);
1825         return 0;
1826 }
1827
1828 /*
1829  * adjust the pointers going up the tree, starting at level
1830  * making sure the right key of each node is points to 'key'.
1831  * This is used after shifting pointers to the left, so it stops
1832  * fixing up pointers when a given leaf/node is not in slot 0 of the
1833  * higher levels
1834  *
1835  * If this fails to write a tree block, it returns -1, but continues
1836  * fixing up the blocks in ram so the tree is consistent.
1837  */
1838 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1839                           struct btrfs_root *root, struct btrfs_path *path,
1840                           struct btrfs_disk_key *key, int level)
1841 {
1842         int i;
1843         int ret = 0;
1844         struct extent_buffer *t;
1845
1846         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1847                 int tslot = path->slots[i];
1848                 if (!path->nodes[i])
1849                         break;
1850                 t = path->nodes[i];
1851                 btrfs_set_node_key(t, key, tslot);
1852                 btrfs_mark_buffer_dirty(path->nodes[i]);
1853                 if (tslot != 0)
1854                         break;
1855         }
1856         return ret;
1857 }
1858
1859 /*
1860  * update item key.
1861  *
1862  * This function isn't completely safe. It's the caller's responsibility
1863  * that the new key won't break the order
1864  */
1865 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1866                             struct btrfs_root *root, struct btrfs_path *path,
1867                             struct btrfs_key *new_key)
1868 {
1869         struct btrfs_disk_key disk_key;
1870         struct extent_buffer *eb;
1871         int slot;
1872
1873         eb = path->nodes[0];
1874         slot = path->slots[0];
1875         if (slot > 0) {
1876                 btrfs_item_key(eb, &disk_key, slot - 1);
1877                 if (comp_keys(&disk_key, new_key) >= 0)
1878                         return -1;
1879         }
1880         if (slot < btrfs_header_nritems(eb) - 1) {
1881                 btrfs_item_key(eb, &disk_key, slot + 1);
1882                 if (comp_keys(&disk_key, new_key) <= 0)
1883                         return -1;
1884         }
1885
1886         btrfs_cpu_key_to_disk(&disk_key, new_key);
1887         btrfs_set_item_key(eb, &disk_key, slot);
1888         btrfs_mark_buffer_dirty(eb);
1889         if (slot == 0)
1890                 fixup_low_keys(trans, root, path, &disk_key, 1);
1891         return 0;
1892 }
1893
1894 /*
1895  * try to push data from one node into the next node left in the
1896  * tree.
1897  *
1898  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1899  * error, and > 0 if there was no room in the left hand block.
1900  */
1901 static int push_node_left(struct btrfs_trans_handle *trans,
1902                           struct btrfs_root *root, struct extent_buffer *dst,
1903                           struct extent_buffer *src, int empty)
1904 {
1905         int push_items = 0;
1906         int src_nritems;
1907         int dst_nritems;
1908         int ret = 0;
1909
1910         src_nritems = btrfs_header_nritems(src);
1911         dst_nritems = btrfs_header_nritems(dst);
1912         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1913         WARN_ON(btrfs_header_generation(src) != trans->transid);
1914         WARN_ON(btrfs_header_generation(dst) != trans->transid);
1915
1916         if (!empty && src_nritems <= 8)
1917                 return 1;
1918
1919         if (push_items <= 0)
1920                 return 1;
1921
1922         if (empty) {
1923                 push_items = min(src_nritems, push_items);
1924                 if (push_items < src_nritems) {
1925                         /* leave at least 8 pointers in the node if
1926                          * we aren't going to empty it
1927                          */
1928                         if (src_nritems - push_items < 8) {
1929                                 if (push_items <= 8)
1930                                         return 1;
1931                                 push_items -= 8;
1932                         }
1933                 }
1934         } else
1935                 push_items = min(src_nritems - 8, push_items);
1936
1937         copy_extent_buffer(dst, src,
1938                            btrfs_node_key_ptr_offset(dst_nritems),
1939                            btrfs_node_key_ptr_offset(0),
1940                            push_items * sizeof(struct btrfs_key_ptr));
1941
1942         if (push_items < src_nritems) {
1943                 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1944                                       btrfs_node_key_ptr_offset(push_items),
1945                                       (src_nritems - push_items) *
1946                                       sizeof(struct btrfs_key_ptr));
1947         }
1948         btrfs_set_header_nritems(src, src_nritems - push_items);
1949         btrfs_set_header_nritems(dst, dst_nritems + push_items);
1950         btrfs_mark_buffer_dirty(src);
1951         btrfs_mark_buffer_dirty(dst);
1952
1953         ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
1954         BUG_ON(ret);
1955
1956         return ret;
1957 }
1958
1959 /*
1960  * try to push data from one node into the next node right in the
1961  * tree.
1962  *
1963  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1964  * error, and > 0 if there was no room in the right hand block.
1965  *
1966  * this will  only push up to 1/2 the contents of the left node over
1967  */
1968 static int balance_node_right(struct btrfs_trans_handle *trans,
1969                               struct btrfs_root *root,
1970                               struct extent_buffer *dst,
1971                               struct extent_buffer *src)
1972 {
1973         int push_items = 0;
1974         int max_push;
1975         int src_nritems;
1976         int dst_nritems;
1977         int ret = 0;
1978
1979         WARN_ON(btrfs_header_generation(src) != trans->transid);
1980         WARN_ON(btrfs_header_generation(dst) != trans->transid);
1981
1982         src_nritems = btrfs_header_nritems(src);
1983         dst_nritems = btrfs_header_nritems(dst);
1984         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1985         if (push_items <= 0)
1986                 return 1;
1987
1988         if (src_nritems < 4)
1989                 return 1;
1990
1991         max_push = src_nritems / 2 + 1;
1992         /* don't try to empty the node */
1993         if (max_push >= src_nritems)
1994                 return 1;
1995
1996         if (max_push < push_items)
1997                 push_items = max_push;
1998
1999         memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2000                                       btrfs_node_key_ptr_offset(0),
2001                                       (dst_nritems) *
2002                                       sizeof(struct btrfs_key_ptr));
2003
2004         copy_extent_buffer(dst, src,
2005                            btrfs_node_key_ptr_offset(0),
2006                            btrfs_node_key_ptr_offset(src_nritems - push_items),
2007                            push_items * sizeof(struct btrfs_key_ptr));
2008
2009         btrfs_set_header_nritems(src, src_nritems - push_items);
2010         btrfs_set_header_nritems(dst, dst_nritems + push_items);
2011
2012         btrfs_mark_buffer_dirty(src);
2013         btrfs_mark_buffer_dirty(dst);
2014
2015         ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
2016         BUG_ON(ret);
2017
2018         return ret;
2019 }
2020
2021 /*
2022  * helper function to insert a new root level in the tree.
2023  * A new node is allocated, and a single item is inserted to
2024  * point to the existing root
2025  *
2026  * returns zero on success or < 0 on failure.
2027  */
2028 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2029                            struct btrfs_root *root,
2030                            struct btrfs_path *path, int level)
2031 {
2032         u64 lower_gen;
2033         struct extent_buffer *lower;
2034         struct extent_buffer *c;
2035         struct extent_buffer *old;
2036         struct btrfs_disk_key lower_key;
2037         int ret;
2038
2039         BUG_ON(path->nodes[level]);
2040         BUG_ON(path->nodes[level-1] != root->node);
2041
2042         lower = path->nodes[level-1];
2043         if (level == 1)
2044                 btrfs_item_key(lower, &lower_key, 0);
2045         else
2046                 btrfs_node_key(lower, &lower_key, 0);
2047
2048         c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2049                                    root->root_key.objectid, trans->transid,
2050                                    level, root->node->start, 0);
2051         if (IS_ERR(c))
2052                 return PTR_ERR(c);
2053
2054         memset_extent_buffer(c, 0, 0, root->nodesize);
2055         btrfs_set_header_nritems(c, 1);
2056         btrfs_set_header_level(c, level);
2057         btrfs_set_header_bytenr(c, c->start);
2058         btrfs_set_header_generation(c, trans->transid);
2059         btrfs_set_header_owner(c, root->root_key.objectid);
2060
2061         write_extent_buffer(c, root->fs_info->fsid,
2062                             (unsigned long)btrfs_header_fsid(c),
2063                             BTRFS_FSID_SIZE);
2064
2065         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2066                             (unsigned long)btrfs_header_chunk_tree_uuid(c),
2067                             BTRFS_UUID_SIZE);
2068
2069         btrfs_set_node_key(c, &lower_key, 0);
2070         btrfs_set_node_blockptr(c, 0, lower->start);
2071         lower_gen = btrfs_header_generation(lower);
2072         WARN_ON(lower_gen != trans->transid);
2073
2074         btrfs_set_node_ptr_generation(c, 0, lower_gen);
2075
2076         btrfs_mark_buffer_dirty(c);
2077
2078         spin_lock(&root->node_lock);
2079         old = root->node;
2080         root->node = c;
2081         spin_unlock(&root->node_lock);
2082
2083         ret = btrfs_update_extent_ref(trans, root, lower->start,
2084                                       lower->len, lower->start, c->start,
2085                                       root->root_key.objectid,
2086                                       trans->transid, level - 1);
2087         BUG_ON(ret);
2088
2089         /* the super has an extra ref to root->node */
2090         free_extent_buffer(old);
2091
2092         add_root_to_dirty_list(root);
2093         extent_buffer_get(c);
2094         path->nodes[level] = c;
2095         path->locks[level] = 1;
2096         path->slots[level] = 0;
2097         return 0;
2098 }
2099
2100 /*
2101  * worker function to insert a single pointer in a node.
2102  * the node should have enough room for the pointer already
2103  *
2104  * slot and level indicate where you want the key to go, and
2105  * blocknr is the block the key points to.
2106  *
2107  * returns zero on success and < 0 on any error
2108  */
2109 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2110                       *root, struct btrfs_path *path, struct btrfs_disk_key
2111                       *key, u64 bytenr, int slot, int level)
2112 {
2113         struct extent_buffer *lower;
2114         int nritems;
2115
2116         BUG_ON(!path->nodes[level]);
2117         lower = path->nodes[level];
2118         nritems = btrfs_header_nritems(lower);
2119         if (slot > nritems)
2120                 BUG();
2121         if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2122                 BUG();
2123         if (slot != nritems) {
2124                 memmove_extent_buffer(lower,
2125                               btrfs_node_key_ptr_offset(slot + 1),
2126                               btrfs_node_key_ptr_offset(slot),
2127                               (nritems - slot) * sizeof(struct btrfs_key_ptr));
2128         }
2129         btrfs_set_node_key(lower, key, slot);
2130         btrfs_set_node_blockptr(lower, slot, bytenr);
2131         WARN_ON(trans->transid == 0);
2132         btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2133         btrfs_set_header_nritems(lower, nritems + 1);
2134         btrfs_mark_buffer_dirty(lower);
2135         return 0;
2136 }
2137
2138 /*
2139  * split the node at the specified level in path in two.
2140  * The path is corrected to point to the appropriate node after the split
2141  *
2142  * Before splitting this tries to make some room in the node by pushing
2143  * left and right, if either one works, it returns right away.
2144  *
2145  * returns 0 on success and < 0 on failure
2146  */
2147 static noinline int split_node(struct btrfs_trans_handle *trans,
2148                                struct btrfs_root *root,
2149                                struct btrfs_path *path, int level)
2150 {
2151         struct extent_buffer *c;
2152         struct extent_buffer *split;
2153         struct btrfs_disk_key disk_key;
2154         int mid;
2155         int ret;
2156         int wret;
2157         u32 c_nritems;
2158
2159         c = path->nodes[level];
2160         WARN_ON(btrfs_header_generation(c) != trans->transid);
2161         if (c == root->node) {
2162                 /* trying to split the root, lets make a new one */
2163                 ret = insert_new_root(trans, root, path, level + 1);
2164                 if (ret)
2165                         return ret;
2166         } else if (!trans->transaction->delayed_refs.flushing) {
2167                 ret = push_nodes_for_insert(trans, root, path, level);
2168                 c = path->nodes[level];
2169                 if (!ret && btrfs_header_nritems(c) <
2170                     BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2171                         return 0;
2172                 if (ret < 0)
2173                         return ret;
2174         }
2175
2176         c_nritems = btrfs_header_nritems(c);
2177
2178         split = btrfs_alloc_free_block(trans, root, root->nodesize,
2179                                         path->nodes[level + 1]->start,
2180                                         root->root_key.objectid,
2181                                         trans->transid, level, c->start, 0);
2182         if (IS_ERR(split))
2183                 return PTR_ERR(split);
2184
2185         btrfs_set_header_flags(split, btrfs_header_flags(c));
2186         btrfs_set_header_level(split, btrfs_header_level(c));
2187         btrfs_set_header_bytenr(split, split->start);
2188         btrfs_set_header_generation(split, trans->transid);
2189         btrfs_set_header_owner(split, root->root_key.objectid);
2190         btrfs_set_header_flags(split, 0);
2191         write_extent_buffer(split, root->fs_info->fsid,
2192                             (unsigned long)btrfs_header_fsid(split),
2193                             BTRFS_FSID_SIZE);
2194         write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2195                             (unsigned long)btrfs_header_chunk_tree_uuid(split),
2196                             BTRFS_UUID_SIZE);
2197
2198         mid = (c_nritems + 1) / 2;
2199
2200         copy_extent_buffer(split, c,
2201                            btrfs_node_key_ptr_offset(0),
2202                            btrfs_node_key_ptr_offset(mid),
2203                            (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2204         btrfs_set_header_nritems(split, c_nritems - mid);
2205         btrfs_set_header_nritems(c, mid);
2206         ret = 0;
2207
2208         btrfs_mark_buffer_dirty(c);
2209         btrfs_mark_buffer_dirty(split);
2210
2211         btrfs_node_key(split, &disk_key, 0);
2212         wret = insert_ptr(trans, root, path, &disk_key, split->start,
2213                           path->slots[level + 1] + 1,
2214                           level + 1);
2215         if (wret)
2216                 ret = wret;
2217
2218         ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
2219         BUG_ON(ret);
2220
2221         if (path->slots[level] >= mid) {
2222                 path->slots[level] -= mid;
2223                 btrfs_tree_unlock(c);
2224                 free_extent_buffer(c);
2225                 path->nodes[level] = split;
2226                 path->slots[level + 1] += 1;
2227         } else {
2228                 btrfs_tree_unlock(split);
2229                 free_extent_buffer(split);
2230         }
2231         return ret;
2232 }
2233
2234 /*
2235  * how many bytes are required to store the items in a leaf.  start
2236  * and nr indicate which items in the leaf to check.  This totals up the
2237  * space used both by the item structs and the item data
2238  */
2239 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2240 {
2241         int data_len;
2242         int nritems = btrfs_header_nritems(l);
2243         int end = min(nritems, start + nr) - 1;
2244
2245         if (!nr)
2246                 return 0;
2247         data_len = btrfs_item_end_nr(l, start);
2248         data_len = data_len - btrfs_item_offset_nr(l, end);
2249         data_len += sizeof(struct btrfs_item) * nr;
2250         WARN_ON(data_len < 0);
2251         return data_len;
2252 }
2253
2254 /*
2255  * The space between the end of the leaf items and
2256  * the start of the leaf data.  IOW, how much room
2257  * the leaf has left for both items and data
2258  */
2259 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2260                                    struct extent_buffer *leaf)
2261 {
2262         int nritems = btrfs_header_nritems(leaf);
2263         int ret;
2264         ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2265         if (ret < 0) {
2266                 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2267                        "used %d nritems %d\n",
2268                        ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2269                        leaf_space_used(leaf, 0, nritems), nritems);
2270         }
2271         return ret;
2272 }
2273
2274 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2275                                       struct btrfs_root *root,
2276                                       struct btrfs_path *path,
2277                                       int data_size, int empty,
2278                                       struct extent_buffer *right,
2279                                       int free_space, u32 left_nritems)
2280 {
2281         struct extent_buffer *left = path->nodes[0];
2282         struct extent_buffer *upper = path->nodes[1];
2283         struct btrfs_disk_key disk_key;
2284         int slot;
2285         u32 i;
2286         int push_space = 0;
2287         int push_items = 0;
2288         struct btrfs_item *item;
2289         u32 nr;
2290         u32 right_nritems;
2291         u32 data_end;
2292         u32 this_item_size;
2293         int ret;
2294
2295         if (empty)
2296                 nr = 0;
2297         else
2298                 nr = 1;
2299
2300         if (path->slots[0] >= left_nritems)
2301                 push_space += data_size;
2302
2303         slot = path->slots[1];
2304         i = left_nritems - 1;
2305         while (i >= nr) {
2306                 item = btrfs_item_nr(left, i);
2307
2308                 if (!empty && push_items > 0) {
2309                         if (path->slots[0] > i)
2310                                 break;
2311                         if (path->slots[0] == i) {
2312                                 int space = btrfs_leaf_free_space(root, left);
2313                                 if (space + push_space * 2 > free_space)
2314                                         break;
2315                         }
2316                 }
2317
2318                 if (path->slots[0] == i)
2319                         push_space += data_size;
2320
2321                 if (!left->map_token) {
2322                         map_extent_buffer(left, (unsigned long)item,
2323                                         sizeof(struct btrfs_item),
2324                                         &left->map_token, &left->kaddr,
2325                                         &left->map_start, &left->map_len,
2326                                         KM_USER1);
2327                 }
2328
2329                 this_item_size = btrfs_item_size(left, item);
2330                 if (this_item_size + sizeof(*item) + push_space > free_space)
2331                         break;
2332
2333                 push_items++;
2334                 push_space += this_item_size + sizeof(*item);
2335                 if (i == 0)
2336                         break;
2337                 i--;
2338         }
2339         if (left->map_token) {
2340                 unmap_extent_buffer(left, left->map_token, KM_USER1);
2341                 left->map_token = NULL;
2342         }
2343
2344         if (push_items == 0)
2345                 goto out_unlock;
2346
2347         if (!empty && push_items == left_nritems)
2348                 WARN_ON(1);
2349
2350         /* push left to right */
2351         right_nritems = btrfs_header_nritems(right);
2352
2353         push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2354         push_space -= leaf_data_end(root, left);
2355
2356         /* make room in the right data area */
2357         data_end = leaf_data_end(root, right);
2358         memmove_extent_buffer(right,
2359                               btrfs_leaf_data(right) + data_end - push_space,
2360                               btrfs_leaf_data(right) + data_end,
2361                               BTRFS_LEAF_DATA_SIZE(root) - data_end);
2362
2363         /* copy from the left data area */
2364         copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2365                      BTRFS_LEAF_DATA_SIZE(root) - push_space,
2366                      btrfs_leaf_data(left) + leaf_data_end(root, left),
2367                      push_space);
2368
2369         memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2370                               btrfs_item_nr_offset(0),
2371                               right_nritems * sizeof(struct btrfs_item));
2372
2373         /* copy the items from left to right */
2374         copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2375                    btrfs_item_nr_offset(left_nritems - push_items),
2376                    push_items * sizeof(struct btrfs_item));
2377
2378         /* update the item pointers */
2379         right_nritems += push_items;
2380         btrfs_set_header_nritems(right, right_nritems);
2381         push_space = BTRFS_LEAF_DATA_SIZE(root);
2382         for (i = 0; i < right_nritems; i++) {
2383                 item = btrfs_item_nr(right, i);
2384                 if (!right->map_token) {
2385                         map_extent_buffer(right, (unsigned long)item,
2386                                         sizeof(struct btrfs_item),
2387                                         &right->map_token, &right->kaddr,
2388                                         &right->map_start, &right->map_len,
2389                                         KM_USER1);
2390                 }
2391                 push_space -= btrfs_item_size(right, item);
2392                 btrfs_set_item_offset(right, item, push_space);
2393         }
2394
2395         if (right->map_token) {
2396                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2397                 right->map_token = NULL;
2398         }
2399         left_nritems -= push_items;
2400         btrfs_set_header_nritems(left, left_nritems);
2401
2402         if (left_nritems)
2403                 btrfs_mark_buffer_dirty(left);
2404         btrfs_mark_buffer_dirty(right);
2405
2406         ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
2407         BUG_ON(ret);
2408
2409         btrfs_item_key(right, &disk_key, 0);
2410         btrfs_set_node_key(upper, &disk_key, slot + 1);
2411         btrfs_mark_buffer_dirty(upper);
2412
2413         /* then fixup the leaf pointer in the path */
2414         if (path->slots[0] >= left_nritems) {
2415                 path->slots[0] -= left_nritems;
2416                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2417                         clean_tree_block(trans, root, path->nodes[0]);
2418                 btrfs_tree_unlock(path->nodes[0]);
2419                 free_extent_buffer(path->nodes[0]);
2420                 path->nodes[0] = right;
2421                 path->slots[1] += 1;
2422         } else {
2423                 btrfs_tree_unlock(right);
2424                 free_extent_buffer(right);
2425         }
2426         return 0;
2427
2428 out_unlock:
2429         btrfs_tree_unlock(right);
2430         free_extent_buffer(right);
2431         return 1;
2432 }
2433
2434 /*
2435  * push some data in the path leaf to the right, trying to free up at
2436  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2437  *
2438  * returns 1 if the push failed because the other node didn't have enough
2439  * room, 0 if everything worked out and < 0 if there were major errors.
2440  */
2441 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2442                            *root, struct btrfs_path *path, int data_size,
2443                            int empty)
2444 {
2445         struct extent_buffer *left = path->nodes[0];
2446         struct extent_buffer *right;
2447         struct extent_buffer *upper;
2448         int slot;
2449         int free_space;
2450         u32 left_nritems;
2451         int ret;
2452
2453         if (!path->nodes[1])
2454                 return 1;
2455
2456         slot = path->slots[1];
2457         upper = path->nodes[1];
2458         if (slot >= btrfs_header_nritems(upper) - 1)
2459                 return 1;
2460
2461         btrfs_assert_tree_locked(path->nodes[1]);
2462
2463         right = read_node_slot(root, upper, slot + 1);
2464         btrfs_tree_lock(right);
2465         btrfs_set_lock_blocking(right);
2466
2467         free_space = btrfs_leaf_free_space(root, right);
2468         if (free_space < data_size)
2469                 goto out_unlock;
2470
2471         /* cow and double check */
2472         ret = btrfs_cow_block(trans, root, right, upper,
2473                               slot + 1, &right);
2474         if (ret)
2475                 goto out_unlock;
2476
2477         free_space = btrfs_leaf_free_space(root, right);
2478         if (free_space < data_size)
2479                 goto out_unlock;
2480
2481         left_nritems = btrfs_header_nritems(left);
2482         if (left_nritems == 0)
2483                 goto out_unlock;
2484
2485         return __push_leaf_right(trans, root, path, data_size, empty,
2486                                 right, free_space, left_nritems);
2487 out_unlock:
2488         btrfs_tree_unlock(right);
2489         free_extent_buffer(right);
2490         return 1;
2491 }
2492
2493 /*
2494  * push some data in the path leaf to the left, trying to free up at
2495  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2496  */
2497 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2498                                      struct btrfs_root *root,
2499                                      struct btrfs_path *path, int data_size,
2500                                      int empty, struct extent_buffer *left,
2501                                      int free_space, int right_nritems)
2502 {
2503         struct btrfs_disk_key disk_key;
2504         struct extent_buffer *right = path->nodes[0];
2505         int slot;
2506         int i;
2507         int push_space = 0;
2508         int push_items = 0;
2509         struct btrfs_item *item;
2510         u32 old_left_nritems;
2511         u32 nr;
2512         int ret = 0;
2513         int wret;
2514         u32 this_item_size;
2515         u32 old_left_item_size;
2516
2517         slot = path->slots[1];
2518
2519         if (empty)
2520                 nr = right_nritems;
2521         else
2522                 nr = right_nritems - 1;
2523
2524         for (i = 0; i < nr; i++) {
2525                 item = btrfs_item_nr(right, i);
2526                 if (!right->map_token) {
2527                         map_extent_buffer(right, (unsigned long)item,
2528                                         sizeof(struct btrfs_item),
2529                                         &right->map_token, &right->kaddr,
2530                                         &right->map_start, &right->map_len,
2531                                         KM_USER1);
2532                 }
2533
2534                 if (!empty && push_items > 0) {
2535                         if (path->slots[0] < i)
2536                                 break;
2537                         if (path->slots[0] == i) {
2538                                 int space = btrfs_leaf_free_space(root, right);
2539                                 if (space + push_space * 2 > free_space)
2540                                         break;
2541                         }
2542                 }
2543
2544                 if (path->slots[0] == i)
2545                         push_space += data_size;
2546
2547                 this_item_size = btrfs_item_size(right, item);
2548                 if (this_item_size + sizeof(*item) + push_space > free_space)
2549                         break;
2550
2551                 push_items++;
2552                 push_space += this_item_size + sizeof(*item);
2553         }
2554
2555         if (right->map_token) {
2556                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2557                 right->map_token = NULL;
2558         }
2559
2560         if (push_items == 0) {
2561                 ret = 1;
2562                 goto out;
2563         }
2564         if (!empty && push_items == btrfs_header_nritems(right))
2565                 WARN_ON(1);
2566
2567         /* push data from right to left */
2568         copy_extent_buffer(left, right,
2569                            btrfs_item_nr_offset(btrfs_header_nritems(left)),
2570                            btrfs_item_nr_offset(0),
2571                            push_items * sizeof(struct btrfs_item));
2572
2573         push_space = BTRFS_LEAF_DATA_SIZE(root) -
2574                      btrfs_item_offset_nr(right, push_items - 1);
2575
2576         copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2577                      leaf_data_end(root, left) - push_space,
2578                      btrfs_leaf_data(right) +
2579                      btrfs_item_offset_nr(right, push_items - 1),
2580                      push_space);
2581         old_left_nritems = btrfs_header_nritems(left);
2582         BUG_ON(old_left_nritems <= 0);
2583
2584         old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2585         for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2586                 u32 ioff;
2587
2588                 item = btrfs_item_nr(left, i);
2589                 if (!left->map_token) {
2590                         map_extent_buffer(left, (unsigned long)item,
2591                                         sizeof(struct btrfs_item),
2592                                         &left->map_token, &left->kaddr,
2593                                         &left->map_start, &left->map_len,
2594                                         KM_USER1);
2595                 }
2596
2597                 ioff = btrfs_item_offset(left, item);
2598                 btrfs_set_item_offset(left, item,
2599                       ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2600         }
2601         btrfs_set_header_nritems(left, old_left_nritems + push_items);
2602         if (left->map_token) {
2603                 unmap_extent_buffer(left, left->map_token, KM_USER1);
2604                 left->map_token = NULL;
2605         }
2606
2607         /* fixup right node */
2608         if (push_items > right_nritems) {
2609                 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2610                        right_nritems);
2611                 WARN_ON(1);
2612         }
2613
2614         if (push_items < right_nritems) {
2615                 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2616                                                   leaf_data_end(root, right);
2617                 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2618                                       BTRFS_LEAF_DATA_SIZE(root) - push_space,
2619                                       btrfs_leaf_data(right) +
2620                                       leaf_data_end(root, right), push_space);
2621
2622                 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2623                               btrfs_item_nr_offset(push_items),
2624                              (btrfs_header_nritems(right) - push_items) *
2625                              sizeof(struct btrfs_item));
2626         }
2627         right_nritems -= push_items;
2628         btrfs_set_header_nritems(right, right_nritems);
2629         push_space = BTRFS_LEAF_DATA_SIZE(root);
2630         for (i = 0; i < right_nritems; i++) {
2631                 item = btrfs_item_nr(right, i);
2632
2633                 if (!right->map_token) {
2634                         map_extent_buffer(right, (unsigned long)item,
2635                                         sizeof(struct btrfs_item),
2636                                         &right->map_token, &right->kaddr,
2637                                         &right->map_start, &right->map_len,
2638                                         KM_USER1);
2639                 }
2640
2641                 push_space = push_space - btrfs_item_size(right, item);
2642                 btrfs_set_item_offset(right, item, push_space);
2643         }
2644         if (right->map_token) {
2645                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2646                 right->map_token = NULL;
2647         }
2648
2649         btrfs_mark_buffer_dirty(left);
2650         if (right_nritems)
2651                 btrfs_mark_buffer_dirty(right);
2652
2653         ret = btrfs_update_ref(trans, root, right, left,
2654                                old_left_nritems, push_items);
2655         BUG_ON(ret);
2656
2657         btrfs_item_key(right, &disk_key, 0);
2658         wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2659         if (wret)
2660                 ret = wret;
2661
2662         /* then fixup the leaf pointer in the path */
2663         if (path->slots[0] < push_items) {
2664                 path->slots[0] += old_left_nritems;
2665                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2666                         clean_tree_block(trans, root, path->nodes[0]);
2667                 btrfs_tree_unlock(path->nodes[0]);
2668                 free_extent_buffer(path->nodes[0]);
2669                 path->nodes[0] = left;
2670                 path->slots[1] -= 1;
2671         } else {
2672                 btrfs_tree_unlock(left);
2673                 free_extent_buffer(left);
2674                 path->slots[0] -= push_items;
2675         }
2676         BUG_ON(path->slots[0] < 0);
2677         return ret;
2678 out:
2679         btrfs_tree_unlock(left);
2680         free_extent_buffer(left);
2681         return ret;
2682 }
2683
2684 /*
2685  * push some data in the path leaf to the left, trying to free up at
2686  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2687  */
2688 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2689                           *root, struct btrfs_path *path, int data_size,
2690                           int empty)
2691 {
2692         struct extent_buffer *right = path->nodes[0];
2693         struct extent_buffer *left;
2694         int slot;
2695         int free_space;
2696         u32 right_nritems;
2697         int ret = 0;
2698
2699         slot = path->slots[1];
2700         if (slot == 0)
2701                 return 1;
2702         if (!path->nodes[1])
2703                 return 1;
2704
2705         right_nritems = btrfs_header_nritems(right);
2706         if (right_nritems == 0)
2707                 return 1;
2708
2709         btrfs_assert_tree_locked(path->nodes[1]);
2710
2711         left = read_node_slot(root, path->nodes[1], slot - 1);
2712         btrfs_tree_lock(left);
2713         btrfs_set_lock_blocking(left);
2714
2715         free_space = btrfs_leaf_free_space(root, left);
2716         if (free_space < data_size) {
2717                 ret = 1;
2718                 goto out;
2719         }
2720
2721         /* cow and double check */
2722         ret = btrfs_cow_block(trans, root, left,
2723                               path->nodes[1], slot - 1, &left);
2724         if (ret) {
2725                 /* we hit -ENOSPC, but it isn't fatal here */
2726                 ret = 1;
2727                 goto out;
2728         }
2729
2730         free_space = btrfs_leaf_free_space(root, left);
2731         if (free_space < data_size) {
2732                 ret = 1;
2733                 goto out;
2734         }
2735
2736         return __push_leaf_left(trans, root, path, data_size,
2737                                empty, left, free_space, right_nritems);
2738 out:
2739         btrfs_tree_unlock(left);
2740         free_extent_buffer(left);
2741         return ret;
2742 }
2743
2744 /*
2745  * split the path's leaf in two, making sure there is at least data_size
2746  * available for the resulting leaf level of the path.
2747  *
2748  * returns 0 if all went well and < 0 on failure.
2749  */
2750 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2751                                struct btrfs_root *root,
2752                                struct btrfs_path *path,
2753                                struct extent_buffer *l,
2754                                struct extent_buffer *right,
2755                                int slot, int mid, int nritems)
2756 {
2757         int data_copy_size;
2758         int rt_data_off;
2759         int i;
2760         int ret = 0;
2761         int wret;
2762         struct btrfs_disk_key disk_key;
2763
2764         nritems = nritems - mid;
2765         btrfs_set_header_nritems(right, nritems);
2766         data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2767
2768         copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2769                            btrfs_item_nr_offset(mid),
2770                            nritems * sizeof(struct btrfs_item));
2771
2772         copy_extent_buffer(right, l,
2773                      btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2774                      data_copy_size, btrfs_leaf_data(l) +
2775                      leaf_data_end(root, l), data_copy_size);
2776
2777         rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2778                       btrfs_item_end_nr(l, mid);
2779
2780         for (i = 0; i < nritems; i++) {
2781                 struct btrfs_item *item = btrfs_item_nr(right, i);
2782                 u32 ioff;
2783
2784                 if (!right->map_token) {
2785                         map_extent_buffer(right, (unsigned long)item,
2786                                         sizeof(struct btrfs_item),
2787                                         &right->map_token, &right->kaddr,
2788                                         &right->map_start, &right->map_len,
2789                                         KM_USER1);
2790                 }
2791
2792                 ioff = btrfs_item_offset(right, item);
2793                 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2794         }
2795
2796         if (right->map_token) {
2797                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2798                 right->map_token = NULL;
2799         }
2800
2801         btrfs_set_header_nritems(l, mid);
2802         ret = 0;
2803         btrfs_item_key(right, &disk_key, 0);
2804         wret = insert_ptr(trans, root, path, &disk_key, right->start,
2805                           path->slots[1] + 1, 1);
2806         if (wret)
2807                 ret = wret;
2808
2809         btrfs_mark_buffer_dirty(right);
2810         btrfs_mark_buffer_dirty(l);
2811         BUG_ON(path->slots[0] != slot);
2812
2813         ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
2814         BUG_ON(ret);
2815
2816         if (mid <= slot) {
2817                 btrfs_tree_unlock(path->nodes[0]);
2818                 free_extent_buffer(path->nodes[0]);
2819                 path->nodes[0] = right;
2820                 path->slots[0] -= mid;
2821                 path->slots[1] += 1;
2822         } else {
2823                 btrfs_tree_unlock(right);
2824                 free_extent_buffer(right);
2825         }
2826
2827         BUG_ON(path->slots[0] < 0);
2828
2829         return ret;
2830 }
2831
2832 /*
2833  * split the path's leaf in two, making sure there is at least data_size
2834  * available for the resulting leaf level of the path.
2835  *
2836  * returns 0 if all went well and < 0 on failure.
2837  */
2838 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2839                                struct btrfs_root *root,
2840                                struct btrfs_key *ins_key,
2841                                struct btrfs_path *path, int data_size,
2842                                int extend)
2843 {
2844         struct extent_buffer *l;
2845         u32 nritems;
2846         int mid;
2847         int slot;
2848         struct extent_buffer *right;
2849         int ret = 0;
2850         int wret;
2851         int double_split;
2852         int num_doubles = 0;
2853
2854         /* first try to make some room by pushing left and right */
2855         if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY &&
2856             !trans->transaction->delayed_refs.flushing) {
2857                 wret = push_leaf_right(trans, root, path, data_size, 0);
2858                 if (wret < 0)
2859                         return wret;
2860                 if (wret) {
2861                         wret = push_leaf_left(trans, root, path, data_size, 0);
2862                         if (wret < 0)
2863                                 return wret;
2864                 }
2865                 l = path->nodes[0];
2866
2867                 /* did the pushes work? */
2868                 if (btrfs_leaf_free_space(root, l) >= data_size)
2869                         return 0;
2870         }
2871
2872         if (!path->nodes[1]) {
2873                 ret = insert_new_root(trans, root, path, 1);
2874                 if (ret)
2875                         return ret;
2876         }
2877 again:
2878         double_split = 0;
2879         l = path->nodes[0];
2880         slot = path->slots[0];
2881         nritems = btrfs_header_nritems(l);
2882         mid = (nritems + 1) / 2;
2883
2884         right = btrfs_alloc_free_block(trans, root, root->leafsize,
2885                                         path->nodes[1]->start,
2886                                         root->root_key.objectid,
2887                                         trans->transid, 0, l->start, 0);
2888         if (IS_ERR(right)) {
2889                 BUG_ON(1);
2890                 return PTR_ERR(right);
2891         }
2892
2893         memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2894         btrfs_set_header_bytenr(right, right->start);
2895         btrfs_set_header_generation(right, trans->transid);
2896         btrfs_set_header_owner(right, root->root_key.objectid);
2897         btrfs_set_header_level(right, 0);
2898         write_extent_buffer(right, root->fs_info->fsid,
2899                             (unsigned long)btrfs_header_fsid(right),
2900                             BTRFS_FSID_SIZE);
2901
2902         write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2903                             (unsigned long)btrfs_header_chunk_tree_uuid(right),
2904                             BTRFS_UUID_SIZE);
2905
2906         if (mid <= slot) {
2907                 if (nritems == 1 ||
2908                     leaf_space_used(l, mid, nritems - mid) + data_size >
2909                         BTRFS_LEAF_DATA_SIZE(root)) {
2910                         if (slot >= nritems) {
2911                                 struct btrfs_disk_key disk_key;
2912
2913                                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2914                                 btrfs_set_header_nritems(right, 0);
2915                                 wret = insert_ptr(trans, root, path,
2916                                                   &disk_key, right->start,
2917                                                   path->slots[1] + 1, 1);
2918                                 if (wret)
2919                                         ret = wret;
2920
2921                                 btrfs_tree_unlock(path->nodes[0]);
2922                                 free_extent_buffer(path->nodes[0]);
2923                                 path->nodes[0] = right;
2924                                 path->slots[0] = 0;
2925                                 path->slots[1] += 1;
2926                                 btrfs_mark_buffer_dirty(right);
2927                                 return ret;
2928                         }
2929                         mid = slot;
2930                         if (mid != nritems &&
2931                             leaf_space_used(l, mid, nritems - mid) +
2932                             data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2933                                 double_split = 1;
2934                         }
2935                 }
2936         } else {
2937                 if (leaf_space_used(l, 0, mid) + data_size >
2938                         BTRFS_LEAF_DATA_SIZE(root)) {
2939                         if (!extend && data_size && slot == 0) {
2940                                 struct btrfs_disk_key disk_key;
2941
2942                                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2943                                 btrfs_set_header_nritems(right, 0);
2944                                 wret = insert_ptr(trans, root, path,
2945                                                   &disk_key,
2946                                                   right->start,
2947                                                   path->slots[1], 1);
2948                                 if (wret)
2949                                         ret = wret;
2950                                 btrfs_tree_unlock(path->nodes[0]);
2951                                 free_extent_buffer(path->nodes[0]);
2952                                 path->nodes[0] = right;
2953                                 path->slots[0] = 0;
2954                                 if (path->slots[1] == 0) {
2955                                         wret = fixup_low_keys(trans, root,
2956                                                       path, &disk_key, 1);
2957                                         if (wret)
2958                                                 ret = wret;
2959                                 }
2960                                 btrfs_mark_buffer_dirty(right);
2961                                 return ret;
2962                         } else if ((extend || !data_size) && slot == 0) {
2963                                 mid = 1;
2964                         } else {
2965                                 mid = slot;
2966                                 if (mid != nritems &&
2967                                     leaf_space_used(l, mid, nritems - mid) +
2968                                     data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2969                                         double_split = 1;
2970                                 }
2971                         }
2972                 }
2973         }
2974
2975         ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2976         BUG_ON(ret);
2977
2978         if (double_split) {
2979                 BUG_ON(num_doubles != 0);
2980                 num_doubles++;
2981                 goto again;
2982         }
2983
2984         return ret;
2985 }
2986
2987 /*
2988  * This function splits a single item into two items,
2989  * giving 'new_key' to the new item and splitting the
2990  * old one at split_offset (from the start of the item).
2991  *
2992  * The path may be released by this operation.  After
2993  * the split, the path is pointing to the old item.  The
2994  * new item is going to be in the same node as the old one.
2995  *
2996  * Note, the item being split must be smaller enough to live alone on
2997  * a tree block with room for one extra struct btrfs_item
2998  *
2999  * This allows us to split the item in place, keeping a lock on the
3000  * leaf the entire time.
3001  */
3002 int btrfs_split_item(struct btrfs_trans_handle *trans,
3003                      struct btrfs_root *root,
3004                      struct btrfs_path *path,
3005                      struct btrfs_key *new_key,
3006                      unsigned long split_offset)
3007 {
3008         u32 item_size;
3009         struct extent_buffer *leaf;
3010         struct btrfs_key orig_key;
3011         struct btrfs_item *item;
3012         struct btrfs_item *new_item;
3013         int ret = 0;
3014         int slot;
3015         u32 nritems;
3016         u32 orig_offset;
3017         struct btrfs_disk_key disk_key;
3018         char *buf;
3019
3020         leaf = path->nodes[0];
3021         btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
3022         if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
3023                 goto split;
3024
3025         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3026         btrfs_release_path(root, path);
3027
3028         path->search_for_split = 1;
3029         path->keep_locks = 1;
3030
3031         ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
3032         path->search_for_split = 0;
3033
3034         /* if our item isn't there or got smaller, return now */
3035         if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
3036                                                         path->slots[0])) {
3037                 path->keep_locks = 0;
3038                 return -EAGAIN;
3039         }
3040
3041         btrfs_set_path_blocking(path);
3042         ret = split_leaf(trans, root, &orig_key, path,
3043                          sizeof(struct btrfs_item), 1);
3044         path->keep_locks = 0;
3045         BUG_ON(ret);
3046
3047         btrfs_unlock_up_safe(path, 1);
3048         leaf = path->nodes[0];
3049         BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3050
3051 split:
3052         /*
3053          * make sure any changes to the path from split_leaf leave it
3054          * in a blocking state
3055          */
3056         btrfs_set_path_blocking(path);
3057
3058         item = btrfs_item_nr(leaf, path->slots[0]);
3059         orig_offset = btrfs_item_offset(leaf, item);
3060         item_size = btrfs_item_size(leaf, item);
3061
3062         buf = kmalloc(item_size, GFP_NOFS);
3063         read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3064                             path->slots[0]), item_size);
3065         slot = path->slots[0] + 1;
3066         leaf = path->nodes[0];
3067
3068         nritems = btrfs_header_nritems(leaf);
3069
3070         if (slot != nritems) {
3071                 /* shift the items */
3072                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3073                               btrfs_item_nr_offset(slot),
3074                               (nritems - slot) * sizeof(struct btrfs_item));
3075
3076         }
3077
3078         btrfs_cpu_key_to_disk(&disk_key, new_key);
3079         btrfs_set_item_key(leaf, &disk_key, slot);
3080
3081         new_item = btrfs_item_nr(leaf, slot);
3082
3083         btrfs_set_item_offset(leaf, new_item, orig_offset);
3084         btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3085
3086         btrfs_set_item_offset(leaf, item,
3087                               orig_offset + item_size - split_offset);
3088         btrfs_set_item_size(leaf, item, split_offset);
3089
3090         btrfs_set_header_nritems(leaf, nritems + 1);
3091
3092         /* write the data for the start of the original item */
3093         write_extent_buffer(leaf, buf,
3094                             btrfs_item_ptr_offset(leaf, path->slots[0]),
3095                             split_offset);
3096
3097         /* write the data for the new item */
3098         write_extent_buffer(leaf, buf + split_offset,
3099                             btrfs_item_ptr_offset(leaf, slot),
3100                             item_size - split_offset);
3101         btrfs_mark_buffer_dirty(leaf);
3102
3103         ret = 0;
3104         if (btrfs_leaf_free_space(root, leaf) < 0) {
3105                 btrfs_print_leaf(root, leaf);
3106                 BUG();
3107         }
3108         kfree(buf);
3109         return ret;
3110 }
3111
3112 /*
3113  * make the item pointed to by the path smaller.  new_size indicates
3114  * how small to make it, and from_end tells us if we just chop bytes
3115  * off the end of the item or if we shift the item to chop bytes off
3116  * the front.
3117  */
3118 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3119                         struct btrfs_root *root,
3120                         struct btrfs_path *path,
3121                         u32 new_size, int from_end)
3122 {
3123         int ret = 0;
3124         int slot;
3125         int slot_orig;
3126         struct extent_buffer *leaf;
3127         struct btrfs_item *item;
3128         u32 nritems;
3129         unsigned int data_end;
3130         unsigned int old_data_start;
3131         unsigned int old_size;
3132         unsigned int size_diff;
3133         int i;
3134
3135         slot_orig = path->slots[0];
3136         leaf = path->nodes[0];
3137         slot = path->slots[0];
3138
3139         old_size = btrfs_item_size_nr(leaf, slot);
3140         if (old_size == new_size)
3141                 return 0;
3142
3143         nritems = btrfs_header_nritems(leaf);
3144         data_end = leaf_data_end(root, leaf);
3145
3146         old_data_start = btrfs_item_offset_nr(leaf, slot);
3147
3148         size_diff = old_size - new_size;
3149
3150         BUG_ON(slot < 0);
3151         BUG_ON(slot >= nritems);
3152
3153         /*
3154          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3155          */
3156         /* first correct the data pointers */
3157         for (i = slot; i < nritems; i++) {
3158                 u32 ioff;
3159                 item = btrfs_item_nr(leaf, i);
3160
3161                 if (!leaf->map_token) {
3162                         map_extent_buffer(leaf, (unsigned long)item,
3163                                         sizeof(struct btrfs_item),
3164                                         &leaf->map_token, &leaf->kaddr,
3165                                         &leaf->map_start, &leaf->map_len,
3166                                         KM_USER1);
3167                 }
3168
3169                 ioff = btrfs_item_offset(leaf, item);
3170                 btrfs_set_item_offset(leaf, item, ioff + size_diff);
3171         }
3172
3173         if (leaf->map_token) {
3174                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3175                 leaf->map_token = NULL;
3176         }
3177
3178         /* shift the data */
3179         if (from_end) {
3180                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3181                               data_end + size_diff, btrfs_leaf_data(leaf) +
3182                               data_end, old_data_start + new_size - data_end);
3183         } else {
3184                 struct btrfs_disk_key disk_key;
3185                 u64 offset;
3186
3187                 btrfs_item_key(leaf, &disk_key, slot);
3188
3189                 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3190                         unsigned long ptr;
3191                         struct btrfs_file_extent_item *fi;
3192
3193                         fi = btrfs_item_ptr(leaf, slot,
3194                                             struct btrfs_file_extent_item);
3195                         fi = (struct btrfs_file_extent_item *)(
3196                              (unsigned long)fi - size_diff);
3197
3198                         if (btrfs_file_extent_type(leaf, fi) ==
3199                             BTRFS_FILE_EXTENT_INLINE) {
3200                                 ptr = btrfs_item_ptr_offset(leaf, slot);
3201                                 memmove_extent_buffer(leaf, ptr,
3202                                       (unsigned long)fi,
3203                                       offsetof(struct btrfs_file_extent_item,
3204                                                  disk_bytenr));
3205                         }
3206                 }
3207
3208                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3209                               data_end + size_diff, btrfs_leaf_data(leaf) +
3210                               data_end, old_data_start - data_end);
3211
3212                 offset = btrfs_disk_key_offset(&disk_key);
3213                 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3214                 btrfs_set_item_key(leaf, &disk_key, slot);
3215                 if (slot == 0)
3216                         fixup_low_keys(trans, root, path, &disk_key, 1);
3217         }
3218
3219         item = btrfs_item_nr(leaf, slot);
3220         btrfs_set_item_size(leaf, item, new_size);
3221         btrfs_mark_buffer_dirty(leaf);
3222
3223         ret = 0;
3224         if (btrfs_leaf_free_space(root, leaf) < 0) {
3225                 btrfs_print_leaf(root, leaf);
3226                 BUG();
3227         }
3228         return ret;
3229 }
3230
3231 /*
3232  * make the item pointed to by the path bigger, data_size is the new size.
3233  */
3234 int btrfs_extend_item(struct btrfs_trans_handle *trans,
3235                       struct btrfs_root *root, struct btrfs_path *path,
3236                       u32 data_size)
3237 {
3238         int ret = 0;
3239         int slot;
3240         int slot_orig;
3241         struct extent_buffer *leaf;
3242         struct btrfs_item *item;
3243         u32 nritems;
3244         unsigned int data_end;
3245         unsigned int old_data;
3246         unsigned int old_size;
3247         int i;
3248
3249         slot_orig = path->slots[0];
3250         leaf = path->nodes[0];
3251
3252         nritems = btrfs_header_nritems(leaf);
3253         data_end = leaf_data_end(root, leaf);
3254
3255         if (btrfs_leaf_free_space(root, leaf) < data_size) {
3256                 btrfs_print_leaf(root, leaf);
3257                 BUG();
3258         }
3259         slot = path->slots[0];
3260         old_data = btrfs_item_end_nr(leaf, slot);
3261
3262         BUG_ON(slot < 0);
3263         if (slot >= nritems) {
3264                 btrfs_print_leaf(root, leaf);
3265                 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3266                        slot, nritems);
3267                 BUG_ON(1);
3268         }
3269
3270         /*
3271          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3272          */
3273         /* first correct the data pointers */
3274         for (i = slot; i < nritems; i++) {
3275                 u32 ioff;
3276                 item = btrfs_item_nr(leaf, i);
3277
3278                 if (!leaf->map_token) {
3279                         map_extent_buffer(leaf, (unsigned long)item,
3280                                         sizeof(struct btrfs_item),
3281                                         &leaf->map_token, &leaf->kaddr,
3282                                         &leaf->map_start, &leaf->map_len,
3283                                         KM_USER1);
3284                 }
3285                 ioff = btrfs_item_offset(leaf, item);
3286                 btrfs_set_item_offset(leaf, item, ioff - data_size);
3287         }
3288
3289         if (leaf->map_token) {
3290                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3291                 leaf->map_token = NULL;
3292         }
3293
3294         /* shift the data */
3295         memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3296                       data_end - data_size, btrfs_leaf_data(leaf) +
3297                       data_end, old_data - data_end);
3298
3299         data_end = old_data;
3300         old_size = btrfs_item_size_nr(leaf, slot);
3301         item = btrfs_item_nr(leaf, slot);
3302         btrfs_set_item_size(leaf, item, old_size + data_size);
3303         btrfs_mark_buffer_dirty(leaf);
3304
3305         ret = 0;
3306         if (btrfs_leaf_free_space(root, leaf) < 0) {
3307                 btrfs_print_leaf(root, leaf);
3308                 BUG();
3309         }
3310         return ret;
3311 }
3312
3313 /*
3314  * Given a key and some data, insert items into the tree.
3315  * This does all the path init required, making room in the tree if needed.
3316  * Returns the number of keys that were inserted.
3317  */
3318 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3319                             struct btrfs_root *root,
3320                             struct btrfs_path *path,
3321                             struct btrfs_key *cpu_key, u32 *data_size,
3322                             int nr)
3323 {
3324         struct extent_buffer *leaf;
3325         struct btrfs_item *item;
3326         int ret = 0;
3327         int slot;
3328         int i;
3329         u32 nritems;
3330         u32 total_data = 0;
3331         u32 total_size = 0;
3332         unsigned int data_end;
3333         struct btrfs_disk_key disk_key;
3334         struct btrfs_key found_key;
3335
3336         for (i = 0; i < nr; i++) {
3337                 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3338                     BTRFS_LEAF_DATA_SIZE(root)) {
3339                         break;
3340                         nr = i;
3341                 }
3342                 total_data += data_size[i];
3343                 total_size += data_size[i] + sizeof(struct btrfs_item);
3344         }
3345         BUG_ON(nr == 0);
3346
3347         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3348         if (ret == 0)
3349                 return -EEXIST;
3350         if (ret < 0)
3351                 goto out;
3352
3353         leaf = path->nodes[0];
3354
3355         nritems = btrfs_header_nritems(leaf);
3356         data_end = leaf_data_end(root, leaf);
3357
3358         if (btrfs_leaf_free_space(root, leaf) < total_size) {
3359                 for (i = nr; i >= 0; i--) {
3360                         total_data -= data_size[i];
3361                         total_size -= data_size[i] + sizeof(struct btrfs_item);
3362                         if (total_size < btrfs_leaf_free_space(root, leaf))
3363                                 break;
3364                 }
3365                 nr = i;
3366         }
3367
3368         slot = path->slots[0];
3369         BUG_ON(slot < 0);
3370
3371         if (slot != nritems) {
3372                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3373
3374                 item = btrfs_item_nr(leaf, slot);
3375                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3376
3377                 /* figure out how many keys we can insert in here */
3378                 total_data = data_size[0];
3379                 for (i = 1; i < nr; i++) {
3380                         if (comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3381                                 break;
3382                         total_data += data_size[i];
3383                 }
3384                 nr = i;
3385
3386                 if (old_data < data_end) {
3387                         btrfs_print_leaf(root, leaf);
3388                         printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3389                                slot, old_data, data_end);
3390                         BUG_ON(1);
3391                 }
3392                 /*
3393                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3394                  */
3395                 /* first correct the data pointers */
3396                 WARN_ON(leaf->map_token);
3397                 for (i = slot; i < nritems; i++) {
3398                         u32 ioff;
3399
3400                         item = btrfs_item_nr(leaf, i);
3401                         if (!leaf->map_token) {
3402                                 map_extent_buffer(leaf, (unsigned long)item,
3403                                         sizeof(struct btrfs_item),
3404                                         &leaf->map_token, &leaf->kaddr,
3405                                         &leaf->map_start, &leaf->map_len,
3406                                         KM_USER1);
3407                         }
3408
3409                         ioff = btrfs_item_offset(leaf, item);
3410                         btrfs_set_item_offset(leaf, item, ioff - total_data);
3411                 }
3412                 if (leaf->map_token) {
3413                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3414                         leaf->map_token = NULL;
3415                 }
3416
3417                 /* shift the items */
3418                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3419                               btrfs_item_nr_offset(slot),
3420                               (nritems - slot) * sizeof(struct btrfs_item));
3421
3422                 /* shift the data */
3423                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3424                               data_end - total_data, btrfs_leaf_data(leaf) +
3425                               data_end, old_data - data_end);
3426                 data_end = old_data;
3427         } else {
3428                 /*
3429                  * this sucks but it has to be done, if we are inserting at
3430                  * the end of the leaf only insert 1 of the items, since we
3431                  * have no way of knowing whats on the next leaf and we'd have
3432                  * to drop our current locks to figure it out
3433                  */
3434                 nr = 1;
3435         }
3436
3437         /* setup the item for the new data */
3438         for (i = 0; i < nr; i++) {
3439                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3440                 btrfs_set_item_key(leaf, &disk_key, slot + i);
3441                 item = btrfs_item_nr(leaf, slot + i);
3442                 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3443                 data_end -= data_size[i];
3444                 btrfs_set_item_size(leaf, item, data_size[i]);
3445         }
3446         btrfs_set_header_nritems(leaf, nritems + nr);
3447         btrfs_mark_buffer_dirty(leaf);
3448
3449         ret = 0;
3450         if (slot == 0) {
3451                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3452                 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3453         }
3454
3455         if (btrfs_leaf_free_space(root, leaf) < 0) {
3456                 btrfs_print_leaf(root, leaf);
3457                 BUG();
3458         }
3459 out:
3460         if (!ret)
3461                 ret = nr;
3462         return ret;
3463 }
3464
3465 /*
3466  * this is a helper for btrfs_insert_empty_items, the main goal here is
3467  * to save stack depth by doing the bulk of the work in a function
3468  * that doesn't call btrfs_search_slot
3469  */
3470 static noinline_for_stack int
3471 setup_items_for_insert(struct btrfs_trans_handle *trans,
3472                       struct btrfs_root *root, struct btrfs_path *path,
3473                       struct btrfs_key *cpu_key, u32 *data_size,
3474                       u32 total_data, u32 total_size, int nr)
3475 {
3476         struct btrfs_item *item;
3477         int i;
3478         u32 nritems;
3479         unsigned int data_end;
3480         struct btrfs_disk_key disk_key;
3481         int ret;
3482         struct extent_buffer *leaf;
3483         int slot;
3484
3485         leaf = path->nodes[0];
3486         slot = path->slots[0];
3487
3488         nritems = btrfs_header_nritems(leaf);
3489         data_end = leaf_data_end(root, leaf);
3490
3491         if (btrfs_leaf_free_space(root, leaf) < total_size) {
3492                 btrfs_print_leaf(root, leaf);
3493                 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3494                        total_size, btrfs_leaf_free_space(root, leaf));
3495                 BUG();
3496         }
3497
3498         if (slot != nritems) {
3499                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3500
3501                 if (old_data < data_end) {
3502                         btrfs_print_leaf(root, leaf);
3503                         printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3504                                slot, old_data, data_end);
3505                         BUG_ON(1);
3506                 }
3507                 /*
3508                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3509                  */
3510                 /* first correct the data pointers */
3511                 WARN_ON(leaf->map_token);
3512                 for (i = slot; i < nritems; i++) {
3513                         u32 ioff;
3514
3515                         item = btrfs_item_nr(leaf, i);
3516                         if (!leaf->map_token) {
3517                                 map_extent_buffer(leaf, (unsigned long)item,
3518                                         sizeof(struct btrfs_item),
3519                                         &leaf->map_token, &leaf->kaddr,
3520                                         &leaf->map_start, &leaf->map_len,
3521                                         KM_USER1);
3522                         }
3523
3524                         ioff = btrfs_item_offset(leaf, item);
3525                         btrfs_set_item_offset(leaf, item, ioff - total_data);
3526                 }
3527                 if (leaf->map_token) {
3528                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3529                         leaf->map_token = NULL;
3530                 }
3531
3532                 /* shift the items */
3533                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3534                               btrfs_item_nr_offset(slot),
3535                               (nritems - slot) * sizeof(struct btrfs_item));
3536
3537                 /* shift the data */
3538                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3539                               data_end - total_data, btrfs_leaf_data(leaf) +
3540                               data_end, old_data - data_end);
3541                 data_end = old_data;
3542         }
3543
3544         /* setup the item for the new data */
3545         for (i = 0; i < nr; i++) {
3546                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3547                 btrfs_set_item_key(leaf, &disk_key, slot + i);
3548                 item = btrfs_item_nr(leaf, slot + i);
3549                 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3550                 data_end -= data_size[i];
3551                 btrfs_set_item_size(leaf, item, data_size[i]);
3552         }
3553
3554         btrfs_set_header_nritems(leaf, nritems + nr);
3555
3556         ret = 0;
3557         if (slot == 0) {
3558                 struct btrfs_disk_key disk_key;
3559                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3560                 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3561         }
3562         btrfs_unlock_up_safe(path, 1);
3563         btrfs_mark_buffer_dirty(leaf);
3564
3565         if (btrfs_leaf_free_space(root, leaf) < 0) {
3566                 btrfs_print_leaf(root, leaf);
3567                 BUG();
3568         }
3569         return ret;
3570 }
3571
3572 /*
3573  * Given a key and some data, insert items into the tree.
3574  * This does all the path init required, making room in the tree if needed.
3575  */
3576 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3577                             struct btrfs_root *root,
3578                             struct btrfs_path *path,
3579                             struct btrfs_key *cpu_key, u32 *data_size,
3580                             int nr)
3581 {
3582         struct extent_buffer *leaf;
3583         int ret = 0;
3584         int slot;
3585         int i;
3586         u32 total_size = 0;
3587         u32 total_data = 0;
3588
3589         for (i = 0; i < nr; i++)
3590                 total_data += data_size[i];
3591
3592         total_size = total_data + (nr * sizeof(struct btrfs_item));
3593         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3594         if (ret == 0)
3595                 return -EEXIST;
3596         if (ret < 0)
3597                 goto out;
3598
3599         leaf = path->nodes[0];
3600         slot = path->slots[0];
3601         BUG_ON(slot < 0);
3602
3603         ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3604                                total_data, total_size, nr);
3605
3606 out:
3607         return ret;
3608 }
3609
3610 /*
3611  * Given a key and some data, insert an item into the tree.
3612  * This does all the path init required, making room in the tree if needed.
3613  */
3614 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3615                       *root, struct btrfs_key *cpu_key, void *data, u32
3616                       data_size)
3617 {
3618         int ret = 0;
3619         struct btrfs_path *path;
3620         struct extent_buffer *leaf;
3621         unsigned long ptr;
3622
3623         path = btrfs_alloc_path();
3624         BUG_ON(!path);
3625         ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3626         if (!ret) {
3627                 leaf = path->nodes[0];
3628                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3629                 write_extent_buffer(leaf, data, ptr, data_size);
3630                 btrfs_mark_buffer_dirty(leaf);
3631         }
3632         btrfs_free_path(path);
3633         return ret;
3634 }
3635
3636 /*
3637  * delete the pointer from a given node.
3638  *
3639  * the tree should have been previously balanced so the deletion does not
3640  * empty a node.
3641  */
3642 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3643                    struct btrfs_path *path, int level, int slot)
3644 {
3645         struct extent_buffer *parent = path->nodes[level];
3646         u32 nritems;
3647         int ret = 0;
3648         int wret;
3649
3650         nritems = btrfs_header_nritems(parent);
3651         if (slot != nritems - 1) {
3652                 memmove_extent_buffer(parent,
3653                               btrfs_node_key_ptr_offset(slot),
3654                               btrfs_node_key_ptr_offset(slot + 1),
3655                               sizeof(struct btrfs_key_ptr) *
3656                               (nritems - slot - 1));
3657         }
3658         nritems--;
3659         btrfs_set_header_nritems(parent, nritems);
3660         if (nritems == 0 && parent == root->node) {
3661                 BUG_ON(btrfs_header_level(root->node) != 1);
3662                 /* just turn the root into a leaf and break */
3663                 btrfs_set_header_level(root->node, 0);
3664         } else if (slot == 0) {
3665                 struct btrfs_disk_key disk_key;
3666
3667                 btrfs_node_key(parent, &disk_key, 0);
3668                 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3669                 if (wret)
3670                         ret = wret;
3671         }
3672         btrfs_mark_buffer_dirty(parent);
3673         return ret;
3674 }
3675
3676 /*
3677  * a helper function to delete the leaf pointed to by path->slots[1] and
3678  * path->nodes[1].  bytenr is the node block pointer, but since the callers
3679  * already know it, it is faster to have them pass it down than to
3680  * read it out of the node again.
3681  *
3682  * This deletes the pointer in path->nodes[1] and frees the leaf
3683  * block extent.  zero is returned if it all worked out, < 0 otherwise.
3684  *
3685  * The path must have already been setup for deleting the leaf, including
3686  * all the proper balancing.  path->nodes[1] must be locked.
3687  */
3688 noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3689                             struct btrfs_root *root,
3690                             struct btrfs_path *path, u64 bytenr)
3691 {
3692         int ret;
3693         u64 root_gen = btrfs_header_generation(path->nodes[1]);
3694         u64 parent_start = path->nodes[1]->start;
3695         u64 parent_owner = btrfs_header_owner(path->nodes[1]);
3696
3697         ret = del_ptr(trans, root, path, 1, path->slots[1]);
3698         if (ret)
3699                 return ret;
3700
3701         /*
3702          * btrfs_free_extent is expensive, we want to make sure we
3703          * aren't holding any locks when we call it
3704          */
3705         btrfs_unlock_up_safe(path, 0);
3706
3707         ret = btrfs_free_extent(trans, root, bytenr,
3708                                 btrfs_level_size(root, 0),
3709                                 parent_start, parent_owner,
3710                                 root_gen, 0, 1);
3711         return ret;
3712 }
3713 /*
3714  * delete the item at the leaf level in path.  If that empties
3715  * the leaf, remove it from the tree
3716  */
3717 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3718                     struct btrfs_path *path, int slot, int nr)
3719 {
3720         struct extent_buffer *leaf;
3721         struct btrfs_item *item;
3722         int last_off;
3723         int dsize = 0;
3724         int ret = 0;
3725         int wret;
3726         int i;
3727         u32 nritems;
3728
3729         leaf = path->nodes[0];
3730         last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3731
3732         for (i = 0; i < nr; i++)
3733                 dsize += btrfs_item_size_nr(leaf, slot + i);
3734
3735         nritems = btrfs_header_nritems(leaf);
3736
3737         if (slot + nr != nritems) {
3738                 int data_end = leaf_data_end(root, leaf);
3739
3740                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3741                               data_end + dsize,
3742                               btrfs_leaf_data(leaf) + data_end,
3743                               last_off - data_end);
3744
3745                 for (i = slot + nr; i < nritems; i++) {
3746                         u32 ioff;
3747
3748                         item = btrfs_item_nr(leaf, i);
3749                         if (!leaf->map_token) {
3750                                 map_extent_buffer(leaf, (unsigned long)item,
3751                                         sizeof(struct btrfs_item),
3752                                         &leaf->map_token, &leaf->kaddr,
3753                                         &leaf->map_start, &leaf->map_len,
3754                                         KM_USER1);
3755                         }
3756                         ioff = btrfs_item_offset(leaf, item);
3757                         btrfs_set_item_offset(leaf, item, ioff + dsize);
3758                 }
3759
3760                 if (leaf->map_token) {
3761                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3762                         leaf->map_token = NULL;
3763                 }
3764
3765                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3766                               btrfs_item_nr_offset(slot + nr),
3767                               sizeof(struct btrfs_item) *
3768                               (nritems - slot - nr));
3769         }
3770         btrfs_set_header_nritems(leaf, nritems - nr);
3771         nritems -= nr;
3772
3773         /* delete the leaf if we've emptied it */
3774         if (nritems == 0) {
3775                 if (leaf == root->node) {
3776                         btrfs_set_header_level(leaf, 0);
3777                 } else {
3778                         ret = btrfs_del_leaf(trans, root, path, leaf->start);
3779                         BUG_ON(ret);
3780                 }
3781         } else {
3782                 int used = leaf_space_used(leaf, 0, nritems);
3783                 if (slot == 0) {
3784                         struct btrfs_disk_key disk_key;
3785
3786                         btrfs_item_key(leaf, &disk_key, 0);
3787                         wret = fixup_low_keys(trans, root, path,
3788                                               &disk_key, 1);
3789                         if (wret)
3790                                 ret = wret;
3791                 }
3792
3793                 /* delete the leaf if it is mostly empty */
3794                 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4 &&
3795                     !trans->transaction->delayed_refs.flushing) {
3796                         /* push_leaf_left fixes the path.
3797                          * make sure the path still points to our leaf
3798                          * for possible call to del_ptr below
3799                          */
3800                         slot = path->slots[1];
3801                         extent_buffer_get(leaf);
3802
3803                         btrfs_set_path_blocking(path);
3804                         wret = push_leaf_left(trans, root, path, 1, 1);
3805                         if (wret < 0 && wret != -ENOSPC)
3806                                 ret = wret;
3807
3808                         if (path->nodes[0] == leaf &&
3809                             btrfs_header_nritems(leaf)) {
3810                                 wret = push_leaf_right(trans, root, path, 1, 1);
3811                                 if (wret < 0 && wret != -ENOSPC)
3812                                         ret = wret;
3813                         }
3814
3815                         if (btrfs_header_nritems(leaf) == 0) {
3816                                 path->slots[1] = slot;
3817                                 ret = btrfs_del_leaf(trans, root, path,
3818                                                      leaf->start);
3819                                 BUG_ON(ret);
3820                                 free_extent_buffer(leaf);
3821                         } else {
3822                                 /* if we're still in the path, make sure
3823                                  * we're dirty.  Otherwise, one of the
3824                                  * push_leaf functions must have already
3825                                  * dirtied this buffer
3826                                  */
3827                                 if (path->nodes[0] == leaf)
3828                                         btrfs_mark_buffer_dirty(leaf);
3829                                 free_extent_buffer(leaf);
3830                         }
3831                 } else {
3832                         btrfs_mark_buffer_dirty(leaf);
3833                 }
3834         }
3835         return ret;
3836 }
3837
3838 /*
3839  * search the tree again to find a leaf with lesser keys
3840  * returns 0 if it found something or 1 if there are no lesser leaves.
3841  * returns < 0 on io errors.
3842  *
3843  * This may release the path, and so you may lose any locks held at the
3844  * time you call it.
3845  */
3846 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3847 {
3848         struct btrfs_key key;
3849         struct btrfs_disk_key found_key;
3850         int ret;
3851
3852         btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3853
3854         if (key.offset > 0)
3855                 key.offset--;
3856         else if (key.type > 0)
3857                 key.type--;
3858         else if (key.objectid > 0)
3859                 key.objectid--;
3860         else
3861                 return 1;
3862
3863         btrfs_release_path(root, path);
3864         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3865         if (ret < 0)
3866                 return ret;
3867         btrfs_item_key(path->nodes[0], &found_key, 0);
3868         ret = comp_keys(&found_key, &key);
3869         if (ret < 0)
3870                 return 0;
3871         return 1;
3872 }
3873
3874 /*
3875  * A helper function to walk down the tree starting at min_key, and looking
3876  * for nodes or leaves that are either in cache or have a minimum
3877  * transaction id.  This is used by the btree defrag code, and tree logging
3878  *
3879  * This does not cow, but it does stuff the starting key it finds back
3880  * into min_key, so you can call btrfs_search_slot with cow=1 on the
3881  * key and get a writable path.
3882  *
3883  * This does lock as it descends, and path->keep_locks should be set
3884  * to 1 by the caller.
3885  *
3886  * This honors path->lowest_level to prevent descent past a given level
3887  * of the tree.
3888  *
3889  * min_trans indicates the oldest transaction that you are interested
3890  * in walking through.  Any nodes or leaves older than min_trans are
3891  * skipped over (without reading them).
3892  *
3893  * returns zero if something useful was found, < 0 on error and 1 if there
3894  * was nothing in the tree that matched the search criteria.
3895  */
3896 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3897                          struct btrfs_key *max_key,
3898                          struct btrfs_path *path, int cache_only,
3899                          u64 min_trans)
3900 {
3901         struct extent_buffer *cur;
3902         struct btrfs_key found_key;
3903         int slot;
3904         int sret;
3905         u32 nritems;
3906         int level;
3907         int ret = 1;
3908
3909         WARN_ON(!path->keep_locks);
3910 again:
3911         cur = btrfs_lock_root_node(root);
3912         level = btrfs_header_level(cur);
3913         WARN_ON(path->nodes[level]);
3914         path->nodes[level] = cur;
3915         path->locks[level] = 1;
3916
3917         if (btrfs_header_generation(cur) < min_trans) {
3918                 ret = 1;
3919                 goto out;
3920         }
3921         while (1) {
3922                 nritems = btrfs_header_nritems(cur);
3923                 level = btrfs_header_level(cur);
3924                 sret = bin_search(cur, min_key, level, &slot);
3925
3926                 /* at the lowest level, we're done, setup the path and exit */
3927                 if (level == path->lowest_level) {
3928                         if (slot >= nritems)
3929                                 goto find_next_key;
3930                         ret = 0;
3931                         path->slots[level] = slot;
3932                         btrfs_item_key_to_cpu(cur, &found_key, slot);
3933                         goto out;
3934                 }
3935                 if (sret && slot > 0)
3936                         slot--;
3937                 /*
3938                  * check this node pointer against the cache_only and
3939                  * min_trans parameters.  If it isn't in cache or is too
3940                  * old, skip to the next one.
3941                  */
3942                 while (slot < nritems) {
3943                         u64 blockptr;
3944                         u64 gen;
3945                         struct extent_buffer *tmp;
3946                         struct btrfs_disk_key disk_key;
3947
3948                         blockptr = btrfs_node_blockptr(cur, slot);
3949                         gen = btrfs_node_ptr_generation(cur, slot);
3950                         if (gen < min_trans) {
3951                                 slot++;
3952                                 continue;
3953                         }
3954                         if (!cache_only)
3955                                 break;
3956
3957                         if (max_key) {
3958                                 btrfs_node_key(cur, &disk_key, slot);
3959                                 if (comp_keys(&disk_key, max_key) >= 0) {
3960                                         ret = 1;
3961                                         goto out;
3962                                 }
3963                         }
3964
3965                         tmp = btrfs_find_tree_block(root, blockptr,
3966                                             btrfs_level_size(root, level - 1));
3967
3968                         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3969                                 free_extent_buffer(tmp);
3970                                 break;
3971                         }
3972                         if (tmp)
3973                                 free_extent_buffer(tmp);
3974                         slot++;
3975                 }
3976 find_next_key:
3977                 /*
3978                  * we didn't find a candidate key in this node, walk forward
3979                  * and find another one
3980                  */
3981                 if (slot >= nritems) {
3982                         path->slots[level] = slot;
3983                         btrfs_set_path_blocking(path);
3984                         sret = btrfs_find_next_key(root, path, min_key, level,
3985                                                   cache_only, min_trans);
3986                         if (sret == 0) {
3987                                 btrfs_release_path(root, path);
3988                                 goto again;
3989                         } else {
3990                                 goto out;
3991                         }
3992                 }
3993                 /* save our key for returning back */
3994                 btrfs_node_key_to_cpu(cur, &found_key, slot);
3995                 path->slots[level] = slot;
3996                 if (level == path->lowest_level) {
3997                         ret = 0;
3998                         unlock_up(path, level, 1);
3999                         goto out;
4000                 }
4001                 btrfs_set_path_blocking(path);
4002                 cur = read_node_slot(root, cur, slot);
4003
4004                 btrfs_tree_lock(cur);
4005
4006                 path->locks[level - 1] = 1;
4007                 path->nodes[level - 1] = cur;
4008                 unlock_up(path, level, 1);
4009                 btrfs_clear_path_blocking(path, NULL);
4010         }
4011 out:
4012         if (ret == 0)
4013                 memcpy(min_key, &found_key, sizeof(found_key));
4014         btrfs_set_path_blocking(path);
4015         return ret;
4016 }
4017
4018 /*
4019  * this is similar to btrfs_next_leaf, but does not try to preserve
4020  * and fixup the path.  It looks for and returns the next key in the
4021  * tree based on the current path and the cache_only and min_trans
4022  * parameters.
4023  *
4024  * 0 is returned if another key is found, < 0 if there are any errors
4025  * and 1 is returned if there are no higher keys in the tree
4026  *
4027  * path->keep_locks should be set to 1 on the search made before
4028  * calling this function.
4029  */
4030 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4031                         struct btrfs_key *key, int lowest_level,
4032                         int cache_only, u64 min_trans)
4033 {
4034         int level = lowest_level;
4035         int slot;
4036         struct extent_buffer *c;
4037
4038         WARN_ON(!path->keep_locks);
4039         while (level < BTRFS_MAX_LEVEL) {
4040                 if (!path->nodes[level])
4041                         return 1;
4042
4043                 slot = path->slots[level] + 1;
4044                 c = path->nodes[level];
4045 next:
4046                 if (slot >= btrfs_header_nritems(c)) {
4047                         level++;
4048                         if (level == BTRFS_MAX_LEVEL)
4049                                 return 1;
4050                         continue;
4051                 }
4052                 if (level == 0)
4053                         btrfs_item_key_to_cpu(c, key, slot);
4054                 else {
4055                         u64 blockptr = btrfs_node_blockptr(c, slot);
4056                         u64 gen = btrfs_node_ptr_generation(c, slot);
4057
4058                         if (cache_only) {
4059                                 struct extent_buffer *cur;
4060                                 cur = btrfs_find_tree_block(root, blockptr,
4061                                             btrfs_level_size(root, level - 1));
4062                                 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4063                                         slot++;
4064                                         if (cur)
4065                                                 free_extent_buffer(cur);
4066                                         goto next;
4067                                 }
4068                                 free_extent_buffer(cur);
4069                         }
4070                         if (gen < min_trans) {
4071                                 slot++;
4072                                 goto next;
4073                         }
4074                         btrfs_node_key_to_cpu(c, key, slot);
4075                 }
4076                 return 0;
4077         }
4078         return 1;
4079 }
4080
4081 /*
4082  * search the tree again to find a leaf with greater keys
4083  * returns 0 if it found something or 1 if there are no greater leaves.
4084  * returns < 0 on io errors.
4085  */
4086 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4087 {
4088         int slot;
4089         int level = 1;
4090         struct extent_buffer *c;
4091         struct extent_buffer *next = NULL;
4092         struct btrfs_key key;
4093         u32 nritems;
4094         int ret;
4095
4096         nritems = btrfs_header_nritems(path->nodes[0]);
4097         if (nritems == 0)
4098                 return 1;
4099
4100         btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4101
4102         btrfs_release_path(root, path);
4103         path->keep_locks = 1;
4104         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4105         path->keep_locks = 0;
4106
4107         if (ret < 0)
4108                 return ret;
4109
4110         btrfs_set_path_blocking(path);
4111         nritems = btrfs_header_nritems(path->nodes[0]);
4112         /*
4113          * by releasing the path above we dropped all our locks.  A balance
4114          * could have added more items next to the key that used to be
4115          * at the very end of the block.  So, check again here and
4116          * advance the path if there are now more items available.
4117          */
4118         if (nritems > 0 && path->slots[0] < nritems - 1) {
4119                 path->slots[0]++;
4120                 goto done;
4121         }
4122
4123         while (level < BTRFS_MAX_LEVEL) {
4124                 if (!path->nodes[level])
4125                         return 1;
4126
4127                 slot = path->slots[level] + 1;
4128                 c = path->nodes[level];
4129                 if (slot >= btrfs_header_nritems(c)) {
4130                         level++;
4131                         if (level == BTRFS_MAX_LEVEL)
4132                                 return 1;
4133                         continue;
4134                 }
4135
4136                 if (next) {
4137                         btrfs_tree_unlock(next);
4138                         free_extent_buffer(next);
4139                 }
4140
4141                 /* the path was set to blocking above */
4142                 if (level == 1 && (path->locks[1] || path->skip_locking) &&
4143                     path->reada)
4144                         reada_for_search(root, path, level, slot, 0);
4145
4146                 next = read_node_slot(root, c, slot);
4147                 if (!path->skip_locking) {
4148                         btrfs_assert_tree_locked(c);
4149                         btrfs_tree_lock(next);
4150                         btrfs_set_lock_blocking(next);
4151                 }
4152                 break;
4153         }
4154         path->slots[level] = slot;
4155         while (1) {
4156                 level--;
4157                 c = path->nodes[level];
4158                 if (path->locks[level])
4159                         btrfs_tree_unlock(c);
4160                 free_extent_buffer(c);
4161                 path->nodes[level] = next;
4162                 path->slots[level] = 0;
4163                 if (!path->skip_locking)
4164                         path->locks[level] = 1;
4165                 if (!level)
4166                         break;
4167
4168                 btrfs_set_path_blocking(path);
4169                 if (level == 1 && path->locks[1] && path->reada)
4170                         reada_for_search(root, path, level, slot, 0);
4171                 next = read_node_slot(root, next, 0);
4172                 if (!path->skip_locking) {
4173                         btrfs_assert_tree_locked(path->nodes[level]);
4174                         btrfs_tree_lock(next);
4175                         btrfs_set_lock_blocking(next);
4176                 }
4177         }
4178 done:
4179         unlock_up(path, 0, 1);
4180         return 0;
4181 }
4182
4183 /*
4184  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4185  * searching until it gets past min_objectid or finds an item of 'type'
4186  *
4187  * returns 0 if something is found, 1 if nothing was found and < 0 on error
4188  */
4189 int btrfs_previous_item(struct btrfs_root *root,
4190                         struct btrfs_path *path, u64 min_objectid,
4191                         int type)
4192 {
4193         struct btrfs_key found_key;
4194         struct extent_buffer *leaf;
4195         u32 nritems;
4196         int ret;
4197
4198         while (1) {
4199                 if (path->slots[0] == 0) {
4200                         btrfs_set_path_blocking(path);
4201                         ret = btrfs_prev_leaf(root, path);
4202                         if (ret != 0)
4203                                 return ret;
4204                 } else {
4205                         path->slots[0]--;
4206                 }
4207                 leaf = path->nodes[0];
4208                 nritems = btrfs_header_nritems(leaf);
4209                 if (nritems == 0)
4210                         return 1;
4211                 if (path->slots[0] == nritems)
4212                         path->slots[0]--;
4213
4214                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4215                 if (found_key.type == type)
4216                         return 0;
4217                 if (found_key.objectid < min_objectid)
4218                         break;
4219                 if (found_key.objectid == min_objectid &&
4220                     found_key.type < type)
4221                         break;
4222         }
4223         return 1;
4224 }