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                                               mid->start, child->start,
 926                                               root->root_key.objectid,
 927                                               trans->transid, level - 1);
 928                 BUG_ON(ret);
 929
 930                 add_root_to_dirty_list(root);
 931                 btrfs_tree_unlock(child);
 932
 933                 path->locks[level] = 0;
 934                 path->nodes[level] = NULL;
 935                 clean_tree_block(trans, root, mid);
 936                 btrfs_tree_unlock(mid);
 937                 /* once for the path */
 938                 free_extent_buffer(mid);
 939                 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
 940                                         mid->start, root->root_key.objectid,
 941                                         btrfs_header_generation(mid),
 942                                         level, 1);
 943                 /* once for the root ptr */
 944                 free_extent_buffer(mid);
 945                 return ret;
 946         }
 947         if (btrfs_header_nritems(mid) >
 948             BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
 949                 return 0;
 950
 951         if (btrfs_header_nritems(mid) < 2)
 952                 err_on_enospc = 1;
 953
 954         left = read_node_slot(root, parent, pslot - 1);
 955         if (left) {
 956                 btrfs_tree_lock(left);
 957                 btrfs_set_lock_blocking(left);
 958                 wret = btrfs_cow_block(trans, root, left,
 959                                        parent, pslot - 1, &left);
 960                 if (wret) {
 961                         ret = wret;
 962                         goto enospc;
 963                 }
 964         }
 965         right = read_node_slot(root, parent, pslot + 1);
 966         if (right) {
 967                 btrfs_tree_lock(right);
 968                 btrfs_set_lock_blocking(right);
 969                 wret = btrfs_cow_block(trans, root, right,
 970                                        parent, pslot + 1, &right);
 971                 if (wret) {
 972                         ret = wret;
 973                         goto enospc;
 974                 }
 975         }
 976
 977         /* first, try to make some room in the middle buffer */
 978         if (left) {
 979                 orig_slot += btrfs_header_nritems(left);
 980                 wret = push_node_left(trans, root, left, mid, 1);
 981                 if (wret < 0)
 982                         ret = wret;
 983                 if (btrfs_header_nritems(mid) < 2)
 984                         err_on_enospc = 1;
 985         }
 986
 987         /*
 988          * then try to empty the right most buffer into the middle
 989          */
 990         if (right) {
 991                 wret = push_node_left(trans, root, mid, right, 1);
 992                 if (wret < 0 && wret != -ENOSPC)
 993                         ret = wret;
 994                 if (btrfs_header_nritems(right) == 0) {
 995                         u64 bytenr = right->start;
 996                         u64 generation = btrfs_header_generation(parent);
 997                         u32 blocksize = right->len;
 998
 999                         clean_tree_block(trans, root, right);
1000                         btrfs_tree_unlock(right);
1001                         free_extent_buffer(right);
1002                         right = NULL;
1003                         wret = del_ptr(trans, root, path, level + 1, pslot +
1004                                        1);
1005                         if (wret)
1006                                 ret = wret;
1007                         wret = btrfs_free_extent(trans, root, bytenr,
1008                                                  blocksize, parent->start,
1009                                                  btrfs_header_owner(parent),
1010                                                  generation, level, 1);
1011                         if (wret)
1012                                 ret = wret;
1013                 } else {
1014                         struct btrfs_disk_key right_key;
1015                         btrfs_node_key(right, &right_key, 0);
1016                         btrfs_set_node_key(parent, &right_key, pslot + 1);
1017                         btrfs_mark_buffer_dirty(parent);
1018                 }
1019         }
1020         if (btrfs_header_nritems(mid) == 1) {
1021                 /*
1022                  * we're not allowed to leave a node with one item in the
1023                  * tree during a delete.  A deletion from lower in the tree
1024                  * could try to delete the only pointer in this node.
1025                  * So, pull some keys from the left.
1026                  * There has to be a left pointer at this point because
1027                  * otherwise we would have pulled some pointers from the
1028                  * right
1029                  */
1030                 BUG_ON(!left);
1031                 wret = balance_node_right(trans, root, mid, left);
1032                 if (wret < 0) {
1033                         ret = wret;
1034                         goto enospc;
1035                 }
1036                 if (wret == 1) {
1037                         wret = push_node_left(trans, root, left, mid, 1);
1038                         if (wret < 0)
1039                                 ret = wret;
1040                 }
1041                 BUG_ON(wret == 1);
1042         }
1043         if (btrfs_header_nritems(mid) == 0) {
1044                 /* we've managed to empty the middle node, drop it */
1045                 u64 root_gen = btrfs_header_generation(parent);
1046                 u64 bytenr = mid->start;
1047                 u32 blocksize = mid->len;
1048
1049                 clean_tree_block(trans, root, mid);
1050                 btrfs_tree_unlock(mid);
1051                 free_extent_buffer(mid);
1052                 mid = NULL;
1053                 wret = del_ptr(trans, root, path, level + 1, pslot);
1054                 if (wret)
1055                         ret = wret;
1056                 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
1057                                          parent->start,
1058                                          btrfs_header_owner(parent),
1059                                          root_gen, level, 1);
1060                 if (wret)
1061                         ret = wret;
1062         } else {
1063                 /* update the parent key to reflect our changes */
1064                 struct btrfs_disk_key mid_key;
1065                 btrfs_node_key(mid, &mid_key, 0);
1066                 btrfs_set_node_key(parent, &mid_key, pslot);
1067                 btrfs_mark_buffer_dirty(parent);
1068         }
1069
1070         /* update the path */
1071         if (left) {
1072                 if (btrfs_header_nritems(left) > orig_slot) {
1073                         extent_buffer_get(left);
1074                         /* left was locked after cow */
1075                         path->nodes[level] = left;
1076                         path->slots[level + 1] -= 1;
1077                         path->slots[level] = orig_slot;
1078                         if (mid) {
1079                                 btrfs_tree_unlock(mid);
1080                                 free_extent_buffer(mid);
1081                         }
1082                 } else {
1083                         orig_slot -= btrfs_header_nritems(left);
1084                         path->slots[level] = orig_slot;
1085                 }
1086         }
1087         /* double check we haven't messed things up */
1088         check_block(root, path, level);
1089         if (orig_ptr !=
1090             btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1091                 BUG();
1092 enospc:
1093         if (right) {
1094                 btrfs_tree_unlock(right);
1095                 free_extent_buffer(right);
1096         }
1097         if (left) {
1098                 if (path->nodes[level] != left)
1099                         btrfs_tree_unlock(left);
1100                 free_extent_buffer(left);
1101         }
1102         return ret;
1103 }
1104
1105 /* Node balancing for insertion.  Here we only split or push nodes around
1106  * when they are completely full.  This is also done top down, so we
1107  * have to be pessimistic.
1108  */
1109 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1110                                           struct btrfs_root *root,
1111                                           struct btrfs_path *path, int level)
1112 {
1113         struct extent_buffer *right = NULL;
1114         struct extent_buffer *mid;
1115         struct extent_buffer *left = NULL;
1116         struct extent_buffer *parent = NULL;
1117         int ret = 0;
1118         int wret;
1119         int pslot;
1120         int orig_slot = path->slots[level];
1121         u64 orig_ptr;
1122
1123         if (level == 0)
1124                 return 1;
1125
1126         mid = path->nodes[level];
1127         WARN_ON(btrfs_header_generation(mid) != trans->transid);
1128         orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1129
1130         if (level < BTRFS_MAX_LEVEL - 1)
1131                 parent = path->nodes[level + 1];
1132         pslot = path->slots[level + 1];
1133
1134         if (!parent)
1135                 return 1;
1136
1137         left = read_node_slot(root, parent, pslot - 1);
1138
1139         /* first, try to make some room in the middle buffer */
1140         if (left) {
1141                 u32 left_nr;
1142
1143                 btrfs_tree_lock(left);
1144                 btrfs_set_lock_blocking(left);
1145
1146                 left_nr = btrfs_header_nritems(left);
1147                 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1148                         wret = 1;
1149                 } else {
1150                         ret = btrfs_cow_block(trans, root, left, parent,
1151                                               pslot - 1, &left);
1152                         if (ret)
1153                                 wret = 1;
1154                         else {
1155                                 wret = push_node_left(trans, root,
1156                                                       left, mid, 0);
1157                         }
1158                 }
1159                 if (wret < 0)
1160                         ret = wret;
1161                 if (wret == 0) {
1162                         struct btrfs_disk_key disk_key;
1163                         orig_slot += left_nr;
1164                         btrfs_node_key(mid, &disk_key, 0);
1165                         btrfs_set_node_key(parent, &disk_key, pslot);
1166                         btrfs_mark_buffer_dirty(parent);
1167                         if (btrfs_header_nritems(left) > orig_slot) {
1168                                 path->nodes[level] = left;
1169                                 path->slots[level + 1] -= 1;
1170                                 path->slots[level] = orig_slot;
1171                                 btrfs_tree_unlock(mid);
1172                                 free_extent_buffer(mid);
1173                         } else {
1174                                 orig_slot -=
1175                                         btrfs_header_nritems(left);
1176                                 path->slots[level] = orig_slot;
1177                                 btrfs_tree_unlock(left);
1178                                 free_extent_buffer(left);
1179                         }
1180                         return 0;
1181                 }
1182                 btrfs_tree_unlock(left);
1183                 free_extent_buffer(left);
1184         }
1185         right = read_node_slot(root, parent, pslot + 1);
1186
1187         /*
1188          * then try to empty the right most buffer into the middle
1189          */
1190         if (right) {
1191                 u32 right_nr;
1192
1193                 btrfs_tree_lock(right);
1194                 btrfs_set_lock_blocking(right);
1195
1196                 right_nr = btrfs_header_nritems(right);
1197                 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1198                         wret = 1;
1199                 } else {
1200                         ret = btrfs_cow_block(trans, root, right,
1201                                               parent, pslot + 1,
1202                                               &right);
1203                         if (ret)
1204                                 wret = 1;
1205                         else {
1206                                 wret = balance_node_right(trans, root,
1207                                                           right, mid);
1208                         }
1209                 }
1210                 if (wret < 0)
1211                         ret = wret;
1212                 if (wret == 0) {
1213                         struct btrfs_disk_key disk_key;
1214
1215                         btrfs_node_key(right, &disk_key, 0);
1216                         btrfs_set_node_key(parent, &disk_key, pslot + 1);
1217                         btrfs_mark_buffer_dirty(parent);
1218
1219                         if (btrfs_header_nritems(mid) <= orig_slot) {
1220                                 path->nodes[level] = right;
1221                                 path->slots[level + 1] += 1;
1222                                 path->slots[level] = orig_slot -
1223                                         btrfs_header_nritems(mid);
1224                                 btrfs_tree_unlock(mid);
1225                                 free_extent_buffer(mid);
1226                         } else {
1227                                 btrfs_tree_unlock(right);
1228                                 free_extent_buffer(right);
1229                         }
1230                         return 0;
1231                 }
1232                 btrfs_tree_unlock(right);
1233                 free_extent_buffer(right);
1234         }
1235         return 1;
1236 }
1237
1238 /*
1239  * readahead one full node of leaves, finding things that are close
1240  * to the block in 'slot', and triggering ra on them.
1241  */
1242 static noinline void reada_for_search(struct btrfs_root *root,
1243                                       struct btrfs_path *path,
1244                                       int level, int slot, u64 objectid)
1245 {
1246         struct extent_buffer *node;
1247         struct btrfs_disk_key disk_key;
1248         u32 nritems;
1249         u64 search;
1250         u64 target;
1251         u64 nread = 0;
1252         int direction = path->reada;
1253         struct extent_buffer *eb;
1254         u32 nr;
1255         u32 blocksize;
1256         u32 nscan = 0;
1257
1258         if (level != 1)
1259                 return;
1260
1261         if (!path->nodes[level])
1262                 return;
1263
1264         node = path->nodes[level];
1265
1266         search = btrfs_node_blockptr(node, slot);
1267         blocksize = btrfs_level_size(root, level - 1);
1268         eb = btrfs_find_tree_block(root, search, blocksize);
1269         if (eb) {
1270                 free_extent_buffer(eb);
1271                 return;
1272         }
1273
1274         target = search;
1275
1276         nritems = btrfs_header_nritems(node);
1277         nr = slot;
1278         while (1) {
1279                 if (direction < 0) {
1280                         if (nr == 0)
1281                                 break;
1282                         nr--;
1283                 } else if (direction > 0) {
1284                         nr++;
1285                         if (nr >= nritems)
1286                                 break;
1287                 }
1288                 if (path->reada < 0 && objectid) {
1289                         btrfs_node_key(node, &disk_key, nr);
1290                         if (btrfs_disk_key_objectid(&disk_key) != objectid)
1291                                 break;
1292                 }
1293                 search = btrfs_node_blockptr(node, nr);
1294                 if ((search <= target && target - search <= 65536) ||
1295                     (search > target && search - target <= 65536)) {
1296                         readahead_tree_block(root, search, blocksize,
1297                                      btrfs_node_ptr_generation(node, nr));
1298                         nread += blocksize;
1299                 }
1300                 nscan++;
1301                 if ((nread > 65536 || nscan > 32))
1302                         break;
1303         }
1304 }
1305
1306 /*
1307  * returns -EAGAIN if it had to drop the path, or zero if everything was in
1308  * cache
1309  */
1310 static noinline int reada_for_balance(struct btrfs_root *root,
1311                                       struct btrfs_path *path, int level)
1312 {
1313         int slot;
1314         int nritems;
1315         struct extent_buffer *parent;
1316         struct extent_buffer *eb;
1317         u64 gen;
1318         u64 block1 = 0;
1319         u64 block2 = 0;
1320         int ret = 0;
1321         int blocksize;
1322
1323         parent = path->nodes[level - 1];
1324         if (!parent)
1325                 return 0;
1326
1327         nritems = btrfs_header_nritems(parent);
1328         slot = path->slots[level];
1329         blocksize = btrfs_level_size(root, level);
1330
1331         if (slot > 0) {
1332                 block1 = btrfs_node_blockptr(parent, slot - 1);
1333                 gen = btrfs_node_ptr_generation(parent, slot - 1);
1334                 eb = btrfs_find_tree_block(root, block1, blocksize);
1335                 if (eb && btrfs_buffer_uptodate(eb, gen))
1336                         block1 = 0;
1337                 free_extent_buffer(eb);
1338         }
1339         if (slot < nritems) {
1340                 block2 = btrfs_node_blockptr(parent, slot + 1);
1341                 gen = btrfs_node_ptr_generation(parent, slot + 1);
1342                 eb = btrfs_find_tree_block(root, block2, blocksize);
1343                 if (eb && btrfs_buffer_uptodate(eb, gen))
1344                         block2 = 0;
1345                 free_extent_buffer(eb);
1346         }
1347         if (block1 || block2) {
1348                 ret = -EAGAIN;
1349                 btrfs_release_path(root, path);
1350                 if (block1)
1351                         readahead_tree_block(root, block1, blocksize, 0);
1352                 if (block2)
1353                         readahead_tree_block(root, block2, blocksize, 0);
1354
1355                 if (block1) {
1356                         eb = read_tree_block(root, block1, blocksize, 0);
1357                         free_extent_buffer(eb);
1358                 }
1359                 if (block1) {
1360                         eb = read_tree_block(root, block2, blocksize, 0);
1361                         free_extent_buffer(eb);
1362                 }
1363         }
1364         return ret;
1365 }
1366
1367
1368 /*
1369  * when we walk down the tree, it is usually safe to unlock the higher layers
1370  * in the tree.  The exceptions are when our path goes through slot 0, because
1371  * operations on the tree might require changing key pointers higher up in the
1372  * tree.
1373  *
1374  * callers might also have set path->keep_locks, which tells this code to keep
1375  * the lock if the path points to the last slot in the block.  This is part of
1376  * walking through the tree, and selecting the next slot in the higher block.
1377  *
1378  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
1379  * if lowest_unlock is 1, level 0 won't be unlocked
1380  */
1381 static noinline void unlock_up(struct btrfs_path *path, int level,
1382                                int lowest_unlock)
1383 {
1384         int i;
1385         int skip_level = level;
1386         int no_skips = 0;
1387         struct extent_buffer *t;
1388
1389         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1390                 if (!path->nodes[i])
1391                         break;
1392                 if (!path->locks[i])
1393                         break;
1394                 if (!no_skips && path->slots[i] == 0) {
1395                         skip_level = i + 1;
1396                         continue;
1397                 }
1398                 if (!no_skips && path->keep_locks) {
1399                         u32 nritems;
1400                         t = path->nodes[i];
1401                         nritems = btrfs_header_nritems(t);
1402                         if (nritems < 1 || path->slots[i] >= nritems - 1) {
1403                                 skip_level = i + 1;
1404                                 continue;
1405                         }
1406                 }
1407                 if (skip_level < i && i >= lowest_unlock)
1408                         no_skips = 1;
1409
1410                 t = path->nodes[i];
1411                 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1412                         btrfs_tree_unlock(t);
1413                         path->locks[i] = 0;
1414                 }
1415         }
1416 }
1417
1418 /*
1419  * This releases any locks held in the path starting at level and
1420  * going all the way up to the root.
1421  *
1422  * btrfs_search_slot will keep the lock held on higher nodes in a few
1423  * corner cases, such as COW of the block at slot zero in the node.  This
1424  * ignores those rules, and it should only be called when there are no
1425  * more updates to be done higher up in the tree.
1426  */
1427 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1428 {
1429         int i;
1430
1431         if (path->keep_locks || path->lowest_level)
1432                 return;
1433
1434         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1435                 if (!path->nodes[i])
1436                         continue;
1437                 if (!path->locks[i])
1438                         continue;
1439                 btrfs_tree_unlock(path->nodes[i]);
1440                 path->locks[i] = 0;
1441         }
1442 }
1443
1444 /*
1445  * look for key in the tree.  path is filled in with nodes along the way
1446  * if key is found, we return zero and you can find the item in the leaf
1447  * level of the path (level 0)
1448  *
1449  * If the key isn't found, the path points to the slot where it should
1450  * be inserted, and 1 is returned.  If there are other errors during the
1451  * search a negative error number is returned.
1452  *
1453  * if ins_len > 0, nodes and leaves will be split as we walk down the
1454  * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
1455  * possible)
1456  */
1457 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1458                       *root, struct btrfs_key *key, struct btrfs_path *p, int
1459                       ins_len, int cow)
1460 {
1461         struct extent_buffer *b;
1462         struct extent_buffer *tmp;
1463         int slot;
1464         int ret;
1465         int level;
1466         int should_reada = p->reada;
1467         int lowest_unlock = 1;
1468         int blocksize;
1469         u8 lowest_level = 0;
1470         u64 blocknr;
1471         u64 gen;
1472
1473         lowest_level = p->lowest_level;
1474         WARN_ON(lowest_level && ins_len > 0);
1475         WARN_ON(p->nodes[0] != NULL);
1476
1477         if (ins_len < 0)
1478                 lowest_unlock = 2;
1479
1480 again:
1481         if (p->skip_locking)
1482                 b = btrfs_root_node(root);
1483         else
1484                 b = btrfs_lock_root_node(root);
1485
1486         while (b) {
1487                 level = btrfs_header_level(b);
1488
1489                 /*
1490                  * setup the path here so we can release it under lock
1491                  * contention with the cow code
1492                  */
1493                 p->nodes[level] = b;
1494                 if (!p->skip_locking)
1495                         p->locks[level] = 1;
1496
1497                 if (cow) {
1498                         int wret;
1499
1500                         /* is a cow on this block not required */
1501                         if (btrfs_header_generation(b) == trans->transid &&
1502                             btrfs_header_owner(b) == root->root_key.objectid &&
1503                             !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1504                                 goto cow_done;
1505                         }
1506                         btrfs_set_path_blocking(p);
1507
1508                         wret = btrfs_cow_block(trans, root, b,
1509                                                p->nodes[level + 1],
1510                                                p->slots[level + 1], &b);
1511                         if (wret) {
1512                                 free_extent_buffer(b);
1513                                 ret = wret;
1514                                 goto done;
1515                         }
1516                 }
1517 cow_done:
1518                 BUG_ON(!cow && ins_len);
1519                 if (level != btrfs_header_level(b))
1520                         WARN_ON(1);
1521                 level = btrfs_header_level(b);
1522
1523                 p->nodes[level] = b;
1524                 if (!p->skip_locking)
1525                         p->locks[level] = 1;
1526
1527                 btrfs_clear_path_blocking(p, NULL);
1528
1529                 /*
1530                  * we have a lock on b and as long as we aren't changing
1531                  * the tree, there is no way to for the items in b to change.
1532                  * It is safe to drop the lock on our parent before we
1533                  * go through the expensive btree search on b.
1534                  *
1535                  * If cow is true, then we might be changing slot zero,
1536                  * which may require changing the parent.  So, we can't
1537                  * drop the lock until after we know which slot we're
1538                  * operating on.
1539                  */
1540                 if (!cow)
1541                         btrfs_unlock_up_safe(p, level + 1);
1542
1543                 ret = check_block(root, p, level);
1544                 if (ret) {
1545                         ret = -1;
1546                         goto done;
1547                 }
1548
1549                 ret = bin_search(b, key, level, &slot);
1550
1551                 if (level != 0) {
1552                         if (ret && slot > 0)
1553                                 slot -= 1;
1554                         p->slots[level] = slot;
1555                         if ((p->search_for_split || ins_len > 0) &&
1556                             btrfs_header_nritems(b) >=
1557                             BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1558                                 int sret;
1559
1560                                 sret = reada_for_balance(root, p, level);
1561                                 if (sret)
1562                                         goto again;
1563
1564                                 btrfs_set_path_blocking(p);
1565                                 sret = split_node(trans, root, p, level);
1566                                 btrfs_clear_path_blocking(p, NULL);
1567
1568                                 BUG_ON(sret > 0);
1569                                 if (sret) {
1570                                         ret = sret;
1571                                         goto done;
1572                                 }
1573                                 b = p->nodes[level];
1574                                 slot = p->slots[level];
1575                         } else if (ins_len < 0 &&
1576                                    btrfs_header_nritems(b) <
1577                                    BTRFS_NODEPTRS_PER_BLOCK(root) / 4) {
1578                                 int sret;
1579
1580                                 sret = reada_for_balance(root, p, level);
1581                                 if (sret)
1582                                         goto again;
1583
1584                                 btrfs_set_path_blocking(p);
1585                                 sret = balance_level(trans, root, p, level);
1586                                 btrfs_clear_path_blocking(p, NULL);
1587
1588                                 if (sret) {
1589                                         ret = sret;
1590                                         goto done;
1591                                 }
1592                                 b = p->nodes[level];
1593                                 if (!b) {
1594                                         btrfs_release_path(NULL, p);
1595                                         goto again;
1596                                 }
1597                                 slot = p->slots[level];
1598                                 BUG_ON(btrfs_header_nritems(b) == 1);
1599                         }
1600                         unlock_up(p, level, lowest_unlock);
1601
1602                         /* this is only true while dropping a snapshot */
1603                         if (level == lowest_level) {
1604                                 ret = 0;
1605                                 goto done;
1606                         }
1607
1608                         blocknr = btrfs_node_blockptr(b, slot);
1609                         gen = btrfs_node_ptr_generation(b, slot);
1610                         blocksize = btrfs_level_size(root, level - 1);
1611
1612                         tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1613                         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1614                                 b = tmp;
1615                         } else {
1616                                 /*
1617                                  * reduce lock contention at high levels
1618                                  * of the btree by dropping locks before
1619                                  * we read.
1620                                  */
1621                                 if (level > 0) {
1622                                         btrfs_release_path(NULL, p);
1623                                         if (tmp)
1624                                                 free_extent_buffer(tmp);
1625                                         if (should_reada)
1626                                                 reada_for_search(root, p,
1627                                                                  level, slot,
1628                                                                  key->objectid);
1629
1630                                         tmp = read_tree_block(root, blocknr,
1631                                                          blocksize, gen);
1632                                         if (tmp)
1633                                                 free_extent_buffer(tmp);
1634                                         goto again;
1635                                 } else {
1636                                         btrfs_set_path_blocking(p);
1637                                         if (tmp)
1638                                                 free_extent_buffer(tmp);
1639                                         if (should_reada)
1640                                                 reada_for_search(root, p,
1641                                                                  level, slot,
1642                                                                  key->objectid);
1643                                         b = read_node_slot(root, b, slot);
1644                                 }
1645                         }
1646                         if (!p->skip_locking) {
1647                                 int lret;
1648
1649                                 btrfs_clear_path_blocking(p, NULL);
1650                                 lret = btrfs_try_spin_lock(b);
1651
1652                                 if (!lret) {
1653                                         btrfs_set_path_blocking(p);
1654                                         btrfs_tree_lock(b);
1655                                         btrfs_clear_path_blocking(p, b);
1656                                 }
1657                         }
1658                 } else {
1659                         p->slots[level] = slot;
1660                         if (ins_len > 0 &&
1661                             btrfs_leaf_free_space(root, b) < ins_len) {
1662                                 int sret;
1663
1664                                 btrfs_set_path_blocking(p);
1665                                 sret = split_leaf(trans, root, key,
1666                                                       p, ins_len, ret == 0);
1667                                 btrfs_clear_path_blocking(p, NULL);
1668
1669                                 BUG_ON(sret > 0);
1670                                 if (sret) {
1671                                         ret = sret;
1672                                         goto done;
1673                                 }
1674                         }
1675                         if (!p->search_for_split)
1676                                 unlock_up(p, level, lowest_unlock);
1677                         goto done;
1678                 }
1679         }
1680         ret = 1;
1681 done:
1682         /*
1683          * we don't really know what they plan on doing with the path
1684          * from here on, so for now just mark it as blocking
1685          */
1686         btrfs_set_path_blocking(p);
1687         return ret;
1688 }
1689
1690 int btrfs_merge_path(struct btrfs_trans_handle *trans,
1691                      struct btrfs_root *root,
1692                      struct btrfs_key *node_keys,
1693                      u64 *nodes, int lowest_level)
1694 {
1695         struct extent_buffer *eb;
1696         struct extent_buffer *parent;
1697         struct btrfs_key key;
1698         u64 bytenr;
1699         u64 generation;
1700         u32 blocksize;
1701         int level;
1702         int slot;
1703         int key_match;
1704         int ret;
1705
1706         eb = btrfs_lock_root_node(root);
1707         ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb);
1708         BUG_ON(ret);
1709
1710         btrfs_set_lock_blocking(eb);
1711
1712         parent = eb;
1713         while (1) {
1714                 level = btrfs_header_level(parent);
1715                 if (level == 0 || level <= lowest_level)
1716                         break;
1717
1718                 ret = bin_search(parent, &node_keys[lowest_level], level,
1719                                  &slot);
1720                 if (ret && slot > 0)
1721                         slot--;
1722
1723                 bytenr = btrfs_node_blockptr(parent, slot);
1724                 if (nodes[level - 1] == bytenr)
1725                         break;
1726
1727                 blocksize = btrfs_level_size(root, level - 1);
1728                 generation = btrfs_node_ptr_generation(parent, slot);
1729                 btrfs_node_key_to_cpu(eb, &key, slot);
1730                 key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
1731
1732                 if (generation == trans->transid) {
1733                         eb = read_tree_block(root, bytenr, blocksize,
1734                                              generation);
1735                         btrfs_tree_lock(eb);
1736                         btrfs_set_lock_blocking(eb);
1737                 }
1738
1739                 /*
1740                  * if node keys match and node pointer hasn't been modified
1741                  * in the running transaction, we can merge the path. for
1742                  * blocks owened by reloc trees, the node pointer check is
1743                  * skipped, this is because these blocks are fully controlled
1744                  * by the space balance code, no one else can modify them.
1745                  */
1746                 if (!nodes[level - 1] || !key_match ||
1747                     (generation == trans->transid &&
1748                      btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
1749                         if (level == 1 || level == lowest_level + 1) {
1750                                 if (generation == trans->transid) {
1751                                         btrfs_tree_unlock(eb);
1752                                         free_extent_buffer(eb);
1753                                 }
1754                                 break;
1755                         }
1756
1757                         if (generation != trans->transid) {
1758                                 eb = read_tree_block(root, bytenr, blocksize,
1759                                                 generation);
1760                                 btrfs_tree_lock(eb);
1761                                 btrfs_set_lock_blocking(eb);
1762                         }
1763
1764                         ret = btrfs_cow_block(trans, root, eb, parent, slot,
1765                                               &eb);
1766                         BUG_ON(ret);
1767
1768                         if (root->root_key.objectid ==
1769                             BTRFS_TREE_RELOC_OBJECTID) {
1770                                 if (!nodes[level - 1]) {
1771                                         nodes[level - 1] = eb->start;
1772                                         memcpy(&node_keys[level - 1], &key,
1773                                                sizeof(node_keys[0]));
1774                                 } else {
1775                                         WARN_ON(1);
1776                                 }
1777                         }
1778
1779                         btrfs_tree_unlock(parent);
1780                         free_extent_buffer(parent);
1781                         parent = eb;
1782                         continue;
1783                 }
1784
1785                 btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
1786                 btrfs_set_node_ptr_generation(parent, slot, trans->transid);
1787                 btrfs_mark_buffer_dirty(parent);
1788
1789                 ret = btrfs_inc_extent_ref(trans, root,
1790                                         nodes[level - 1],
1791                                         blocksize, parent->start,
1792                                         btrfs_header_owner(parent),
1793                                         btrfs_header_generation(parent),
1794                                         level - 1);
1795                 BUG_ON(ret);
1796
1797                 /*
1798                  * If the block was created in the running transaction,
1799                  * it's possible this is the last reference to it, so we
1800                  * should drop the subtree.
1801                  */
1802                 if (generation == trans->transid) {
1803                         ret = btrfs_drop_subtree(trans, root, eb, parent);
1804                         BUG_ON(ret);
1805                         btrfs_tree_unlock(eb);
1806                         free_extent_buffer(eb);
1807                 } else {
1808                         ret = btrfs_free_extent(trans, root, bytenr,
1809                                         blocksize, parent->start,
1810                                         btrfs_header_owner(parent),
1811                                         btrfs_header_generation(parent),
1812                                         level - 1, 1);
1813                         BUG_ON(ret);
1814                 }
1815                 break;
1816         }
1817         btrfs_tree_unlock(parent);
1818         free_extent_buffer(parent);
1819         return 0;
1820 }
1821
1822 /*
1823  * adjust the pointers going up the tree, starting at level
1824  * making sure the right key of each node is points to 'key'.
1825  * This is used after shifting pointers to the left, so it stops
1826  * fixing up pointers when a given leaf/node is not in slot 0 of the
1827  * higher levels
1828  *
1829  * If this fails to write a tree block, it returns -1, but continues
1830  * fixing up the blocks in ram so the tree is consistent.
1831  */
1832 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1833                           struct btrfs_root *root, struct btrfs_path *path,
1834                           struct btrfs_disk_key *key, int level)
1835 {
1836         int i;
1837         int ret = 0;
1838         struct extent_buffer *t;
1839
1840         for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1841                 int tslot = path->slots[i];
1842                 if (!path->nodes[i])
1843                         break;
1844                 t = path->nodes[i];
1845                 btrfs_set_node_key(t, key, tslot);
1846                 btrfs_mark_buffer_dirty(path->nodes[i]);
1847                 if (tslot != 0)
1848                         break;
1849         }
1850         return ret;
1851 }
1852
1853 /*
1854  * update item key.
1855  *
1856  * This function isn't completely safe. It's the caller's responsibility
1857  * that the new key won't break the order
1858  */
1859 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1860                             struct btrfs_root *root, struct btrfs_path *path,
1861                             struct btrfs_key *new_key)
1862 {
1863         struct btrfs_disk_key disk_key;
1864         struct extent_buffer *eb;
1865         int slot;
1866
1867         eb = path->nodes[0];
1868         slot = path->slots[0];
1869         if (slot > 0) {
1870                 btrfs_item_key(eb, &disk_key, slot - 1);
1871                 if (comp_keys(&disk_key, new_key) >= 0)
1872                         return -1;
1873         }
1874         if (slot < btrfs_header_nritems(eb) - 1) {
1875                 btrfs_item_key(eb, &disk_key, slot + 1);
1876                 if (comp_keys(&disk_key, new_key) <= 0)
1877                         return -1;
1878         }
1879
1880         btrfs_cpu_key_to_disk(&disk_key, new_key);
1881         btrfs_set_item_key(eb, &disk_key, slot);
1882         btrfs_mark_buffer_dirty(eb);
1883         if (slot == 0)
1884                 fixup_low_keys(trans, root, path, &disk_key, 1);
1885         return 0;
1886 }
1887
1888 /*
1889  * try to push data from one node into the next node left in the
1890  * tree.
1891  *
1892  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1893  * error, and > 0 if there was no room in the left hand block.
1894  */
1895 static int push_node_left(struct btrfs_trans_handle *trans,
1896                           struct btrfs_root *root, struct extent_buffer *dst,
1897                           struct extent_buffer *src, int empty)
1898 {
1899         int push_items = 0;
1900         int src_nritems;
1901         int dst_nritems;
1902         int ret = 0;
1903
1904         src_nritems = btrfs_header_nritems(src);
1905         dst_nritems = btrfs_header_nritems(dst);
1906         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1907         WARN_ON(btrfs_header_generation(src) != trans->transid);
1908         WARN_ON(btrfs_header_generation(dst) != trans->transid);
1909
1910         if (!empty && src_nritems <= 8)
1911                 return 1;
1912
1913         if (push_items <= 0)
1914                 return 1;
1915
1916         if (empty) {
1917                 push_items = min(src_nritems, push_items);
1918                 if (push_items < src_nritems) {
1919                         /* leave at least 8 pointers in the node if
1920                          * we aren't going to empty it
1921                          */
1922                         if (src_nritems - push_items < 8) {
1923                                 if (push_items <= 8)
1924                                         return 1;
1925                                 push_items -= 8;
1926                         }
1927                 }
1928         } else
1929                 push_items = min(src_nritems - 8, push_items);
1930
1931         copy_extent_buffer(dst, src,
1932                            btrfs_node_key_ptr_offset(dst_nritems),
1933                            btrfs_node_key_ptr_offset(0),
1934                            push_items * sizeof(struct btrfs_key_ptr));
1935
1936         if (push_items < src_nritems) {
1937                 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1938                                       btrfs_node_key_ptr_offset(push_items),
1939                                       (src_nritems - push_items) *
1940                                       sizeof(struct btrfs_key_ptr));
1941         }
1942         btrfs_set_header_nritems(src, src_nritems - push_items);
1943         btrfs_set_header_nritems(dst, dst_nritems + push_items);
1944         btrfs_mark_buffer_dirty(src);
1945         btrfs_mark_buffer_dirty(dst);
1946
1947         ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
1948         BUG_ON(ret);
1949
1950         return ret;
1951 }
1952
1953 /*
1954  * try to push data from one node into the next node right in the
1955  * tree.
1956  *
1957  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1958  * error, and > 0 if there was no room in the right hand block.
1959  *
1960  * this will  only push up to 1/2 the contents of the left node over
1961  */
1962 static int balance_node_right(struct btrfs_trans_handle *trans,
1963                               struct btrfs_root *root,
1964                               struct extent_buffer *dst,
1965                               struct extent_buffer *src)
1966 {
1967         int push_items = 0;
1968         int max_push;
1969         int src_nritems;
1970         int dst_nritems;
1971         int ret = 0;
1972
1973         WARN_ON(btrfs_header_generation(src) != trans->transid);
1974         WARN_ON(btrfs_header_generation(dst) != trans->transid);
1975
1976         src_nritems = btrfs_header_nritems(src);
1977         dst_nritems = btrfs_header_nritems(dst);
1978         push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1979         if (push_items <= 0)
1980                 return 1;
1981
1982         if (src_nritems < 4)
1983                 return 1;
1984
1985         max_push = src_nritems / 2 + 1;
1986         /* don't try to empty the node */
1987         if (max_push >= src_nritems)
1988                 return 1;
1989
1990         if (max_push < push_items)
1991                 push_items = max_push;
1992
1993         memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1994                                       btrfs_node_key_ptr_offset(0),
1995                                       (dst_nritems) *
1996                                       sizeof(struct btrfs_key_ptr));
1997
1998         copy_extent_buffer(dst, src,
1999                            btrfs_node_key_ptr_offset(0),
2000                            btrfs_node_key_ptr_offset(src_nritems - push_items),
2001                            push_items * sizeof(struct btrfs_key_ptr));
2002
2003         btrfs_set_header_nritems(src, src_nritems - push_items);
2004         btrfs_set_header_nritems(dst, dst_nritems + push_items);
2005
2006         btrfs_mark_buffer_dirty(src);
2007         btrfs_mark_buffer_dirty(dst);
2008
2009         ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
2010         BUG_ON(ret);
2011
2012         return ret;
2013 }
2014
2015 /*
2016  * helper function to insert a new root level in the tree.
2017  * A new node is allocated, and a single item is inserted to
2018  * point to the existing root
2019  *
2020  * returns zero on success or < 0 on failure.
2021  */
2022 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2023                            struct btrfs_root *root,
2024                            struct btrfs_path *path, int level)
2025 {
2026         u64 lower_gen;
2027         struct extent_buffer *lower;
2028         struct extent_buffer *c;
2029         struct extent_buffer *old;
2030         struct btrfs_disk_key lower_key;
2031         int ret;
2032
2033         BUG_ON(path->nodes[level]);
2034         BUG_ON(path->nodes[level-1] != root->node);
2035
2036         lower = path->nodes[level-1];
2037         if (level == 1)
2038                 btrfs_item_key(lower, &lower_key, 0);
2039         else
2040                 btrfs_node_key(lower, &lower_key, 0);
2041
2042         c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2043                                    root->root_key.objectid, trans->transid,
2044                                    level, root->node->start, 0);
2045         if (IS_ERR(c))
2046                 return PTR_ERR(c);
2047
2048         memset_extent_buffer(c, 0, 0, root->nodesize);
2049         btrfs_set_header_nritems(c, 1);
2050         btrfs_set_header_level(c, level);
2051         btrfs_set_header_bytenr(c, c->start);
2052         btrfs_set_header_generation(c, trans->transid);
2053         btrfs_set_header_owner(c, root->root_key.objectid);
2054
2055         write_extent_buffer(c, root->fs_info->fsid,
2056                             (unsigned long)btrfs_header_fsid(c),
2057                             BTRFS_FSID_SIZE);
2058
2059         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2060                             (unsigned long)btrfs_header_chunk_tree_uuid(c),
2061                             BTRFS_UUID_SIZE);
2062
2063         btrfs_set_node_key(c, &lower_key, 0);
2064         btrfs_set_node_blockptr(c, 0, lower->start);
2065         lower_gen = btrfs_header_generation(lower);
2066         WARN_ON(lower_gen != trans->transid);
2067
2068         btrfs_set_node_ptr_generation(c, 0, lower_gen);
2069
2070         btrfs_mark_buffer_dirty(c);
2071
2072         spin_lock(&root->node_lock);
2073         old = root->node;
2074         root->node = c;
2075         spin_unlock(&root->node_lock);
2076
2077         ret = btrfs_update_extent_ref(trans, root, lower->start,
2078                                       lower->start, c->start,
2079                                       root->root_key.objectid,
2080                                       trans->transid, level - 1);
2081         BUG_ON(ret);
2082
2083         /* the super has an extra ref to root->node */
2084         free_extent_buffer(old);
2085
2086         add_root_to_dirty_list(root);
2087         extent_buffer_get(c);
2088         path->nodes[level] = c;
2089         path->locks[level] = 1;
2090         path->slots[level] = 0;
2091         return 0;
2092 }
2093
2094 /*
2095  * worker function to insert a single pointer in a node.
2096  * the node should have enough room for the pointer already
2097  *
2098  * slot and level indicate where you want the key to go, and
2099  * blocknr is the block the key points to.
2100  *
2101  * returns zero on success and < 0 on any error
2102  */
2103 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2104                       *root, struct btrfs_path *path, struct btrfs_disk_key
2105                       *key, u64 bytenr, int slot, int level)
2106 {
2107         struct extent_buffer *lower;
2108         int nritems;
2109
2110         BUG_ON(!path->nodes[level]);
2111         lower = path->nodes[level];
2112         nritems = btrfs_header_nritems(lower);
2113         if (slot > nritems)
2114                 BUG();
2115         if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2116                 BUG();
2117         if (slot != nritems) {
2118                 memmove_extent_buffer(lower,
2119                               btrfs_node_key_ptr_offset(slot + 1),
2120                               btrfs_node_key_ptr_offset(slot),
2121                               (nritems - slot) * sizeof(struct btrfs_key_ptr));
2122         }
2123         btrfs_set_node_key(lower, key, slot);
2124         btrfs_set_node_blockptr(lower, slot, bytenr);
2125         WARN_ON(trans->transid == 0);
2126         btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2127         btrfs_set_header_nritems(lower, nritems + 1);
2128         btrfs_mark_buffer_dirty(lower);
2129         return 0;
2130 }
2131
2132 /*
2133  * split the node at the specified level in path in two.
2134  * The path is corrected to point to the appropriate node after the split
2135  *
2136  * Before splitting this tries to make some room in the node by pushing
2137  * left and right, if either one works, it returns right away.
2138  *
2139  * returns 0 on success and < 0 on failure
2140  */
2141 static noinline int split_node(struct btrfs_trans_handle *trans,
2142                                struct btrfs_root *root,
2143                                struct btrfs_path *path, int level)
2144 {
2145         struct extent_buffer *c;
2146         struct extent_buffer *split;
2147         struct btrfs_disk_key disk_key;
2148         int mid;
2149         int ret;
2150         int wret;
2151         u32 c_nritems;
2152
2153         c = path->nodes[level];
2154         WARN_ON(btrfs_header_generation(c) != trans->transid);
2155         if (c == root->node) {
2156                 /* trying to split the root, lets make a new one */
2157                 ret = insert_new_root(trans, root, path, level + 1);
2158                 if (ret)
2159                         return ret;
2160         } else {
2161                 ret = push_nodes_for_insert(trans, root, path, level);
2162                 c = path->nodes[level];
2163                 if (!ret && btrfs_header_nritems(c) <
2164                     BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2165                         return 0;
2166                 if (ret < 0)
2167                         return ret;
2168         }
2169
2170         c_nritems = btrfs_header_nritems(c);
2171
2172         split = btrfs_alloc_free_block(trans, root, root->nodesize,
2173                                         path->nodes[level + 1]->start,
2174                                         root->root_key.objectid,
2175                                         trans->transid, level, c->start, 0);
2176         if (IS_ERR(split))
2177                 return PTR_ERR(split);
2178
2179         btrfs_set_header_flags(split, btrfs_header_flags(c));
2180         btrfs_set_header_level(split, btrfs_header_level(c));
2181         btrfs_set_header_bytenr(split, split->start);
2182         btrfs_set_header_generation(split, trans->transid);
2183         btrfs_set_header_owner(split, root->root_key.objectid);
2184         btrfs_set_header_flags(split, 0);
2185         write_extent_buffer(split, root->fs_info->fsid,
2186                             (unsigned long)btrfs_header_fsid(split),
2187                             BTRFS_FSID_SIZE);
2188         write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2189                             (unsigned long)btrfs_header_chunk_tree_uuid(split),
2190                             BTRFS_UUID_SIZE);
2191
2192         mid = (c_nritems + 1) / 2;
2193
2194         copy_extent_buffer(split, c,
2195                            btrfs_node_key_ptr_offset(0),
2196                            btrfs_node_key_ptr_offset(mid),
2197                            (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2198         btrfs_set_header_nritems(split, c_nritems - mid);
2199         btrfs_set_header_nritems(c, mid);
2200         ret = 0;
2201
2202         btrfs_mark_buffer_dirty(c);
2203         btrfs_mark_buffer_dirty(split);
2204
2205         btrfs_node_key(split, &disk_key, 0);
2206         wret = insert_ptr(trans, root, path, &disk_key, split->start,
2207                           path->slots[level + 1] + 1,
2208                           level + 1);
2209         if (wret)
2210                 ret = wret;
2211
2212         ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
2213         BUG_ON(ret);
2214
2215         if (path->slots[level] >= mid) {
2216                 path->slots[level] -= mid;
2217                 btrfs_tree_unlock(c);
2218                 free_extent_buffer(c);
2219                 path->nodes[level] = split;
2220                 path->slots[level + 1] += 1;
2221         } else {
2222                 btrfs_tree_unlock(split);
2223                 free_extent_buffer(split);
2224         }
2225         return ret;
2226 }
2227
2228 /*
2229  * how many bytes are required to store the items in a leaf.  start
2230  * and nr indicate which items in the leaf to check.  This totals up the
2231  * space used both by the item structs and the item data
2232  */
2233 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2234 {
2235         int data_len;
2236         int nritems = btrfs_header_nritems(l);
2237         int end = min(nritems, start + nr) - 1;
2238
2239         if (!nr)
2240                 return 0;
2241         data_len = btrfs_item_end_nr(l, start);
2242         data_len = data_len - btrfs_item_offset_nr(l, end);
2243         data_len += sizeof(struct btrfs_item) * nr;
2244         WARN_ON(data_len < 0);
2245         return data_len;
2246 }
2247
2248 /*
2249  * The space between the end of the leaf items and
2250  * the start of the leaf data.  IOW, how much room
2251  * the leaf has left for both items and data
2252  */
2253 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2254                                    struct extent_buffer *leaf)
2255 {
2256         int nritems = btrfs_header_nritems(leaf);
2257         int ret;
2258         ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2259         if (ret < 0) {
2260                 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2261                        "used %d nritems %d\n",
2262                        ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2263                        leaf_space_used(leaf, 0, nritems), nritems);
2264         }
2265         return ret;
2266 }
2267
2268 /*
2269  * push some data in the path leaf to the right, trying to free up at
2270  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2271  *
2272  * returns 1 if the push failed because the other node didn't have enough
2273  * room, 0 if everything worked out and < 0 if there were major errors.
2274  */
2275 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2276                            *root, struct btrfs_path *path, int data_size,
2277                            int empty)
2278 {
2279         struct extent_buffer *left = path->nodes[0];
2280         struct extent_buffer *right;
2281         struct extent_buffer *upper;
2282         struct btrfs_disk_key disk_key;
2283         int slot;
2284         u32 i;
2285         int free_space;
2286         int push_space = 0;
2287         int push_items = 0;
2288         struct btrfs_item *item;
2289         u32 left_nritems;
2290         u32 nr;
2291         u32 right_nritems;
2292         u32 data_end;
2293         u32 this_item_size;
2294         int ret;
2295
2296         slot = path->slots[1];
2297         if (!path->nodes[1])
2298                 return 1;
2299
2300         upper = path->nodes[1];
2301         if (slot >= btrfs_header_nritems(upper) - 1)
2302                 return 1;
2303
2304         btrfs_assert_tree_locked(path->nodes[1]);
2305
2306         right = read_node_slot(root, upper, slot + 1);
2307         btrfs_tree_lock(right);
2308         btrfs_set_lock_blocking(right);
2309
2310         free_space = btrfs_leaf_free_space(root, right);
2311         if (free_space < data_size)
2312                 goto out_unlock;
2313
2314         /* cow and double check */
2315         ret = btrfs_cow_block(trans, root, right, upper,
2316                               slot + 1, &right);
2317         if (ret)
2318                 goto out_unlock;
2319
2320         free_space = btrfs_leaf_free_space(root, right);
2321         if (free_space < data_size)
2322                 goto out_unlock;
2323
2324         left_nritems = btrfs_header_nritems(left);
2325         if (left_nritems == 0)
2326                 goto out_unlock;
2327
2328         if (empty)
2329                 nr = 0;
2330         else
2331                 nr = 1;
2332
2333         if (path->slots[0] >= left_nritems)
2334                 push_space += data_size;
2335
2336         i = left_nritems - 1;
2337         while (i >= nr) {
2338                 item = btrfs_item_nr(left, i);
2339
2340                 if (!empty && push_items > 0) {
2341                         if (path->slots[0] > i)
2342                                 break;
2343                         if (path->slots[0] == i) {
2344                                 int space = btrfs_leaf_free_space(root, left);
2345                                 if (space + push_space * 2 > free_space)
2346                                         break;
2347                         }
2348                 }
2349
2350                 if (path->slots[0] == i)
2351                         push_space += data_size;
2352
2353                 if (!left->map_token) {
2354                         map_extent_buffer(left, (unsigned long)item,
2355                                         sizeof(struct btrfs_item),
2356                                         &left->map_token, &left->kaddr,
2357                                         &left->map_start, &left->map_len,
2358                                         KM_USER1);
2359                 }
2360
2361                 this_item_size = btrfs_item_size(left, item);
2362                 if (this_item_size + sizeof(*item) + push_space > free_space)
2363                         break;
2364
2365                 push_items++;
2366                 push_space += this_item_size + sizeof(*item);
2367                 if (i == 0)
2368                         break;
2369                 i--;
2370         }
2371         if (left->map_token) {
2372                 unmap_extent_buffer(left, left->map_token, KM_USER1);
2373                 left->map_token = NULL;
2374         }
2375
2376         if (push_items == 0)
2377                 goto out_unlock;
2378
2379         if (!empty && push_items == left_nritems)
2380                 WARN_ON(1);
2381
2382         /* push left to right */
2383         right_nritems = btrfs_header_nritems(right);
2384
2385         push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2386         push_space -= leaf_data_end(root, left);
2387
2388         /* make room in the right data area */
2389         data_end = leaf_data_end(root, right);
2390         memmove_extent_buffer(right,
2391                               btrfs_leaf_data(right) + data_end - push_space,
2392                               btrfs_leaf_data(right) + data_end,
2393                               BTRFS_LEAF_DATA_SIZE(root) - data_end);
2394
2395         /* copy from the left data area */
2396         copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2397                      BTRFS_LEAF_DATA_SIZE(root) - push_space,
2398                      btrfs_leaf_data(left) + leaf_data_end(root, left),
2399                      push_space);
2400
2401         memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2402                               btrfs_item_nr_offset(0),
2403                               right_nritems * sizeof(struct btrfs_item));
2404
2405         /* copy the items from left to right */
2406         copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2407                    btrfs_item_nr_offset(left_nritems - push_items),
2408                    push_items * sizeof(struct btrfs_item));
2409
2410         /* update the item pointers */
2411         right_nritems += push_items;
2412         btrfs_set_header_nritems(right, right_nritems);
2413         push_space = BTRFS_LEAF_DATA_SIZE(root);
2414         for (i = 0; i < right_nritems; i++) {
2415                 item = btrfs_item_nr(right, i);
2416                 if (!right->map_token) {
2417                         map_extent_buffer(right, (unsigned long)item,
2418                                         sizeof(struct btrfs_item),
2419                                         &right->map_token, &right->kaddr,
2420                                         &right->map_start, &right->map_len,
2421                                         KM_USER1);
2422                 }
2423                 push_space -= btrfs_item_size(right, item);
2424                 btrfs_set_item_offset(right, item, push_space);
2425         }
2426
2427         if (right->map_token) {
2428                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2429                 right->map_token = NULL;
2430         }
2431         left_nritems -= push_items;
2432         btrfs_set_header_nritems(left, left_nritems);
2433
2434         if (left_nritems)
2435                 btrfs_mark_buffer_dirty(left);
2436         btrfs_mark_buffer_dirty(right);
2437
2438         ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
2439         BUG_ON(ret);
2440
2441         btrfs_item_key(right, &disk_key, 0);
2442         btrfs_set_node_key(upper, &disk_key, slot + 1);
2443         btrfs_mark_buffer_dirty(upper);
2444
2445         /* then fixup the leaf pointer in the path */
2446         if (path->slots[0] >= left_nritems) {
2447                 path->slots[0] -= left_nritems;
2448                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2449                         clean_tree_block(trans, root, path->nodes[0]);
2450                 btrfs_tree_unlock(path->nodes[0]);
2451                 free_extent_buffer(path->nodes[0]);
2452                 path->nodes[0] = right;
2453                 path->slots[1] += 1;
2454         } else {
2455                 btrfs_tree_unlock(right);
2456                 free_extent_buffer(right);
2457         }
2458         return 0;
2459
2460 out_unlock:
2461         btrfs_tree_unlock(right);
2462         free_extent_buffer(right);
2463         return 1;
2464 }
2465
2466 /*
2467  * push some data in the path leaf to the left, trying to free up at
2468  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2469  */
2470 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2471                           *root, struct btrfs_path *path, int data_size,
2472                           int empty)
2473 {
2474         struct btrfs_disk_key disk_key;
2475         struct extent_buffer *right = path->nodes[0];
2476         struct extent_buffer *left;
2477         int slot;
2478         int i;
2479         int free_space;
2480         int push_space = 0;
2481         int push_items = 0;
2482         struct btrfs_item *item;
2483         u32 old_left_nritems;
2484         u32 right_nritems;
2485         u32 nr;
2486         int ret = 0;
2487         int wret;
2488         u32 this_item_size;
2489         u32 old_left_item_size;
2490
2491         slot = path->slots[1];
2492         if (slot == 0)
2493                 return 1;
2494         if (!path->nodes[1])
2495                 return 1;
2496
2497         right_nritems = btrfs_header_nritems(right);
2498         if (right_nritems == 0)
2499                 return 1;
2500
2501         btrfs_assert_tree_locked(path->nodes[1]);
2502
2503         left = read_node_slot(root, path->nodes[1], slot - 1);
2504         btrfs_tree_lock(left);
2505         btrfs_set_lock_blocking(left);
2506
2507         free_space = btrfs_leaf_free_space(root, left);
2508         if (free_space < data_size) {
2509                 ret = 1;
2510                 goto out;
2511         }
2512
2513         /* cow and double check */
2514         ret = btrfs_cow_block(trans, root, left,
2515                               path->nodes[1], slot - 1, &left);
2516         if (ret) {
2517                 /* we hit -ENOSPC, but it isn't fatal here */
2518                 ret = 1;
2519                 goto out;
2520         }
2521
2522         free_space = btrfs_leaf_free_space(root, left);
2523         if (free_space < data_size) {
2524                 ret = 1;
2525                 goto out;
2526         }
2527
2528         if (empty)
2529                 nr = right_nritems;
2530         else
2531                 nr = right_nritems - 1;
2532
2533         for (i = 0; i < nr; i++) {
2534                 item = btrfs_item_nr(right, i);
2535                 if (!right->map_token) {
2536                         map_extent_buffer(right, (unsigned long)item,
2537                                         sizeof(struct btrfs_item),
2538                                         &right->map_token, &right->kaddr,
2539                                         &right->map_start, &right->map_len,
2540                                         KM_USER1);
2541                 }
2542
2543                 if (!empty && push_items > 0) {
2544                         if (path->slots[0] < i)
2545                                 break;
2546                         if (path->slots[0] == i) {
2547                                 int space = btrfs_leaf_free_space(root, right);
2548                                 if (space + push_space * 2 > free_space)
2549                                         break;
2550                         }
2551                 }
2552
2553                 if (path->slots[0] == i)
2554                         push_space += data_size;
2555
2556                 this_item_size = btrfs_item_size(right, item);
2557                 if (this_item_size + sizeof(*item) + push_space > free_space)
2558                         break;
2559
2560                 push_items++;
2561                 push_space += this_item_size + sizeof(*item);
2562         }
2563
2564         if (right->map_token) {
2565                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2566                 right->map_token = NULL;
2567         }
2568
2569         if (push_items == 0) {
2570                 ret = 1;
2571                 goto out;
2572         }
2573         if (!empty && push_items == btrfs_header_nritems(right))
2574                 WARN_ON(1);
2575
2576         /* push data from right to left */
2577         copy_extent_buffer(left, right,
2578                            btrfs_item_nr_offset(btrfs_header_nritems(left)),
2579                            btrfs_item_nr_offset(0),
2580                            push_items * sizeof(struct btrfs_item));
2581
2582         push_space = BTRFS_LEAF_DATA_SIZE(root) -
2583                      btrfs_item_offset_nr(right, push_items - 1);
2584
2585         copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2586                      leaf_data_end(root, left) - push_space,
2587                      btrfs_leaf_data(right) +
2588                      btrfs_item_offset_nr(right, push_items - 1),
2589                      push_space);
2590         old_left_nritems = btrfs_header_nritems(left);
2591         BUG_ON(old_left_nritems <= 0);
2592
2593         old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2594         for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2595                 u32 ioff;
2596
2597                 item = btrfs_item_nr(left, i);
2598                 if (!left->map_token) {
2599                         map_extent_buffer(left, (unsigned long)item,
2600                                         sizeof(struct btrfs_item),
2601                                         &left->map_token, &left->kaddr,
2602                                         &left->map_start, &left->map_len,
2603                                         KM_USER1);
2604                 }
2605
2606                 ioff = btrfs_item_offset(left, item);
2607                 btrfs_set_item_offset(left, item,
2608                       ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2609         }
2610         btrfs_set_header_nritems(left, old_left_nritems + push_items);
2611         if (left->map_token) {
2612                 unmap_extent_buffer(left, left->map_token, KM_USER1);
2613                 left->map_token = NULL;
2614         }
2615
2616         /* fixup right node */
2617         if (push_items > right_nritems) {
2618                 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2619                        right_nritems);
2620                 WARN_ON(1);
2621         }
2622
2623         if (push_items < right_nritems) {
2624                 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2625                                                   leaf_data_end(root, right);
2626                 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2627                                       BTRFS_LEAF_DATA_SIZE(root) - push_space,
2628                                       btrfs_leaf_data(right) +
2629                                       leaf_data_end(root, right), push_space);
2630
2631                 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2632                               btrfs_item_nr_offset(push_items),
2633                              (btrfs_header_nritems(right) - push_items) *
2634                              sizeof(struct btrfs_item));
2635         }
2636         right_nritems -= push_items;
2637         btrfs_set_header_nritems(right, right_nritems);
2638         push_space = BTRFS_LEAF_DATA_SIZE(root);
2639         for (i = 0; i < right_nritems; i++) {
2640                 item = btrfs_item_nr(right, i);
2641
2642                 if (!right->map_token) {
2643                         map_extent_buffer(right, (unsigned long)item,
2644                                         sizeof(struct btrfs_item),
2645                                         &right->map_token, &right->kaddr,
2646                                         &right->map_start, &right->map_len,
2647                                         KM_USER1);
2648                 }
2649
2650                 push_space = push_space - btrfs_item_size(right, item);
2651                 btrfs_set_item_offset(right, item, push_space);
2652         }
2653         if (right->map_token) {
2654                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2655                 right->map_token = NULL;
2656         }
2657
2658         btrfs_mark_buffer_dirty(left);
2659         if (right_nritems)
2660                 btrfs_mark_buffer_dirty(right);
2661
2662         ret = btrfs_update_ref(trans, root, right, left,
2663                                old_left_nritems, push_items);
2664         BUG_ON(ret);
2665
2666         btrfs_item_key(right, &disk_key, 0);
2667         wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2668         if (wret)
2669                 ret = wret;
2670
2671         /* then fixup the leaf pointer in the path */
2672         if (path->slots[0] < push_items) {
2673                 path->slots[0] += old_left_nritems;
2674                 if (btrfs_header_nritems(path->nodes[0]) == 0)
2675                         clean_tree_block(trans, root, path->nodes[0]);
2676                 btrfs_tree_unlock(path->nodes[0]);
2677                 free_extent_buffer(path->nodes[0]);
2678                 path->nodes[0] = left;
2679                 path->slots[1] -= 1;
2680         } else {
2681                 btrfs_tree_unlock(left);
2682                 free_extent_buffer(left);
2683                 path->slots[0] -= push_items;
2684         }
2685         BUG_ON(path->slots[0] < 0);
2686         return ret;
2687 out:
2688         btrfs_tree_unlock(left);
2689         free_extent_buffer(left);
2690         return ret;
2691 }
2692
2693 /*
2694  * split the path's leaf in two, making sure there is at least data_size
2695  * available for the resulting leaf level of the path.
2696  *
2697  * returns 0 if all went well and < 0 on failure.
2698  */
2699 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2700                                struct btrfs_root *root,
2701                                struct btrfs_key *ins_key,
2702                                struct btrfs_path *path, int data_size,
2703                                int extend)
2704 {
2705         struct extent_buffer *l;
2706         u32 nritems;
2707         int mid;
2708         int slot;
2709         struct extent_buffer *right;
2710         int data_copy_size;
2711         int rt_data_off;
2712         int i;
2713         int ret = 0;
2714         int wret;
2715         int double_split;
2716         int num_doubles = 0;
2717         struct btrfs_disk_key disk_key;
2718
2719         /* first try to make some room by pushing left and right */
2720         if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2721                 wret = push_leaf_right(trans, root, path, data_size, 0);
2722                 if (wret < 0)
2723                         return wret;
2724                 if (wret) {
2725                         wret = push_leaf_left(trans, root, path, data_size, 0);
2726                         if (wret < 0)
2727                                 return wret;
2728                 }
2729                 l = path->nodes[0];
2730
2731                 /* did the pushes work? */
2732                 if (btrfs_leaf_free_space(root, l) >= data_size)
2733                         return 0;
2734         }
2735
2736         if (!path->nodes[1]) {
2737                 ret = insert_new_root(trans, root, path, 1);
2738                 if (ret)
2739                         return ret;
2740         }
2741 again:
2742         double_split = 0;
2743         l = path->nodes[0];
2744         slot = path->slots[0];
2745         nritems = btrfs_header_nritems(l);
2746         mid = (nritems + 1) / 2;
2747
2748         right = btrfs_alloc_free_block(trans, root, root->leafsize,
2749                                         path->nodes[1]->start,
2750                                         root->root_key.objectid,
2751                                         trans->transid, 0, l->start, 0);
2752         if (IS_ERR(right)) {
2753                 BUG_ON(1);
2754                 return PTR_ERR(right);
2755         }
2756
2757         memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2758         btrfs_set_header_bytenr(right, right->start);
2759         btrfs_set_header_generation(right, trans->transid);
2760         btrfs_set_header_owner(right, root->root_key.objectid);
2761         btrfs_set_header_level(right, 0);
2762         write_extent_buffer(right, root->fs_info->fsid,
2763                             (unsigned long)btrfs_header_fsid(right),
2764                             BTRFS_FSID_SIZE);
2765
2766         write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2767                             (unsigned long)btrfs_header_chunk_tree_uuid(right),
2768                             BTRFS_UUID_SIZE);
2769         if (mid <= slot) {
2770                 if (nritems == 1 ||
2771                     leaf_space_used(l, mid, nritems - mid) + data_size >
2772                         BTRFS_LEAF_DATA_SIZE(root)) {
2773                         if (slot >= nritems) {
2774                                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2775                                 btrfs_set_header_nritems(right, 0);
2776                                 wret = insert_ptr(trans, root, path,
2777                                                   &disk_key, right->start,
2778                                                   path->slots[1] + 1, 1);
2779                                 if (wret)
2780                                         ret = wret;
2781
2782                                 btrfs_tree_unlock(path->nodes[0]);
2783                                 free_extent_buffer(path->nodes[0]);
2784                                 path->nodes[0] = right;
2785                                 path->slots[0] = 0;
2786                                 path->slots[1] += 1;
2787                                 btrfs_mark_buffer_dirty(right);
2788                                 return ret;
2789                         }
2790                         mid = slot;
2791                         if (mid != nritems &&
2792                             leaf_space_used(l, mid, nritems - mid) +
2793                             data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2794                                 double_split = 1;
2795                         }
2796                 }
2797         } else {
2798                 if (leaf_space_used(l, 0, mid) + data_size >
2799                         BTRFS_LEAF_DATA_SIZE(root)) {
2800                         if (!extend && data_size && slot == 0) {
2801                                 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2802                                 btrfs_set_header_nritems(right, 0);
2803                                 wret = insert_ptr(trans, root, path,
2804                                                   &disk_key,
2805                                                   right->start,
2806                                                   path->slots[1], 1);
2807                                 if (wret)
2808                                         ret = wret;
2809                                 btrfs_tree_unlock(path->nodes[0]);
2810                                 free_extent_buffer(path->nodes[0]);
2811                                 path->nodes[0] = right;
2812                                 path->slots[0] = 0;
2813                                 if (path->slots[1] == 0) {
2814                                         wret = fixup_low_keys(trans, root,
2815                                                       path, &disk_key, 1);
2816                                         if (wret)
2817                                                 ret = wret;
2818                                 }
2819                                 btrfs_mark_buffer_dirty(right);
2820                                 return ret;
2821                         } else if ((extend || !data_size) && slot == 0) {
2822                                 mid = 1;
2823                         } else {
2824                                 mid = slot;
2825                                 if (mid != nritems &&
2826                                     leaf_space_used(l, mid, nritems - mid) +
2827                                     data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2828                                         double_split = 1;
2829                                 }
2830                         }
2831                 }
2832         }
2833         nritems = nritems - mid;
2834         btrfs_set_header_nritems(right, nritems);
2835         data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2836
2837         copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2838                            btrfs_item_nr_offset(mid),
2839                            nritems * sizeof(struct btrfs_item));
2840
2841         copy_extent_buffer(right, l,
2842                      btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2843                      data_copy_size, btrfs_leaf_data(l) +
2844                      leaf_data_end(root, l), data_copy_size);
2845
2846         rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2847                       btrfs_item_end_nr(l, mid);
2848
2849         for (i = 0; i < nritems; i++) {
2850                 struct btrfs_item *item = btrfs_item_nr(right, i);
2851                 u32 ioff;
2852
2853                 if (!right->map_token) {
2854                         map_extent_buffer(right, (unsigned long)item,
2855                                         sizeof(struct btrfs_item),
2856                                         &right->map_token, &right->kaddr,
2857                                         &right->map_start, &right->map_len,
2858                                         KM_USER1);
2859                 }
2860
2861                 ioff = btrfs_item_offset(right, item);
2862                 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2863         }
2864
2865         if (right->map_token) {
2866                 unmap_extent_buffer(right, right->map_token, KM_USER1);
2867                 right->map_token = NULL;
2868         }
2869
2870         btrfs_set_header_nritems(l, mid);
2871         ret = 0;
2872         btrfs_item_key(right, &disk_key, 0);
2873         wret = insert_ptr(trans, root, path, &disk_key, right->start,
2874                           path->slots[1] + 1, 1);
2875         if (wret)
2876                 ret = wret;
2877
2878         btrfs_mark_buffer_dirty(right);
2879         btrfs_mark_buffer_dirty(l);
2880         BUG_ON(path->slots[0] != slot);
2881
2882         ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
2883         BUG_ON(ret);
2884
2885         if (mid <= slot) {
2886                 btrfs_tree_unlock(path->nodes[0]);
2887                 free_extent_buffer(path->nodes[0]);
2888                 path->nodes[0] = right;
2889                 path->slots[0] -= mid;
2890                 path->slots[1] += 1;
2891         } else {
2892                 btrfs_tree_unlock(right);
2893                 free_extent_buffer(right);
2894         }
2895
2896         BUG_ON(path->slots[0] < 0);
2897
2898         if (double_split) {
2899                 BUG_ON(num_doubles != 0);
2900                 num_doubles++;
2901                 goto again;
2902         }
2903         return ret;
2904 }
2905
2906 /*
2907  * This function splits a single item into two items,
2908  * giving 'new_key' to the new item and splitting the
2909  * old one at split_offset (from the start of the item).
2910  *
2911  * The path may be released by this operation.  After
2912  * the split, the path is pointing to the old item.  The
2913  * new item is going to be in the same node as the old one.
2914  *
2915  * Note, the item being split must be smaller enough to live alone on
2916  * a tree block with room for one extra struct btrfs_item
2917  *
2918  * This allows us to split the item in place, keeping a lock on the
2919  * leaf the entire time.
2920  */
2921 int btrfs_split_item(struct btrfs_trans_handle *trans,
2922                      struct btrfs_root *root,
2923                      struct btrfs_path *path,
2924                      struct btrfs_key *new_key,
2925                      unsigned long split_offset)
2926 {
2927         u32 item_size;
2928         struct extent_buffer *leaf;
2929         struct btrfs_key orig_key;
2930         struct btrfs_item *item;
2931         struct btrfs_item *new_item;
2932         int ret = 0;
2933         int slot;
2934         u32 nritems;
2935         u32 orig_offset;
2936         struct btrfs_disk_key disk_key;
2937         char *buf;
2938
2939         leaf = path->nodes[0];
2940         btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
2941         if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
2942                 goto split;
2943
2944         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2945         btrfs_release_path(root, path);
2946
2947         path->search_for_split = 1;
2948         path->keep_locks = 1;
2949
2950         ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
2951         path->search_for_split = 0;
2952
2953         /* if our item isn't there or got smaller, return now */
2954         if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
2955                                                         path->slots[0])) {
2956                 path->keep_locks = 0;
2957                 return -EAGAIN;
2958         }
2959
2960         ret = split_leaf(trans, root, &orig_key, path,
2961                          sizeof(struct btrfs_item), 1);
2962         path->keep_locks = 0;
2963         BUG_ON(ret);
2964
2965         /*
2966          * make sure any changes to the path from split_leaf leave it
2967          * in a blocking state
2968          */
2969         btrfs_set_path_blocking(path);
2970
2971         leaf = path->nodes[0];
2972         BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
2973
2974 split:
2975         item = btrfs_item_nr(leaf, path->slots[0]);
2976         orig_offset = btrfs_item_offset(leaf, item);
2977         item_size = btrfs_item_size(leaf, item);
2978
2979
2980         buf = kmalloc(item_size, GFP_NOFS);
2981         read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
2982                             path->slots[0]), item_size);
2983         slot = path->slots[0] + 1;
2984         leaf = path->nodes[0];
2985
2986         nritems = btrfs_header_nritems(leaf);
2987
2988         if (slot != nritems) {
2989                 /* shift the items */
2990                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
2991                               btrfs_item_nr_offset(slot),
2992                               (nritems - slot) * sizeof(struct btrfs_item));
2993
2994         }
2995
2996         btrfs_cpu_key_to_disk(&disk_key, new_key);
2997         btrfs_set_item_key(leaf, &disk_key, slot);
2998
2999         new_item = btrfs_item_nr(leaf, slot);
3000
3001         btrfs_set_item_offset(leaf, new_item, orig_offset);
3002         btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3003
3004         btrfs_set_item_offset(leaf, item,
3005                               orig_offset + item_size - split_offset);
3006         btrfs_set_item_size(leaf, item, split_offset);
3007
3008         btrfs_set_header_nritems(leaf, nritems + 1);
3009
3010         /* write the data for the start of the original item */
3011         write_extent_buffer(leaf, buf,
3012                             btrfs_item_ptr_offset(leaf, path->slots[0]),
3013                             split_offset);
3014
3015         /* write the data for the new item */
3016         write_extent_buffer(leaf, buf + split_offset,
3017                             btrfs_item_ptr_offset(leaf, slot),
3018                             item_size - split_offset);
3019         btrfs_mark_buffer_dirty(leaf);
3020
3021         ret = 0;
3022         if (btrfs_leaf_free_space(root, leaf) < 0) {
3023                 btrfs_print_leaf(root, leaf);
3024                 BUG();
3025         }
3026         kfree(buf);
3027         return ret;
3028 }
3029
3030 /*
3031  * make the item pointed to by the path smaller.  new_size indicates
3032  * how small to make it, and from_end tells us if we just chop bytes
3033  * off the end of the item or if we shift the item to chop bytes off
3034  * the front.
3035  */
3036 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3037                         struct btrfs_root *root,
3038                         struct btrfs_path *path,
3039                         u32 new_size, int from_end)
3040 {
3041         int ret = 0;
3042         int slot;
3043         int slot_orig;
3044         struct extent_buffer *leaf;
3045         struct btrfs_item *item;
3046         u32 nritems;
3047         unsigned int data_end;
3048         unsigned int old_data_start;
3049         unsigned int old_size;
3050         unsigned int size_diff;
3051         int i;
3052
3053         slot_orig = path->slots[0];
3054         leaf = path->nodes[0];
3055         slot = path->slots[0];
3056
3057         old_size = btrfs_item_size_nr(leaf, slot);
3058         if (old_size == new_size)
3059                 return 0;
3060
3061         nritems = btrfs_header_nritems(leaf);
3062         data_end = leaf_data_end(root, leaf);
3063
3064         old_data_start = btrfs_item_offset_nr(leaf, slot);
3065
3066         size_diff = old_size - new_size;
3067
3068         BUG_ON(slot < 0);
3069         BUG_ON(slot >= nritems);
3070
3071         /*
3072          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3073          */
3074         /* first correct the data pointers */
3075         for (i = slot; i < nritems; i++) {
3076                 u32 ioff;
3077                 item = btrfs_item_nr(leaf, i);
3078
3079                 if (!leaf->map_token) {
3080                         map_extent_buffer(leaf, (unsigned long)item,
3081                                         sizeof(struct btrfs_item),
3082                                         &leaf->map_token, &leaf->kaddr,
3083                                         &leaf->map_start, &leaf->map_len,
3084                                         KM_USER1);
3085                 }
3086
3087                 ioff = btrfs_item_offset(leaf, item);
3088                 btrfs_set_item_offset(leaf, item, ioff + size_diff);
3089         }
3090
3091         if (leaf->map_token) {
3092                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3093                 leaf->map_token = NULL;
3094         }
3095
3096         /* shift the data */
3097         if (from_end) {
3098                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3099                               data_end + size_diff, btrfs_leaf_data(leaf) +
3100                               data_end, old_data_start + new_size - data_end);
3101         } else {
3102                 struct btrfs_disk_key disk_key;
3103                 u64 offset;
3104
3105                 btrfs_item_key(leaf, &disk_key, slot);
3106
3107                 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3108                         unsigned long ptr;
3109                         struct btrfs_file_extent_item *fi;
3110
3111                         fi = btrfs_item_ptr(leaf, slot,
3112                                             struct btrfs_file_extent_item);
3113                         fi = (struct btrfs_file_extent_item *)(
3114                              (unsigned long)fi - size_diff);
3115
3116                         if (btrfs_file_extent_type(leaf, fi) ==
3117                             BTRFS_FILE_EXTENT_INLINE) {
3118                                 ptr = btrfs_item_ptr_offset(leaf, slot);
3119                                 memmove_extent_buffer(leaf, ptr,
3120                                       (unsigned long)fi,
3121                                       offsetof(struct btrfs_file_extent_item,
3122                                                  disk_bytenr));
3123                         }
3124                 }
3125
3126                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3127                               data_end + size_diff, btrfs_leaf_data(leaf) +
3128                               data_end, old_data_start - data_end);
3129
3130                 offset = btrfs_disk_key_offset(&disk_key);
3131                 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3132                 btrfs_set_item_key(leaf, &disk_key, slot);
3133                 if (slot == 0)
3134                         fixup_low_keys(trans, root, path, &disk_key, 1);
3135         }
3136
3137         item = btrfs_item_nr(leaf, slot);
3138         btrfs_set_item_size(leaf, item, new_size);
3139         btrfs_mark_buffer_dirty(leaf);
3140
3141         ret = 0;
3142         if (btrfs_leaf_free_space(root, leaf) < 0) {
3143                 btrfs_print_leaf(root, leaf);
3144                 BUG();
3145         }
3146         return ret;
3147 }
3148
3149 /*
3150  * make the item pointed to by the path bigger, data_size is the new size.
3151  */
3152 int btrfs_extend_item(struct btrfs_trans_handle *trans,
3153                       struct btrfs_root *root, struct btrfs_path *path,
3154                       u32 data_size)
3155 {
3156         int ret = 0;
3157         int slot;
3158         int slot_orig;
3159         struct extent_buffer *leaf;
3160         struct btrfs_item *item;
3161         u32 nritems;
3162         unsigned int data_end;
3163         unsigned int old_data;
3164         unsigned int old_size;
3165         int i;
3166
3167         slot_orig = path->slots[0];
3168         leaf = path->nodes[0];
3169
3170         nritems = btrfs_header_nritems(leaf);
3171         data_end = leaf_data_end(root, leaf);
3172
3173         if (btrfs_leaf_free_space(root, leaf) < data_size) {
3174                 btrfs_print_leaf(root, leaf);
3175                 BUG();
3176         }
3177         slot = path->slots[0];
3178         old_data = btrfs_item_end_nr(leaf, slot);
3179
3180         BUG_ON(slot < 0);
3181         if (slot >= nritems) {
3182                 btrfs_print_leaf(root, leaf);
3183                 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3184                        slot, nritems);
3185                 BUG_ON(1);
3186         }
3187
3188         /*
3189          * item0..itemN ... dataN.offset..dataN.size .. data0.size
3190          */
3191         /* first correct the data pointers */
3192         for (i = slot; i < nritems; i++) {
3193                 u32 ioff;
3194                 item = btrfs_item_nr(leaf, i);
3195
3196                 if (!leaf->map_token) {
3197                         map_extent_buffer(leaf, (unsigned long)item,
3198                                         sizeof(struct btrfs_item),
3199                                         &leaf->map_token, &leaf->kaddr,
3200                                         &leaf->map_start, &leaf->map_len,
3201                                         KM_USER1);
3202                 }
3203                 ioff = btrfs_item_offset(leaf, item);
3204                 btrfs_set_item_offset(leaf, item, ioff - data_size);
3205         }
3206
3207         if (leaf->map_token) {
3208                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3209                 leaf->map_token = NULL;
3210         }
3211
3212         /* shift the data */
3213         memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3214                       data_end - data_size, btrfs_leaf_data(leaf) +
3215                       data_end, old_data - data_end);
3216
3217         data_end = old_data;
3218         old_size = btrfs_item_size_nr(leaf, slot);
3219         item = btrfs_item_nr(leaf, slot);
3220         btrfs_set_item_size(leaf, item, old_size + data_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  * Given a key and some data, insert items into the tree.
3233  * This does all the path init required, making room in the tree if needed.
3234  * Returns the number of keys that were inserted.
3235  */
3236 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3237                             struct btrfs_root *root,
3238                             struct btrfs_path *path,
3239                             struct btrfs_key *cpu_key, u32 *data_size,
3240                             int nr)
3241 {
3242         struct extent_buffer *leaf;
3243         struct btrfs_item *item;
3244         int ret = 0;
3245         int slot;
3246         int i;
3247         u32 nritems;
3248         u32 total_data = 0;
3249         u32 total_size = 0;
3250         unsigned int data_end;
3251         struct btrfs_disk_key disk_key;
3252         struct btrfs_key found_key;
3253
3254         for (i = 0; i < nr; i++) {
3255                 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3256                     BTRFS_LEAF_DATA_SIZE(root)) {
3257                         break;
3258                         nr = i;
3259                 }
3260                 total_data += data_size[i];
3261                 total_size += data_size[i] + sizeof(struct btrfs_item);
3262         }
3263         BUG_ON(nr == 0);
3264
3265         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3266         if (ret == 0)
3267                 return -EEXIST;
3268         if (ret < 0)
3269                 goto out;
3270
3271         leaf = path->nodes[0];
3272
3273         nritems = btrfs_header_nritems(leaf);
3274         data_end = leaf_data_end(root, leaf);
3275
3276         if (btrfs_leaf_free_space(root, leaf) < total_size) {
3277                 for (i = nr; i >= 0; i--) {
3278                         total_data -= data_size[i];
3279                         total_size -= data_size[i] + sizeof(struct btrfs_item);
3280                         if (total_size < btrfs_leaf_free_space(root, leaf))
3281                                 break;
3282                 }
3283                 nr = i;
3284         }
3285
3286         slot = path->slots[0];
3287         BUG_ON(slot < 0);
3288
3289         if (slot != nritems) {
3290                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3291
3292                 item = btrfs_item_nr(leaf, slot);
3293                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3294
3295                 /* figure out how many keys we can insert in here */
3296                 total_data = data_size[0];
3297                 for (i = 1; i < nr; i++) {
3298                         if (comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3299                                 break;
3300                         total_data += data_size[i];
3301                 }
3302                 nr = i;
3303
3304                 if (old_data < data_end) {
3305                         btrfs_print_leaf(root, leaf);
3306                         printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3307                                slot, old_data, data_end);
3308                         BUG_ON(1);
3309                 }
3310                 /*
3311                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3312                  */
3313                 /* first correct the data pointers */
3314                 WARN_ON(leaf->map_token);
3315                 for (i = slot; i < nritems; i++) {
3316                         u32 ioff;
3317
3318                         item = btrfs_item_nr(leaf, i);
3319                         if (!leaf->map_token) {
3320                                 map_extent_buffer(leaf, (unsigned long)item,
3321                                         sizeof(struct btrfs_item),
3322                                         &leaf->map_token, &leaf->kaddr,
3323                                         &leaf->map_start, &leaf->map_len,
3324                                         KM_USER1);
3325                         }
3326
3327                         ioff = btrfs_item_offset(leaf, item);
3328                         btrfs_set_item_offset(leaf, item, ioff - total_data);
3329                 }
3330                 if (leaf->map_token) {
3331                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3332                         leaf->map_token = NULL;
3333                 }
3334
3335                 /* shift the items */
3336                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3337                               btrfs_item_nr_offset(slot),
3338                               (nritems - slot) * sizeof(struct btrfs_item));
3339
3340                 /* shift the data */
3341                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3342                               data_end - total_data, btrfs_leaf_data(leaf) +
3343                               data_end, old_data - data_end);
3344                 data_end = old_data;
3345         } else {
3346                 /*
3347                  * this sucks but it has to be done, if we are inserting at
3348                  * the end of the leaf only insert 1 of the items, since we
3349                  * have no way of knowing whats on the next leaf and we'd have
3350                  * to drop our current locks to figure it out
3351                  */
3352                 nr = 1;
3353         }
3354
3355         /* setup the item for the new data */
3356         for (i = 0; i < nr; i++) {
3357                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3358                 btrfs_set_item_key(leaf, &disk_key, slot + i);
3359                 item = btrfs_item_nr(leaf, slot + i);
3360                 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3361                 data_end -= data_size[i];
3362                 btrfs_set_item_size(leaf, item, data_size[i]);
3363         }
3364         btrfs_set_header_nritems(leaf, nritems + nr);
3365         btrfs_mark_buffer_dirty(leaf);
3366
3367         ret = 0;
3368         if (slot == 0) {
3369                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3370                 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3371         }
3372
3373         if (btrfs_leaf_free_space(root, leaf) < 0) {
3374                 btrfs_print_leaf(root, leaf);
3375                 BUG();
3376         }
3377 out:
3378         if (!ret)
3379                 ret = nr;
3380         return ret;
3381 }
3382
3383 /*
3384  * Given a key and some data, insert items into the tree.
3385  * This does all the path init required, making room in the tree if needed.
3386  */
3387 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3388                             struct btrfs_root *root,
3389                             struct btrfs_path *path,
3390                             struct btrfs_key *cpu_key, u32 *data_size,
3391                             int nr)
3392 {
3393         struct extent_buffer *leaf;
3394         struct btrfs_item *item;
3395         int ret = 0;
3396         int slot;
3397         int slot_orig;
3398         int i;
3399         u32 nritems;
3400         u32 total_size = 0;
3401         u32 total_data = 0;
3402         unsigned int data_end;
3403         struct btrfs_disk_key disk_key;
3404
3405         for (i = 0; i < nr; i++)
3406                 total_data += data_size[i];
3407
3408         total_size = total_data + (nr * sizeof(struct btrfs_item));
3409         ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3410         if (ret == 0)
3411                 return -EEXIST;
3412         if (ret < 0)
3413                 goto out;
3414
3415         slot_orig = path->slots[0];
3416         leaf = path->nodes[0];
3417
3418         nritems = btrfs_header_nritems(leaf);
3419         data_end = leaf_data_end(root, leaf);
3420
3421         if (btrfs_leaf_free_space(root, leaf) < total_size) {
3422                 btrfs_print_leaf(root, leaf);
3423                 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3424                        total_size, btrfs_leaf_free_space(root, leaf));
3425                 BUG();
3426         }
3427
3428         slot = path->slots[0];
3429         BUG_ON(slot < 0);
3430
3431         if (slot != nritems) {
3432                 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3433
3434                 if (old_data < data_end) {
3435                         btrfs_print_leaf(root, leaf);
3436                         printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3437                                slot, old_data, data_end);
3438                         BUG_ON(1);
3439                 }
3440                 /*
3441                  * item0..itemN ... dataN.offset..dataN.size .. data0.size
3442                  */
3443                 /* first correct the data pointers */
3444                 WARN_ON(leaf->map_token);
3445                 for (i = slot; i < nritems; i++) {
3446                         u32 ioff;
3447
3448                         item = btrfs_item_nr(leaf, i);
3449                         if (!leaf->map_token) {
3450                                 map_extent_buffer(leaf, (unsigned long)item,
3451                                         sizeof(struct btrfs_item),
3452                                         &leaf->map_token, &leaf->kaddr,
3453                                         &leaf->map_start, &leaf->map_len,
3454                                         KM_USER1);
3455                         }
3456
3457                         ioff = btrfs_item_offset(leaf, item);
3458                         btrfs_set_item_offset(leaf, item, ioff - total_data);
3459                 }
3460                 if (leaf->map_token) {
3461                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3462                         leaf->map_token = NULL;
3463                 }
3464
3465                 /* shift the items */
3466                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3467                               btrfs_item_nr_offset(slot),
3468                               (nritems - slot) * sizeof(struct btrfs_item));
3469
3470                 /* shift the data */
3471                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3472                               data_end - total_data, btrfs_leaf_data(leaf) +
3473                               data_end, old_data - data_end);
3474                 data_end = old_data;
3475         }
3476
3477         /* setup the item for the new data */
3478         for (i = 0; i < nr; i++) {
3479                 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3480                 btrfs_set_item_key(leaf, &disk_key, slot + i);
3481                 item = btrfs_item_nr(leaf, slot + i);
3482                 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3483                 data_end -= data_size[i];
3484                 btrfs_set_item_size(leaf, item, data_size[i]);
3485         }
3486         btrfs_set_header_nritems(leaf, nritems + nr);
3487         btrfs_mark_buffer_dirty(leaf);
3488
3489         ret = 0;
3490         if (slot == 0) {
3491                 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3492                 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3493         }
3494
3495         if (btrfs_leaf_free_space(root, leaf) < 0) {
3496                 btrfs_print_leaf(root, leaf);
3497                 BUG();
3498         }
3499 out:
3500         btrfs_unlock_up_safe(path, 1);
3501         return ret;
3502 }
3503
3504 /*
3505  * Given a key and some data, insert an item into the tree.
3506  * This does all the path init required, making room in the tree if needed.
3507  */
3508 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3509                       *root, struct btrfs_key *cpu_key, void *data, u32
3510                       data_size)
3511 {
3512         int ret = 0;
3513         struct btrfs_path *path;
3514         struct extent_buffer *leaf;
3515         unsigned long ptr;
3516
3517         path = btrfs_alloc_path();
3518         BUG_ON(!path);
3519         ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3520         if (!ret) {
3521                 leaf = path->nodes[0];
3522                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3523                 write_extent_buffer(leaf, data, ptr, data_size);
3524                 btrfs_mark_buffer_dirty(leaf);
3525         }
3526         btrfs_free_path(path);
3527         return ret;
3528 }
3529
3530 /*
3531  * delete the pointer from a given node.
3532  *
3533  * the tree should have been previously balanced so the deletion does not
3534  * empty a node.
3535  */
3536 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3537                    struct btrfs_path *path, int level, int slot)
3538 {
3539         struct extent_buffer *parent = path->nodes[level];
3540         u32 nritems;
3541         int ret = 0;
3542         int wret;
3543
3544         nritems = btrfs_header_nritems(parent);
3545         if (slot != nritems - 1) {
3546                 memmove_extent_buffer(parent,
3547                               btrfs_node_key_ptr_offset(slot),
3548                               btrfs_node_key_ptr_offset(slot + 1),
3549                               sizeof(struct btrfs_key_ptr) *
3550                               (nritems - slot - 1));
3551         }
3552         nritems--;
3553         btrfs_set_header_nritems(parent, nritems);
3554         if (nritems == 0 && parent == root->node) {
3555                 BUG_ON(btrfs_header_level(root->node) != 1);
3556                 /* just turn the root into a leaf and break */
3557                 btrfs_set_header_level(root->node, 0);
3558         } else if (slot == 0) {
3559                 struct btrfs_disk_key disk_key;
3560
3561                 btrfs_node_key(parent, &disk_key, 0);
3562                 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3563                 if (wret)
3564                         ret = wret;
3565         }
3566         btrfs_mark_buffer_dirty(parent);
3567         return ret;
3568 }
3569
3570 /*
3571  * a helper function to delete the leaf pointed to by path->slots[1] and
3572  * path->nodes[1].  bytenr is the node block pointer, but since the callers
3573  * already know it, it is faster to have them pass it down than to
3574  * read it out of the node again.
3575  *
3576  * This deletes the pointer in path->nodes[1] and frees the leaf
3577  * block extent.  zero is returned if it all worked out, < 0 otherwise.
3578  *
3579  * The path must have already been setup for deleting the leaf, including
3580  * all the proper balancing.  path->nodes[1] must be locked.
3581  */
3582 noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3583                             struct btrfs_root *root,
3584                             struct btrfs_path *path, u64 bytenr)
3585 {
3586         int ret;
3587         u64 root_gen = btrfs_header_generation(path->nodes[1]);
3588         u64 parent_start = path->nodes[1]->start;
3589         u64 parent_owner = btrfs_header_owner(path->nodes[1]);
3590
3591         ret = del_ptr(trans, root, path, 1, path->slots[1]);
3592         if (ret)
3593                 return ret;
3594
3595         /*
3596          * btrfs_free_extent is expensive, we want to make sure we
3597          * aren't holding any locks when we call it
3598          */
3599         btrfs_unlock_up_safe(path, 0);
3600
3601         ret = btrfs_free_extent(trans, root, bytenr,
3602                                 btrfs_level_size(root, 0),
3603                                 parent_start, parent_owner,
3604                                 root_gen, 0, 1);
3605         return ret;
3606 }
3607 /*
3608  * delete the item at the leaf level in path.  If that empties
3609  * the leaf, remove it from the tree
3610  */
3611 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3612                     struct btrfs_path *path, int slot, int nr)
3613 {
3614         struct extent_buffer *leaf;
3615         struct btrfs_item *item;
3616         int last_off;
3617         int dsize = 0;
3618         int ret = 0;
3619         int wret;
3620         int i;
3621         u32 nritems;
3622
3623         leaf = path->nodes[0];
3624         last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3625
3626         for (i = 0; i < nr; i++)
3627                 dsize += btrfs_item_size_nr(leaf, slot + i);
3628
3629         nritems = btrfs_header_nritems(leaf);
3630
3631         if (slot + nr != nritems) {
3632                 int data_end = leaf_data_end(root, leaf);
3633
3634                 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3635                               data_end + dsize,
3636                               btrfs_leaf_data(leaf) + data_end,
3637                               last_off - data_end);
3638
3639                 for (i = slot + nr; i < nritems; i++) {
3640                         u32 ioff;
3641
3642                         item = btrfs_item_nr(leaf, i);
3643                         if (!leaf->map_token) {
3644                                 map_extent_buffer(leaf, (unsigned long)item,
3645                                         sizeof(struct btrfs_item),
3646                                         &leaf->map_token, &leaf->kaddr,
3647                                         &leaf->map_start, &leaf->map_len,
3648                                         KM_USER1);
3649                         }
3650                         ioff = btrfs_item_offset(leaf, item);
3651                         btrfs_set_item_offset(leaf, item, ioff + dsize);
3652                 }
3653
3654                 if (leaf->map_token) {
3655                         unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3656                         leaf->map_token = NULL;
3657                 }
3658
3659                 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3660                               btrfs_item_nr_offset(slot + nr),
3661                               sizeof(struct btrfs_item) *
3662                               (nritems - slot - nr));
3663         }
3664         btrfs_set_header_nritems(leaf, nritems - nr);
3665         nritems -= nr;
3666
3667         /* delete the leaf if we've emptied it */
3668         if (nritems == 0) {
3669                 if (leaf == root->node) {
3670                         btrfs_set_header_level(leaf, 0);
3671                 } else {
3672                         ret = btrfs_del_leaf(trans, root, path, leaf->start);
3673                         BUG_ON(ret);
3674                 }
3675         } else {
3676                 int used = leaf_space_used(leaf, 0, nritems);
3677                 if (slot == 0) {
3678                         struct btrfs_disk_key disk_key;
3679
3680                         btrfs_item_key(leaf, &disk_key, 0);
3681                         wret = fixup_low_keys(trans, root, path,
3682                                               &disk_key, 1);
3683                         if (wret)
3684                                 ret = wret;
3685                 }
3686
3687                 /* delete the leaf if it is mostly empty */
3688                 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
3689                         /* push_leaf_left fixes the path.
3690                          * make sure the path still points to our leaf
3691                          * for possible call to del_ptr below
3692                          */
3693                         slot = path->slots[1];
3694                         extent_buffer_get(leaf);
3695
3696                         wret = push_leaf_left(trans, root, path, 1, 1);
3697                         if (wret < 0 && wret != -ENOSPC)
3698                                 ret = wret;
3699
3700                         if (path->nodes[0] == leaf &&
3701                             btrfs_header_nritems(leaf)) {
3702                                 wret = push_leaf_right(trans, root, path, 1, 1);
3703                                 if (wret < 0 && wret != -ENOSPC)
3704                                         ret = wret;
3705                         }
3706
3707                         if (btrfs_header_nritems(leaf) == 0) {
3708                                 path->slots[1] = slot;
3709                                 ret = btrfs_del_leaf(trans, root, path,
3710                                                      leaf->start);
3711                                 BUG_ON(ret);
3712                                 free_extent_buffer(leaf);
3713                         } else {
3714                                 /* if we're still in the path, make sure
3715                                  * we're dirty.  Otherwise, one of the
3716                                  * push_leaf functions must have already
3717                                  * dirtied this buffer
3718                                  */
3719                                 if (path->nodes[0] == leaf)
3720                                         btrfs_mark_buffer_dirty(leaf);
3721                                 free_extent_buffer(leaf);
3722                         }
3723                 } else {
3724                         btrfs_mark_buffer_dirty(leaf);
3725                 }
3726         }
3727         return ret;
3728 }
3729
3730 /*
3731  * search the tree again to find a leaf with lesser keys
3732  * returns 0 if it found something or 1 if there are no lesser leaves.
3733  * returns < 0 on io errors.
3734  *
3735  * This may release the path, and so you may lose any locks held at the
3736  * time you call it.
3737  */
3738 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3739 {
3740         struct btrfs_key key;
3741         struct btrfs_disk_key found_key;
3742         int ret;
3743
3744         btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3745
3746         if (key.offset > 0)
3747                 key.offset--;
3748         else if (key.type > 0)
3749                 key.type--;
3750         else if (key.objectid > 0)
3751                 key.objectid--;
3752         else
3753                 return 1;
3754
3755         btrfs_release_path(root, path);
3756         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3757         if (ret < 0)
3758                 return ret;
3759         btrfs_item_key(path->nodes[0], &found_key, 0);
3760         ret = comp_keys(&found_key, &key);
3761         if (ret < 0)
3762                 return 0;
3763         return 1;
3764 }
3765
3766 /*
3767  * A helper function to walk down the tree starting at min_key, and looking
3768  * for nodes or leaves that are either in cache or have a minimum
3769  * transaction id.  This is used by the btree defrag code, and tree logging
3770  *
3771  * This does not cow, but it does stuff the starting key it finds back
3772  * into min_key, so you can call btrfs_search_slot with cow=1 on the
3773  * key and get a writable path.
3774  *
3775  * This does lock as it descends, and path->keep_locks should be set
3776  * to 1 by the caller.
3777  *
3778  * This honors path->lowest_level to prevent descent past a given level
3779  * of the tree.
3780  *
3781  * min_trans indicates the oldest transaction that you are interested
3782  * in walking through.  Any nodes or leaves older than min_trans are
3783  * skipped over (without reading them).
3784  *
3785  * returns zero if something useful was found, < 0 on error and 1 if there
3786  * was nothing in the tree that matched the search criteria.
3787  */
3788 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3789                          struct btrfs_key *max_key,
3790                          struct btrfs_path *path, int cache_only,
3791                          u64 min_trans)
3792 {
3793         struct extent_buffer *cur;
3794         struct btrfs_key found_key;
3795         int slot;
3796         int sret;
3797         u32 nritems;
3798         int level;
3799         int ret = 1;
3800
3801         WARN_ON(!path->keep_locks);
3802 again:
3803         cur = btrfs_lock_root_node(root);
3804         level = btrfs_header_level(cur);
3805         WARN_ON(path->nodes[level]);
3806         path->nodes[level] = cur;
3807         path->locks[level] = 1;
3808
3809         if (btrfs_header_generation(cur) < min_trans) {
3810                 ret = 1;
3811                 goto out;
3812         }
3813         while (1) {
3814                 nritems = btrfs_header_nritems(cur);
3815                 level = btrfs_header_level(cur);
3816                 sret = bin_search(cur, min_key, level, &slot);
3817
3818                 /* at the lowest level, we're done, setup the path and exit */
3819                 if (level == path->lowest_level) {
3820                         if (slot >= nritems)
3821                                 goto find_next_key;
3822                         ret = 0;
3823                         path->slots[level] = slot;
3824                         btrfs_item_key_to_cpu(cur, &found_key, slot);
3825                         goto out;
3826                 }
3827                 if (sret && slot > 0)
3828                         slot--;
3829                 /*
3830                  * check this node pointer against the cache_only and
3831                  * min_trans parameters.  If it isn't in cache or is too
3832                  * old, skip to the next one.
3833                  */
3834                 while (slot < nritems) {
3835                         u64 blockptr;
3836                         u64 gen;
3837                         struct extent_buffer *tmp;
3838                         struct btrfs_disk_key disk_key;
3839
3840                         blockptr = btrfs_node_blockptr(cur, slot);
3841                         gen = btrfs_node_ptr_generation(cur, slot);
3842                         if (gen < min_trans) {
3843                                 slot++;
3844                                 continue;
3845                         }
3846                         if (!cache_only)
3847                                 break;
3848
3849                         if (max_key) {
3850                                 btrfs_node_key(cur, &disk_key, slot);
3851                                 if (comp_keys(&disk_key, max_key) >= 0) {
3852                                         ret = 1;
3853                                         goto out;
3854                                 }
3855                         }
3856
3857                         tmp = btrfs_find_tree_block(root, blockptr,
3858                                             btrfs_level_size(root, level - 1));
3859
3860                         if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3861                                 free_extent_buffer(tmp);
3862                                 break;
3863                         }
3864                         if (tmp)
3865                                 free_extent_buffer(tmp);
3866                         slot++;
3867                 }
3868 find_next_key:
3869                 /*
3870                  * we didn't find a candidate key in this node, walk forward
3871                  * and find another one
3872                  */
3873                 if (slot >= nritems) {
3874                         path->slots[level] = slot;
3875                         btrfs_set_path_blocking(path);
3876                         sret = btrfs_find_next_key(root, path, min_key, level,
3877                                                   cache_only, min_trans);
3878                         if (sret == 0) {
3879                                 btrfs_release_path(root, path);
3880                                 goto again;
3881                         } else {
3882                                 goto out;
3883                         }
3884                 }
3885                 /* save our key for returning back */
3886                 btrfs_node_key_to_cpu(cur, &found_key, slot);
3887                 path->slots[level] = slot;
3888                 if (level == path->lowest_level) {
3889                         ret = 0;
3890                         unlock_up(path, level, 1);
3891                         goto out;
3892                 }
3893                 btrfs_set_path_blocking(path);
3894                 cur = read_node_slot(root, cur, slot);
3895
3896                 btrfs_tree_lock(cur);
3897
3898                 path->locks[level - 1] = 1;
3899                 path->nodes[level - 1] = cur;
3900                 unlock_up(path, level, 1);
3901                 btrfs_clear_path_blocking(path, NULL);
3902         }
3903 out:
3904         if (ret == 0)
3905                 memcpy(min_key, &found_key, sizeof(found_key));
3906         btrfs_set_path_blocking(path);
3907         return ret;
3908 }
3909
3910 /*
3911  * this is similar to btrfs_next_leaf, but does not try to preserve
3912  * and fixup the path.  It looks for and returns the next key in the
3913  * tree based on the current path and the cache_only and min_trans
3914  * parameters.
3915  *
3916  * 0 is returned if another key is found, < 0 if there are any errors
3917  * and 1 is returned if there are no higher keys in the tree
3918  *
3919  * path->keep_locks should be set to 1 on the search made before
3920  * calling this function.
3921  */
3922 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
3923                         struct btrfs_key *key, int lowest_level,
3924                         int cache_only, u64 min_trans)
3925 {
3926         int level = lowest_level;
3927         int slot;
3928         struct extent_buffer *c;
3929
3930         WARN_ON(!path->keep_locks);
3931         while (level < BTRFS_MAX_LEVEL) {
3932                 if (!path->nodes[level])
3933                         return 1;
3934
3935                 slot = path->slots[level] + 1;
3936                 c = path->nodes[level];
3937 next:
3938                 if (slot >= btrfs_header_nritems(c)) {
3939                         level++;
3940                         if (level == BTRFS_MAX_LEVEL)
3941                                 return 1;
3942                         continue;
3943                 }
3944                 if (level == 0)
3945                         btrfs_item_key_to_cpu(c, key, slot);
3946                 else {
3947                         u64 blockptr = btrfs_node_blockptr(c, slot);
3948                         u64 gen = btrfs_node_ptr_generation(c, slot);
3949
3950                         if (cache_only) {
3951                                 struct extent_buffer *cur;
3952                                 cur = btrfs_find_tree_block(root, blockptr,
3953                                             btrfs_level_size(root, level - 1));
3954                                 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
3955                                         slot++;
3956                                         if (cur)
3957                                                 free_extent_buffer(cur);
3958                                         goto next;
3959                                 }
3960                                 free_extent_buffer(cur);
3961                         }
3962                         if (gen < min_trans) {
3963                                 slot++;
3964                                 goto next;
3965                         }
3966                         btrfs_node_key_to_cpu(c, key, slot);
3967                 }
3968                 return 0;
3969         }
3970         return 1;
3971 }
3972
3973 /*
3974  * search the tree again to find a leaf with greater keys
3975  * returns 0 if it found something or 1 if there are no greater leaves.
3976  * returns < 0 on io errors.
3977  */
3978 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
3979 {
3980         int slot;
3981         int level = 1;
3982         struct extent_buffer *c;
3983         struct extent_buffer *next = NULL;
3984         struct btrfs_key key;
3985         u32 nritems;
3986         int ret;
3987
3988         nritems = btrfs_header_nritems(path->nodes[0]);
3989         if (nritems == 0)
3990                 return 1;
3991
3992         btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
3993
3994         btrfs_release_path(root, path);
3995         path->keep_locks = 1;
3996         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3997         path->keep_locks = 0;
3998
3999         if (ret < 0)
4000                 return ret;
4001
4002         btrfs_set_path_blocking(path);
4003         nritems = btrfs_header_nritems(path->nodes[0]);
4004         /*
4005          * by releasing the path above we dropped all our locks.  A balance
4006          * could have added more items next to the key that used to be
4007          * at the very end of the block.  So, check again here and
4008          * advance the path if there are now more items available.
4009          */
4010         if (nritems > 0 && path->slots[0] < nritems - 1) {
4011                 path->slots[0]++;
4012                 goto done;
4013         }
4014
4015         while (level < BTRFS_MAX_LEVEL) {
4016                 if (!path->nodes[level])
4017                         return 1;
4018
4019                 slot = path->slots[level] + 1;
4020                 c = path->nodes[level];
4021                 if (slot >= btrfs_header_nritems(c)) {
4022                         level++;
4023                         if (level == BTRFS_MAX_LEVEL)
4024                                 return 1;
4025                         continue;
4026                 }
4027
4028                 if (next) {
4029                         btrfs_tree_unlock(next);
4030                         free_extent_buffer(next);
4031                 }
4032
4033                 /* the path was set to blocking above */
4034                 if (level == 1 && (path->locks[1] || path->skip_locking) &&
4035                     path->reada)
4036                         reada_for_search(root, path, level, slot, 0);
4037
4038                 next = read_node_slot(root, c, slot);
4039                 if (!path->skip_locking) {
4040                         btrfs_assert_tree_locked(c);
4041                         btrfs_tree_lock(next);
4042                         btrfs_set_lock_blocking(next);
4043                 }
4044                 break;
4045         }
4046         path->slots[level] = slot;
4047         while (1) {
4048                 level--;
4049                 c = path->nodes[level];
4050                 if (path->locks[level])
4051                         btrfs_tree_unlock(c);
4052                 free_extent_buffer(c);
4053                 path->nodes[level] = next;
4054                 path->slots[level] = 0;
4055                 if (!path->skip_locking)
4056                         path->locks[level] = 1;
4057                 if (!level)
4058                         break;
4059
4060                 btrfs_set_path_blocking(path);
4061                 if (level == 1 && path->locks[1] && path->reada)
4062                         reada_for_search(root, path, level, slot, 0);
4063                 next = read_node_slot(root, next, 0);
4064                 if (!path->skip_locking) {
4065                         btrfs_assert_tree_locked(path->nodes[level]);
4066                         btrfs_tree_lock(next);
4067                         btrfs_set_lock_blocking(next);
4068                 }
4069         }
4070 done:
4071         unlock_up(path, 0, 1);
4072         return 0;
4073 }
4074
4075 /*
4076  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4077  * searching until it gets past min_objectid or finds an item of 'type'
4078  *
4079  * returns 0 if something is found, 1 if nothing was found and < 0 on error
4080  */
4081 int btrfs_previous_item(struct btrfs_root *root,
4082                         struct btrfs_path *path, u64 min_objectid,
4083                         int type)
4084 {
4085         struct btrfs_key found_key;
4086         struct extent_buffer *leaf;
4087         u32 nritems;
4088         int ret;
4089
4090         while (1) {
4091                 if (path->slots[0] == 0) {
4092                         btrfs_set_path_blocking(path);
4093                         ret = btrfs_prev_leaf(root, path);
4094                         if (ret != 0)
4095                                 return ret;
4096                 } else {
4097                         path->slots[0]--;
4098                 }
4099                 leaf = path->nodes[0];
4100                 nritems = btrfs_header_nritems(leaf);
4101                 if (nritems == 0)
4102                         return 1;
4103                 if (path->slots[0] == nritems)
4104                         path->slots[0]--;
4105
4106                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4107                 if (found_key.type == type)
4108                         return 0;
4109                 if (found_key.objectid < min_objectid)
4110                         break;
4111                 if (found_key.objectid == min_objectid &&
4112                     found_key.type < type)
4113                         break;
4114         }
4115         return 1;
4116 }