2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
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.
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.
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.
19 #include <linux/sched.h>
22 #include "transaction.h"
23 #include "print-tree.h"
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);
41 inline void btrfs_init_path(struct btrfs_path *p)
43 memset(p, 0, sizeof(*p));
46 struct btrfs_path *btrfs_alloc_path(void)
48 struct btrfs_path *path;
49 path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS);
51 btrfs_init_path(path);
57 /* this also releases the path */
58 void btrfs_free_path(struct btrfs_path *p)
60 btrfs_release_path(NULL, p);
61 kmem_cache_free(btrfs_path_cachep, p);
65 * path release drops references on the extent buffers in the path
66 * and it drops any locks held by this path
68 * It is safe to call this on paths that no locks or extent buffers held.
70 void noinline btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
74 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
79 btrfs_tree_unlock(p->nodes[i]);
82 free_extent_buffer(p->nodes[i]);
88 * safely gets a reference on the root node of a tree. A lock
89 * is not taken, so a concurrent writer may put a different node
90 * at the root of the tree. See btrfs_lock_root_node for the
93 * The extent buffer returned by this has a reference taken, so
94 * it won't disappear. It may stop being the root of the tree
95 * at any time because there are no locks held.
97 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
99 struct extent_buffer *eb;
100 spin_lock(&root->node_lock);
102 extent_buffer_get(eb);
103 spin_unlock(&root->node_lock);
107 /* loop around taking references on and locking the root node of the
108 * tree until you end up with a lock on the root. A locked buffer
109 * is returned, with a reference held.
111 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
113 struct extent_buffer *eb;
116 eb = btrfs_root_node(root);
119 spin_lock(&root->node_lock);
120 if (eb == root->node) {
121 spin_unlock(&root->node_lock);
124 spin_unlock(&root->node_lock);
126 btrfs_tree_unlock(eb);
127 free_extent_buffer(eb);
132 /* cowonly root (everything not a reference counted cow subvolume), just get
133 * put onto a simple dirty list. transaction.c walks this to make sure they
134 * get properly updated on disk.
136 static void add_root_to_dirty_list(struct btrfs_root *root)
138 if (root->track_dirty && list_empty(&root->dirty_list)) {
139 list_add(&root->dirty_list,
140 &root->fs_info->dirty_cowonly_roots);
145 * used by snapshot creation to make a copy of a root for a tree with
146 * a given objectid. The buffer with the new root node is returned in
147 * cow_ret, and this func returns zero on success or a negative error code.
149 int btrfs_copy_root(struct btrfs_trans_handle *trans,
150 struct btrfs_root *root,
151 struct extent_buffer *buf,
152 struct extent_buffer **cow_ret, u64 new_root_objectid)
154 struct extent_buffer *cow;
158 struct btrfs_root *new_root;
160 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
164 memcpy(new_root, root, sizeof(*new_root));
165 new_root->root_key.objectid = new_root_objectid;
167 WARN_ON(root->ref_cows && trans->transid !=
168 root->fs_info->running_transaction->transid);
169 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
171 level = btrfs_header_level(buf);
172 nritems = btrfs_header_nritems(buf);
174 cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0,
175 new_root_objectid, trans->transid,
176 level, buf->start, 0);
182 copy_extent_buffer(cow, buf, 0, 0, cow->len);
183 btrfs_set_header_bytenr(cow, cow->start);
184 btrfs_set_header_generation(cow, trans->transid);
185 btrfs_set_header_owner(cow, new_root_objectid);
186 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
188 WARN_ON(btrfs_header_generation(buf) > trans->transid);
189 ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL);
195 btrfs_mark_buffer_dirty(cow);
201 * does the dirty work in cow of a single block. The parent block
202 * (if supplied) is updated to point to the new cow copy. The new
203 * buffer is marked dirty and returned locked. If you modify the block
204 * it needs to be marked dirty again.
206 * search_start -- an allocation hint for the new block
208 * empty_size -- a hint that you plan on doing more cow. This is the size in bytes
209 * the allocator should try to find free next to the block it returns. This is
210 * just a hint and may be ignored by the allocator.
212 * prealloc_dest -- if you have already reserved a destination for the cow,
213 * this uses that block instead of allocating a new one. btrfs_alloc_reserved_extent
214 * is used to finish the allocation.
216 int noinline __btrfs_cow_block(struct btrfs_trans_handle *trans,
217 struct btrfs_root *root,
218 struct extent_buffer *buf,
219 struct extent_buffer *parent, int parent_slot,
220 struct extent_buffer **cow_ret,
221 u64 search_start, u64 empty_size,
225 struct extent_buffer *cow;
234 WARN_ON(!btrfs_tree_locked(buf));
237 parent_start = parent->start;
241 WARN_ON(root->ref_cows && trans->transid !=
242 root->fs_info->running_transaction->transid);
243 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
245 level = btrfs_header_level(buf);
246 nritems = btrfs_header_nritems(buf);
249 struct btrfs_key ins;
251 ins.objectid = prealloc_dest;
252 ins.offset = buf->len;
253 ins.type = BTRFS_EXTENT_ITEM_KEY;
255 ret = btrfs_alloc_reserved_extent(trans, root, parent_start,
256 root->root_key.objectid,
257 trans->transid, level, &ins);
259 cow = btrfs_init_new_buffer(trans, root, prealloc_dest,
262 cow = btrfs_alloc_free_block(trans, root, buf->len,
264 root->root_key.objectid,
265 trans->transid, level,
266 search_start, empty_size);
271 copy_extent_buffer(cow, buf, 0, 0, cow->len);
272 btrfs_set_header_bytenr(cow, cow->start);
273 btrfs_set_header_generation(cow, trans->transid);
274 btrfs_set_header_owner(cow, root->root_key.objectid);
275 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
277 WARN_ON(btrfs_header_generation(buf) > trans->transid);
278 if (btrfs_header_generation(buf) != trans->transid) {
280 ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents);
284 ret = btrfs_cache_ref(trans, root, buf, nr_extents);
286 } else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) {
288 * There are only two places that can drop reference to
289 * tree blocks owned by living reloc trees, one is here,
290 * the other place is btrfs_drop_subtree. In both places,
291 * we check reference count while tree block is locked.
292 * Furthermore, if reference count is one, it won't get
293 * increased by someone else.
296 ret = btrfs_lookup_extent_ref(trans, root, buf->start,
300 ret = btrfs_update_ref(trans, root, buf, cow,
302 clean_tree_block(trans, root, buf);
304 ret = btrfs_inc_ref(trans, root, buf, cow, NULL);
308 ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems);
311 clean_tree_block(trans, root, buf);
314 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
315 ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start);
319 if (buf == root->node) {
320 WARN_ON(parent && parent != buf);
322 spin_lock(&root->node_lock);
324 extent_buffer_get(cow);
325 spin_unlock(&root->node_lock);
327 if (buf != root->commit_root) {
328 btrfs_free_extent(trans, root, buf->start,
329 buf->len, buf->start,
330 root->root_key.objectid,
331 btrfs_header_generation(buf),
334 free_extent_buffer(buf);
335 add_root_to_dirty_list(root);
337 btrfs_set_node_blockptr(parent, parent_slot,
339 WARN_ON(trans->transid == 0);
340 btrfs_set_node_ptr_generation(parent, parent_slot,
342 btrfs_mark_buffer_dirty(parent);
343 WARN_ON(btrfs_header_generation(parent) != trans->transid);
344 btrfs_free_extent(trans, root, buf->start, buf->len,
345 parent_start, btrfs_header_owner(parent),
346 btrfs_header_generation(parent), level, 1);
349 btrfs_tree_unlock(buf);
350 free_extent_buffer(buf);
351 btrfs_mark_buffer_dirty(cow);
357 * cows a single block, see __btrfs_cow_block for the real work.
358 * This version of it has extra checks so that a block isn't cow'd more than
359 * once per transaction, as long as it hasn't been written yet
361 int noinline btrfs_cow_block(struct btrfs_trans_handle *trans,
362 struct btrfs_root *root, struct extent_buffer *buf,
363 struct extent_buffer *parent, int parent_slot,
364 struct extent_buffer **cow_ret, u64 prealloc_dest)
369 if (trans->transaction != root->fs_info->running_transaction) {
370 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
371 root->fs_info->running_transaction->transid);
374 if (trans->transid != root->fs_info->generation) {
375 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
376 root->fs_info->generation);
380 spin_lock(&root->fs_info->hash_lock);
381 if (btrfs_header_generation(buf) == trans->transid &&
382 btrfs_header_owner(buf) == root->root_key.objectid &&
383 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
385 spin_unlock(&root->fs_info->hash_lock);
386 WARN_ON(prealloc_dest);
389 spin_unlock(&root->fs_info->hash_lock);
390 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
391 ret = __btrfs_cow_block(trans, root, buf, parent,
392 parent_slot, cow_ret, search_start, 0,
398 * helper function for defrag to decide if two blocks pointed to by a
399 * node are actually close by
401 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
403 if (blocknr < other && other - (blocknr + blocksize) < 32768)
405 if (blocknr > other && blocknr - (other + blocksize) < 32768)
411 * compare two keys in a memcmp fashion
413 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
417 btrfs_disk_key_to_cpu(&k1, disk);
419 if (k1.objectid > k2->objectid)
421 if (k1.objectid < k2->objectid)
423 if (k1.type > k2->type)
425 if (k1.type < k2->type)
427 if (k1.offset > k2->offset)
429 if (k1.offset < k2->offset)
436 * this is used by the defrag code to go through all the
437 * leaves pointed to by a node and reallocate them so that
438 * disk order is close to key order
440 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
441 struct btrfs_root *root, struct extent_buffer *parent,
442 int start_slot, int cache_only, u64 *last_ret,
443 struct btrfs_key *progress)
445 struct extent_buffer *cur;
448 u64 search_start = *last_ret;
458 int progress_passed = 0;
459 struct btrfs_disk_key disk_key;
461 parent_level = btrfs_header_level(parent);
462 if (cache_only && parent_level != 1)
465 if (trans->transaction != root->fs_info->running_transaction) {
466 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
467 root->fs_info->running_transaction->transid);
470 if (trans->transid != root->fs_info->generation) {
471 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
472 root->fs_info->generation);
476 parent_nritems = btrfs_header_nritems(parent);
477 blocksize = btrfs_level_size(root, parent_level - 1);
478 end_slot = parent_nritems;
480 if (parent_nritems == 1)
483 for (i = start_slot; i < end_slot; i++) {
486 if (!parent->map_token) {
487 map_extent_buffer(parent,
488 btrfs_node_key_ptr_offset(i),
489 sizeof(struct btrfs_key_ptr),
490 &parent->map_token, &parent->kaddr,
491 &parent->map_start, &parent->map_len,
494 btrfs_node_key(parent, &disk_key, i);
495 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
499 blocknr = btrfs_node_blockptr(parent, i);
500 gen = btrfs_node_ptr_generation(parent, i);
502 last_block = blocknr;
505 other = btrfs_node_blockptr(parent, i - 1);
506 close = close_blocks(blocknr, other, blocksize);
508 if (!close && i < end_slot - 2) {
509 other = btrfs_node_blockptr(parent, i + 1);
510 close = close_blocks(blocknr, other, blocksize);
513 last_block = blocknr;
516 if (parent->map_token) {
517 unmap_extent_buffer(parent, parent->map_token,
519 parent->map_token = NULL;
522 cur = btrfs_find_tree_block(root, blocknr, blocksize);
524 uptodate = btrfs_buffer_uptodate(cur, gen);
527 if (!cur || !uptodate) {
529 free_extent_buffer(cur);
533 cur = read_tree_block(root, blocknr,
535 } else if (!uptodate) {
536 btrfs_read_buffer(cur, gen);
539 if (search_start == 0)
540 search_start = last_block;
542 btrfs_tree_lock(cur);
543 err = __btrfs_cow_block(trans, root, cur, parent, i,
546 (end_slot - i) * blocksize), 0);
548 btrfs_tree_unlock(cur);
549 free_extent_buffer(cur);
552 search_start = cur->start;
553 last_block = cur->start;
554 *last_ret = search_start;
555 btrfs_tree_unlock(cur);
556 free_extent_buffer(cur);
558 if (parent->map_token) {
559 unmap_extent_buffer(parent, parent->map_token,
561 parent->map_token = NULL;
567 * The leaf data grows from end-to-front in the node.
568 * this returns the address of the start of the last item,
569 * which is the stop of the leaf data stack
571 static inline unsigned int leaf_data_end(struct btrfs_root *root,
572 struct extent_buffer *leaf)
574 u32 nr = btrfs_header_nritems(leaf);
576 return BTRFS_LEAF_DATA_SIZE(root);
577 return btrfs_item_offset_nr(leaf, nr - 1);
581 * extra debugging checks to make sure all the items in a key are
582 * well formed and in the proper order
584 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
587 struct extent_buffer *parent = NULL;
588 struct extent_buffer *node = path->nodes[level];
589 struct btrfs_disk_key parent_key;
590 struct btrfs_disk_key node_key;
593 struct btrfs_key cpukey;
594 u32 nritems = btrfs_header_nritems(node);
596 if (path->nodes[level + 1])
597 parent = path->nodes[level + 1];
599 slot = path->slots[level];
600 BUG_ON(nritems == 0);
602 parent_slot = path->slots[level + 1];
603 btrfs_node_key(parent, &parent_key, parent_slot);
604 btrfs_node_key(node, &node_key, 0);
605 BUG_ON(memcmp(&parent_key, &node_key,
606 sizeof(struct btrfs_disk_key)));
607 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
608 btrfs_header_bytenr(node));
610 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
612 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
613 btrfs_node_key(node, &node_key, slot);
614 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
616 if (slot < nritems - 1) {
617 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
618 btrfs_node_key(node, &node_key, slot);
619 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
625 * extra checking to make sure all the items in a leaf are
626 * well formed and in the proper order
628 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
631 struct extent_buffer *leaf = path->nodes[level];
632 struct extent_buffer *parent = NULL;
634 struct btrfs_key cpukey;
635 struct btrfs_disk_key parent_key;
636 struct btrfs_disk_key leaf_key;
637 int slot = path->slots[0];
639 u32 nritems = btrfs_header_nritems(leaf);
641 if (path->nodes[level + 1])
642 parent = path->nodes[level + 1];
648 parent_slot = path->slots[level + 1];
649 btrfs_node_key(parent, &parent_key, parent_slot);
650 btrfs_item_key(leaf, &leaf_key, 0);
652 BUG_ON(memcmp(&parent_key, &leaf_key,
653 sizeof(struct btrfs_disk_key)));
654 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
655 btrfs_header_bytenr(leaf));
658 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
659 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
660 btrfs_item_key(leaf, &leaf_key, i);
661 if (comp_keys(&leaf_key, &cpukey) >= 0) {
662 btrfs_print_leaf(root, leaf);
663 printk("slot %d offset bad key\n", i);
666 if (btrfs_item_offset_nr(leaf, i) !=
667 btrfs_item_end_nr(leaf, i + 1)) {
668 btrfs_print_leaf(root, leaf);
669 printk("slot %d offset bad\n", i);
673 if (btrfs_item_offset_nr(leaf, i) +
674 btrfs_item_size_nr(leaf, i) !=
675 BTRFS_LEAF_DATA_SIZE(root)) {
676 btrfs_print_leaf(root, leaf);
677 printk("slot %d first offset bad\n", i);
683 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
684 btrfs_print_leaf(root, leaf);
685 printk("slot %d bad size \n", nritems - 1);
690 if (slot != 0 && slot < nritems - 1) {
691 btrfs_item_key(leaf, &leaf_key, slot);
692 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
693 if (comp_keys(&leaf_key, &cpukey) <= 0) {
694 btrfs_print_leaf(root, leaf);
695 printk("slot %d offset bad key\n", slot);
698 if (btrfs_item_offset_nr(leaf, slot - 1) !=
699 btrfs_item_end_nr(leaf, slot)) {
700 btrfs_print_leaf(root, leaf);
701 printk("slot %d offset bad\n", slot);
705 if (slot < nritems - 1) {
706 btrfs_item_key(leaf, &leaf_key, slot);
707 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
708 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
709 if (btrfs_item_offset_nr(leaf, slot) !=
710 btrfs_item_end_nr(leaf, slot + 1)) {
711 btrfs_print_leaf(root, leaf);
712 printk("slot %d offset bad\n", slot);
716 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
717 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
721 static int noinline check_block(struct btrfs_root *root,
722 struct btrfs_path *path, int level)
726 if (btrfs_header_level(path->nodes[level]) != level)
727 printk("warning: bad level %Lu wanted %d found %d\n",
728 path->nodes[level]->start, level,
729 btrfs_header_level(path->nodes[level]));
730 found_start = btrfs_header_bytenr(path->nodes[level]);
731 if (found_start != path->nodes[level]->start) {
732 printk("warning: bad bytentr %Lu found %Lu\n",
733 path->nodes[level]->start, found_start);
736 struct extent_buffer *buf = path->nodes[level];
738 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
739 (unsigned long)btrfs_header_fsid(buf),
741 printk("warning bad block %Lu\n", buf->start);
746 return check_leaf(root, path, level);
747 return check_node(root, path, level);
751 * search for key in the extent_buffer. The items start at offset p,
752 * and they are item_size apart. There are 'max' items in p.
754 * the slot in the array is returned via slot, and it points to
755 * the place where you would insert key if it is not found in
758 * slot may point to max if the key is bigger than all of the keys
760 static noinline int generic_bin_search(struct extent_buffer *eb,
762 int item_size, struct btrfs_key *key,
769 struct btrfs_disk_key *tmp = NULL;
770 struct btrfs_disk_key unaligned;
771 unsigned long offset;
772 char *map_token = NULL;
774 unsigned long map_start = 0;
775 unsigned long map_len = 0;
779 mid = (low + high) / 2;
780 offset = p + mid * item_size;
782 if (!map_token || offset < map_start ||
783 (offset + sizeof(struct btrfs_disk_key)) >
784 map_start + map_len) {
786 unmap_extent_buffer(eb, map_token, KM_USER0);
789 err = map_extent_buffer(eb, offset,
790 sizeof(struct btrfs_disk_key),
792 &map_start, &map_len, KM_USER0);
795 tmp = (struct btrfs_disk_key *)(kaddr + offset -
798 read_extent_buffer(eb, &unaligned,
799 offset, sizeof(unaligned));
804 tmp = (struct btrfs_disk_key *)(kaddr + offset -
807 ret = comp_keys(tmp, key);
816 unmap_extent_buffer(eb, map_token, KM_USER0);
822 unmap_extent_buffer(eb, map_token, KM_USER0);
827 * simple bin_search frontend that does the right thing for
830 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
831 int level, int *slot)
834 return generic_bin_search(eb,
835 offsetof(struct btrfs_leaf, items),
836 sizeof(struct btrfs_item),
837 key, btrfs_header_nritems(eb),
840 return generic_bin_search(eb,
841 offsetof(struct btrfs_node, ptrs),
842 sizeof(struct btrfs_key_ptr),
843 key, btrfs_header_nritems(eb),
849 /* given a node and slot number, this reads the blocks it points to. The
850 * extent buffer is returned with a reference taken (but unlocked).
851 * NULL is returned on error.
853 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
854 struct extent_buffer *parent, int slot)
856 int level = btrfs_header_level(parent);
859 if (slot >= btrfs_header_nritems(parent))
864 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
865 btrfs_level_size(root, level - 1),
866 btrfs_node_ptr_generation(parent, slot));
870 * node level balancing, used to make sure nodes are in proper order for
871 * item deletion. We balance from the top down, so we have to make sure
872 * that a deletion won't leave an node completely empty later on.
874 static noinline int balance_level(struct btrfs_trans_handle *trans,
875 struct btrfs_root *root,
876 struct btrfs_path *path, int level)
878 struct extent_buffer *right = NULL;
879 struct extent_buffer *mid;
880 struct extent_buffer *left = NULL;
881 struct extent_buffer *parent = NULL;
885 int orig_slot = path->slots[level];
886 int err_on_enospc = 0;
892 mid = path->nodes[level];
893 WARN_ON(!path->locks[level]);
894 WARN_ON(btrfs_header_generation(mid) != trans->transid);
896 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
898 if (level < BTRFS_MAX_LEVEL - 1)
899 parent = path->nodes[level + 1];
900 pslot = path->slots[level + 1];
903 * deal with the case where there is only one pointer in the root
904 * by promoting the node below to a root
907 struct extent_buffer *child;
909 if (btrfs_header_nritems(mid) != 1)
912 /* promote the child to a root */
913 child = read_node_slot(root, mid, 0);
914 btrfs_tree_lock(child);
916 ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0);
919 spin_lock(&root->node_lock);
921 spin_unlock(&root->node_lock);
923 ret = btrfs_update_extent_ref(trans, root, child->start,
924 mid->start, child->start,
925 root->root_key.objectid,
926 trans->transid, level - 1);
929 add_root_to_dirty_list(root);
930 btrfs_tree_unlock(child);
931 path->locks[level] = 0;
932 path->nodes[level] = NULL;
933 clean_tree_block(trans, root, mid);
934 btrfs_tree_unlock(mid);
935 /* once for the path */
936 free_extent_buffer(mid);
937 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
938 mid->start, root->root_key.objectid,
939 btrfs_header_generation(mid),
941 /* once for the root ptr */
942 free_extent_buffer(mid);
945 if (btrfs_header_nritems(mid) >
946 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
949 if (btrfs_header_nritems(mid) < 2)
952 left = read_node_slot(root, parent, pslot - 1);
954 btrfs_tree_lock(left);
955 wret = btrfs_cow_block(trans, root, left,
956 parent, pslot - 1, &left, 0);
962 right = read_node_slot(root, parent, pslot + 1);
964 btrfs_tree_lock(right);
965 wret = btrfs_cow_block(trans, root, right,
966 parent, pslot + 1, &right, 0);
973 /* first, try to make some room in the middle buffer */
975 orig_slot += btrfs_header_nritems(left);
976 wret = push_node_left(trans, root, left, mid, 1);
979 if (btrfs_header_nritems(mid) < 2)
984 * then try to empty the right most buffer into the middle
987 wret = push_node_left(trans, root, mid, right, 1);
988 if (wret < 0 && wret != -ENOSPC)
990 if (btrfs_header_nritems(right) == 0) {
991 u64 bytenr = right->start;
992 u64 generation = btrfs_header_generation(parent);
993 u32 blocksize = right->len;
995 clean_tree_block(trans, root, right);
996 btrfs_tree_unlock(right);
997 free_extent_buffer(right);
999 wret = del_ptr(trans, root, path, level + 1, pslot +
1003 wret = btrfs_free_extent(trans, root, bytenr,
1004 blocksize, parent->start,
1005 btrfs_header_owner(parent),
1006 generation, level, 1);
1010 struct btrfs_disk_key right_key;
1011 btrfs_node_key(right, &right_key, 0);
1012 btrfs_set_node_key(parent, &right_key, pslot + 1);
1013 btrfs_mark_buffer_dirty(parent);
1016 if (btrfs_header_nritems(mid) == 1) {
1018 * we're not allowed to leave a node with one item in the
1019 * tree during a delete. A deletion from lower in the tree
1020 * could try to delete the only pointer in this node.
1021 * So, pull some keys from the left.
1022 * There has to be a left pointer at this point because
1023 * otherwise we would have pulled some pointers from the
1027 wret = balance_node_right(trans, root, mid, left);
1033 wret = push_node_left(trans, root, left, mid, 1);
1039 if (btrfs_header_nritems(mid) == 0) {
1040 /* we've managed to empty the middle node, drop it */
1041 u64 root_gen = btrfs_header_generation(parent);
1042 u64 bytenr = mid->start;
1043 u32 blocksize = mid->len;
1045 clean_tree_block(trans, root, mid);
1046 btrfs_tree_unlock(mid);
1047 free_extent_buffer(mid);
1049 wret = del_ptr(trans, root, path, level + 1, pslot);
1052 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
1054 btrfs_header_owner(parent),
1055 root_gen, level, 1);
1059 /* update the parent key to reflect our changes */
1060 struct btrfs_disk_key mid_key;
1061 btrfs_node_key(mid, &mid_key, 0);
1062 btrfs_set_node_key(parent, &mid_key, pslot);
1063 btrfs_mark_buffer_dirty(parent);
1066 /* update the path */
1068 if (btrfs_header_nritems(left) > orig_slot) {
1069 extent_buffer_get(left);
1070 /* left was locked after cow */
1071 path->nodes[level] = left;
1072 path->slots[level + 1] -= 1;
1073 path->slots[level] = orig_slot;
1075 btrfs_tree_unlock(mid);
1076 free_extent_buffer(mid);
1079 orig_slot -= btrfs_header_nritems(left);
1080 path->slots[level] = orig_slot;
1083 /* double check we haven't messed things up */
1084 check_block(root, path, level);
1086 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1090 btrfs_tree_unlock(right);
1091 free_extent_buffer(right);
1094 if (path->nodes[level] != left)
1095 btrfs_tree_unlock(left);
1096 free_extent_buffer(left);
1101 /* Node balancing for insertion. Here we only split or push nodes around
1102 * when they are completely full. This is also done top down, so we
1103 * have to be pessimistic.
1105 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
1106 struct btrfs_root *root,
1107 struct btrfs_path *path, int level)
1109 struct extent_buffer *right = NULL;
1110 struct extent_buffer *mid;
1111 struct extent_buffer *left = NULL;
1112 struct extent_buffer *parent = NULL;
1116 int orig_slot = path->slots[level];
1122 mid = path->nodes[level];
1123 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1124 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1126 if (level < BTRFS_MAX_LEVEL - 1)
1127 parent = path->nodes[level + 1];
1128 pslot = path->slots[level + 1];
1133 left = read_node_slot(root, parent, pslot - 1);
1135 /* first, try to make some room in the middle buffer */
1139 btrfs_tree_lock(left);
1140 left_nr = btrfs_header_nritems(left);
1141 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1144 ret = btrfs_cow_block(trans, root, left, parent,
1145 pslot - 1, &left, 0);
1149 wret = push_node_left(trans, root,
1156 struct btrfs_disk_key disk_key;
1157 orig_slot += left_nr;
1158 btrfs_node_key(mid, &disk_key, 0);
1159 btrfs_set_node_key(parent, &disk_key, pslot);
1160 btrfs_mark_buffer_dirty(parent);
1161 if (btrfs_header_nritems(left) > orig_slot) {
1162 path->nodes[level] = left;
1163 path->slots[level + 1] -= 1;
1164 path->slots[level] = orig_slot;
1165 btrfs_tree_unlock(mid);
1166 free_extent_buffer(mid);
1169 btrfs_header_nritems(left);
1170 path->slots[level] = orig_slot;
1171 btrfs_tree_unlock(left);
1172 free_extent_buffer(left);
1176 btrfs_tree_unlock(left);
1177 free_extent_buffer(left);
1179 right = read_node_slot(root, parent, pslot + 1);
1182 * then try to empty the right most buffer into the middle
1186 btrfs_tree_lock(right);
1187 right_nr = btrfs_header_nritems(right);
1188 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1191 ret = btrfs_cow_block(trans, root, right,
1197 wret = balance_node_right(trans, root,
1204 struct btrfs_disk_key disk_key;
1206 btrfs_node_key(right, &disk_key, 0);
1207 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1208 btrfs_mark_buffer_dirty(parent);
1210 if (btrfs_header_nritems(mid) <= orig_slot) {
1211 path->nodes[level] = right;
1212 path->slots[level + 1] += 1;
1213 path->slots[level] = orig_slot -
1214 btrfs_header_nritems(mid);
1215 btrfs_tree_unlock(mid);
1216 free_extent_buffer(mid);
1218 btrfs_tree_unlock(right);
1219 free_extent_buffer(right);
1223 btrfs_tree_unlock(right);
1224 free_extent_buffer(right);
1230 * readahead one full node of leaves, finding things that are close
1231 * to the block in 'slot', and triggering ra on them.
1233 static noinline void reada_for_search(struct btrfs_root *root,
1234 struct btrfs_path *path,
1235 int level, int slot, u64 objectid)
1237 struct extent_buffer *node;
1238 struct btrfs_disk_key disk_key;
1244 int direction = path->reada;
1245 struct extent_buffer *eb;
1253 if (!path->nodes[level])
1256 node = path->nodes[level];
1258 search = btrfs_node_blockptr(node, slot);
1259 blocksize = btrfs_level_size(root, level - 1);
1260 eb = btrfs_find_tree_block(root, search, blocksize);
1262 free_extent_buffer(eb);
1266 highest_read = search;
1267 lowest_read = search;
1269 nritems = btrfs_header_nritems(node);
1272 if (direction < 0) {
1276 } else if (direction > 0) {
1281 if (path->reada < 0 && objectid) {
1282 btrfs_node_key(node, &disk_key, nr);
1283 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1286 search = btrfs_node_blockptr(node, nr);
1287 if ((search >= lowest_read && search <= highest_read) ||
1288 (search < lowest_read && lowest_read - search <= 32768) ||
1289 (search > highest_read && search - highest_read <= 32768)) {
1290 readahead_tree_block(root, search, blocksize,
1291 btrfs_node_ptr_generation(node, nr));
1295 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1297 if(nread > (1024 * 1024) || nscan > 128)
1300 if (search < lowest_read)
1301 lowest_read = search;
1302 if (search > highest_read)
1303 highest_read = search;
1308 * when we walk down the tree, it is usually safe to unlock the higher layers in
1309 * the tree. The exceptions are when our path goes through slot 0, because operations
1310 * on the tree might require changing key pointers higher up in the tree.
1312 * callers might also have set path->keep_locks, which tells this code to
1313 * keep the lock if the path points to the last slot in the block. This is
1314 * part of walking through the tree, and selecting the next slot in the higher
1317 * lowest_unlock sets the lowest level in the tree we're allowed to unlock.
1318 * so if lowest_unlock is 1, level 0 won't be unlocked
1320 static noinline void unlock_up(struct btrfs_path *path, int level,
1324 int skip_level = level;
1326 struct extent_buffer *t;
1328 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1329 if (!path->nodes[i])
1331 if (!path->locks[i])
1333 if (!no_skips && path->slots[i] == 0) {
1337 if (!no_skips && path->keep_locks) {
1340 nritems = btrfs_header_nritems(t);
1341 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1346 if (skip_level < i && i >= lowest_unlock)
1350 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1351 btrfs_tree_unlock(t);
1358 * look for key in the tree. path is filled in with nodes along the way
1359 * if key is found, we return zero and you can find the item in the leaf
1360 * level of the path (level 0)
1362 * If the key isn't found, the path points to the slot where it should
1363 * be inserted, and 1 is returned. If there are other errors during the
1364 * search a negative error number is returned.
1366 * if ins_len > 0, nodes and leaves will be split as we walk down the
1367 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1370 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1371 *root, struct btrfs_key *key, struct btrfs_path *p, int
1374 struct extent_buffer *b;
1375 struct extent_buffer *tmp;
1379 int should_reada = p->reada;
1380 int lowest_unlock = 1;
1382 u8 lowest_level = 0;
1385 struct btrfs_key prealloc_block;
1387 lowest_level = p->lowest_level;
1388 WARN_ON(lowest_level && ins_len > 0);
1389 WARN_ON(p->nodes[0] != NULL);
1390 WARN_ON(cow && root == root->fs_info->extent_root &&
1391 !mutex_is_locked(&root->fs_info->alloc_mutex));
1395 prealloc_block.objectid = 0;
1398 if (p->skip_locking)
1399 b = btrfs_root_node(root);
1401 b = btrfs_lock_root_node(root);
1404 level = btrfs_header_level(b);
1407 * setup the path here so we can release it under lock
1408 * contention with the cow code
1410 p->nodes[level] = b;
1411 if (!p->skip_locking)
1412 p->locks[level] = 1;
1417 /* is a cow on this block not required */
1418 spin_lock(&root->fs_info->hash_lock);
1419 if (btrfs_header_generation(b) == trans->transid &&
1420 btrfs_header_owner(b) == root->root_key.objectid &&
1421 !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1422 spin_unlock(&root->fs_info->hash_lock);
1425 spin_unlock(&root->fs_info->hash_lock);
1427 /* ok, we have to cow, is our old prealloc the right
1430 if (prealloc_block.objectid &&
1431 prealloc_block.offset != b->len) {
1432 btrfs_free_reserved_extent(root,
1433 prealloc_block.objectid,
1434 prealloc_block.offset);
1435 prealloc_block.objectid = 0;
1439 * for higher level blocks, try not to allocate blocks
1440 * with the block and the parent locks held.
1442 if (level > 1 && !prealloc_block.objectid &&
1443 btrfs_path_lock_waiting(p, level)) {
1445 u64 hint = b->start;
1447 btrfs_release_path(root, p);
1448 ret = btrfs_reserve_extent(trans, root,
1451 &prealloc_block, 0);
1456 wret = btrfs_cow_block(trans, root, b,
1457 p->nodes[level + 1],
1458 p->slots[level + 1],
1459 &b, prealloc_block.objectid);
1460 prealloc_block.objectid = 0;
1462 free_extent_buffer(b);
1468 BUG_ON(!cow && ins_len);
1469 if (level != btrfs_header_level(b))
1471 level = btrfs_header_level(b);
1473 p->nodes[level] = b;
1474 if (!p->skip_locking)
1475 p->locks[level] = 1;
1477 ret = check_block(root, p, level);
1483 ret = bin_search(b, key, level, &slot);
1485 if (ret && slot > 0)
1487 p->slots[level] = slot;
1488 if (ins_len > 0 && btrfs_header_nritems(b) >=
1489 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1490 int sret = split_node(trans, root, p, level);
1496 b = p->nodes[level];
1497 slot = p->slots[level];
1498 } else if (ins_len < 0) {
1499 int sret = balance_level(trans, root, p,
1505 b = p->nodes[level];
1507 btrfs_release_path(NULL, p);
1510 slot = p->slots[level];
1511 BUG_ON(btrfs_header_nritems(b) == 1);
1513 unlock_up(p, level, lowest_unlock);
1515 /* this is only true while dropping a snapshot */
1516 if (level == lowest_level) {
1521 blocknr = btrfs_node_blockptr(b, slot);
1522 gen = btrfs_node_ptr_generation(b, slot);
1523 blocksize = btrfs_level_size(root, level - 1);
1525 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1526 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1530 * reduce lock contention at high levels
1531 * of the btree by dropping locks before
1535 btrfs_release_path(NULL, p);
1537 free_extent_buffer(tmp);
1539 reada_for_search(root, p,
1543 tmp = read_tree_block(root, blocknr,
1546 free_extent_buffer(tmp);
1550 free_extent_buffer(tmp);
1552 reada_for_search(root, p,
1555 b = read_node_slot(root, b, slot);
1558 if (!p->skip_locking)
1561 p->slots[level] = slot;
1562 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1563 sizeof(struct btrfs_item) + ins_len) {
1564 int sret = split_leaf(trans, root, key,
1565 p, ins_len, ret == 0);
1572 unlock_up(p, level, lowest_unlock);
1578 if (prealloc_block.objectid) {
1579 btrfs_free_reserved_extent(root,
1580 prealloc_block.objectid,
1581 prealloc_block.offset);
1587 int btrfs_merge_path(struct btrfs_trans_handle *trans,
1588 struct btrfs_root *root,
1589 struct btrfs_key *node_keys,
1590 u64 *nodes, int lowest_level)
1592 struct extent_buffer *eb;
1593 struct extent_buffer *parent;
1594 struct btrfs_key key;
1603 eb = btrfs_lock_root_node(root);
1604 ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb, 0);
1609 level = btrfs_header_level(parent);
1610 if (level == 0 || level <= lowest_level)
1613 ret = bin_search(parent, &node_keys[lowest_level], level,
1615 if (ret && slot > 0)
1618 bytenr = btrfs_node_blockptr(parent, slot);
1619 if (nodes[level - 1] == bytenr)
1622 blocksize = btrfs_level_size(root, level - 1);
1623 generation = btrfs_node_ptr_generation(parent, slot);
1624 btrfs_node_key_to_cpu(eb, &key, slot);
1625 key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
1627 if (generation == trans->transid) {
1628 eb = read_tree_block(root, bytenr, blocksize,
1630 btrfs_tree_lock(eb);
1634 * if node keys match and node pointer hasn't been modified
1635 * in the running transaction, we can merge the path. for
1636 * blocks owened by reloc trees, the node pointer check is
1637 * skipped, this is because these blocks are fully controlled
1638 * by the space balance code, no one else can modify them.
1640 if (!nodes[level - 1] || !key_match ||
1641 (generation == trans->transid &&
1642 btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
1643 if (level == 1 || level == lowest_level + 1) {
1644 if (generation == trans->transid) {
1645 btrfs_tree_unlock(eb);
1646 free_extent_buffer(eb);
1651 if (generation != trans->transid) {
1652 eb = read_tree_block(root, bytenr, blocksize,
1654 btrfs_tree_lock(eb);
1657 ret = btrfs_cow_block(trans, root, eb, parent, slot,
1661 if (root->root_key.objectid ==
1662 BTRFS_TREE_RELOC_OBJECTID) {
1663 if (!nodes[level - 1]) {
1664 nodes[level - 1] = eb->start;
1665 memcpy(&node_keys[level - 1], &key,
1666 sizeof(node_keys[0]));
1672 btrfs_tree_unlock(parent);
1673 free_extent_buffer(parent);
1678 btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
1679 btrfs_set_node_ptr_generation(parent, slot, trans->transid);
1680 btrfs_mark_buffer_dirty(parent);
1682 ret = btrfs_inc_extent_ref(trans, root,
1684 blocksize, parent->start,
1685 btrfs_header_owner(parent),
1686 btrfs_header_generation(parent),
1691 * If the block was created in the running transaction,
1692 * it's possible this is the last reference to it, so we
1693 * should drop the subtree.
1695 if (generation == trans->transid) {
1696 ret = btrfs_drop_subtree(trans, root, eb, parent);
1698 btrfs_tree_unlock(eb);
1699 free_extent_buffer(eb);
1701 ret = btrfs_free_extent(trans, root, bytenr,
1702 blocksize, parent->start,
1703 btrfs_header_owner(parent),
1704 btrfs_header_generation(parent),
1710 btrfs_tree_unlock(parent);
1711 free_extent_buffer(parent);
1716 * adjust the pointers going up the tree, starting at level
1717 * making sure the right key of each node is points to 'key'.
1718 * This is used after shifting pointers to the left, so it stops
1719 * fixing up pointers when a given leaf/node is not in slot 0 of the
1722 * If this fails to write a tree block, it returns -1, but continues
1723 * fixing up the blocks in ram so the tree is consistent.
1725 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1726 struct btrfs_root *root, struct btrfs_path *path,
1727 struct btrfs_disk_key *key, int level)
1731 struct extent_buffer *t;
1733 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1734 int tslot = path->slots[i];
1735 if (!path->nodes[i])
1738 btrfs_set_node_key(t, key, tslot);
1739 btrfs_mark_buffer_dirty(path->nodes[i]);
1749 * This function isn't completely safe. It's the caller's responsibility
1750 * that the new key won't break the order
1752 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1753 struct btrfs_root *root, struct btrfs_path *path,
1754 struct btrfs_key *new_key)
1756 struct btrfs_disk_key disk_key;
1757 struct extent_buffer *eb;
1760 eb = path->nodes[0];
1761 slot = path->slots[0];
1763 btrfs_item_key(eb, &disk_key, slot - 1);
1764 if (comp_keys(&disk_key, new_key) >= 0)
1767 if (slot < btrfs_header_nritems(eb) - 1) {
1768 btrfs_item_key(eb, &disk_key, slot + 1);
1769 if (comp_keys(&disk_key, new_key) <= 0)
1773 btrfs_cpu_key_to_disk(&disk_key, new_key);
1774 btrfs_set_item_key(eb, &disk_key, slot);
1775 btrfs_mark_buffer_dirty(eb);
1777 fixup_low_keys(trans, root, path, &disk_key, 1);
1782 * try to push data from one node into the next node left in the
1785 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1786 * error, and > 0 if there was no room in the left hand block.
1788 static int push_node_left(struct btrfs_trans_handle *trans,
1789 struct btrfs_root *root, struct extent_buffer *dst,
1790 struct extent_buffer *src, int empty)
1797 src_nritems = btrfs_header_nritems(src);
1798 dst_nritems = btrfs_header_nritems(dst);
1799 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1800 WARN_ON(btrfs_header_generation(src) != trans->transid);
1801 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1803 if (!empty && src_nritems <= 8)
1806 if (push_items <= 0) {
1811 push_items = min(src_nritems, push_items);
1812 if (push_items < src_nritems) {
1813 /* leave at least 8 pointers in the node if
1814 * we aren't going to empty it
1816 if (src_nritems - push_items < 8) {
1817 if (push_items <= 8)
1823 push_items = min(src_nritems - 8, push_items);
1825 copy_extent_buffer(dst, src,
1826 btrfs_node_key_ptr_offset(dst_nritems),
1827 btrfs_node_key_ptr_offset(0),
1828 push_items * sizeof(struct btrfs_key_ptr));
1830 if (push_items < src_nritems) {
1831 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1832 btrfs_node_key_ptr_offset(push_items),
1833 (src_nritems - push_items) *
1834 sizeof(struct btrfs_key_ptr));
1836 btrfs_set_header_nritems(src, src_nritems - push_items);
1837 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1838 btrfs_mark_buffer_dirty(src);
1839 btrfs_mark_buffer_dirty(dst);
1841 ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
1848 * try to push data from one node into the next node right in the
1851 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1852 * error, and > 0 if there was no room in the right hand block.
1854 * this will only push up to 1/2 the contents of the left node over
1856 static int balance_node_right(struct btrfs_trans_handle *trans,
1857 struct btrfs_root *root,
1858 struct extent_buffer *dst,
1859 struct extent_buffer *src)
1867 WARN_ON(btrfs_header_generation(src) != trans->transid);
1868 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1870 src_nritems = btrfs_header_nritems(src);
1871 dst_nritems = btrfs_header_nritems(dst);
1872 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1873 if (push_items <= 0) {
1877 if (src_nritems < 4) {
1881 max_push = src_nritems / 2 + 1;
1882 /* don't try to empty the node */
1883 if (max_push >= src_nritems) {
1887 if (max_push < push_items)
1888 push_items = max_push;
1890 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1891 btrfs_node_key_ptr_offset(0),
1893 sizeof(struct btrfs_key_ptr));
1895 copy_extent_buffer(dst, src,
1896 btrfs_node_key_ptr_offset(0),
1897 btrfs_node_key_ptr_offset(src_nritems - push_items),
1898 push_items * sizeof(struct btrfs_key_ptr));
1900 btrfs_set_header_nritems(src, src_nritems - push_items);
1901 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1903 btrfs_mark_buffer_dirty(src);
1904 btrfs_mark_buffer_dirty(dst);
1906 ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
1913 * helper function to insert a new root level in the tree.
1914 * A new node is allocated, and a single item is inserted to
1915 * point to the existing root
1917 * returns zero on success or < 0 on failure.
1919 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1920 struct btrfs_root *root,
1921 struct btrfs_path *path, int level)
1924 struct extent_buffer *lower;
1925 struct extent_buffer *c;
1926 struct extent_buffer *old;
1927 struct btrfs_disk_key lower_key;
1930 BUG_ON(path->nodes[level]);
1931 BUG_ON(path->nodes[level-1] != root->node);
1933 lower = path->nodes[level-1];
1935 btrfs_item_key(lower, &lower_key, 0);
1937 btrfs_node_key(lower, &lower_key, 0);
1939 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
1940 root->root_key.objectid, trans->transid,
1941 level, root->node->start, 0);
1945 memset_extent_buffer(c, 0, 0, root->nodesize);
1946 btrfs_set_header_nritems(c, 1);
1947 btrfs_set_header_level(c, level);
1948 btrfs_set_header_bytenr(c, c->start);
1949 btrfs_set_header_generation(c, trans->transid);
1950 btrfs_set_header_owner(c, root->root_key.objectid);
1952 write_extent_buffer(c, root->fs_info->fsid,
1953 (unsigned long)btrfs_header_fsid(c),
1956 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1957 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1960 btrfs_set_node_key(c, &lower_key, 0);
1961 btrfs_set_node_blockptr(c, 0, lower->start);
1962 lower_gen = btrfs_header_generation(lower);
1963 WARN_ON(lower_gen != trans->transid);
1965 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1967 btrfs_mark_buffer_dirty(c);
1969 spin_lock(&root->node_lock);
1972 spin_unlock(&root->node_lock);
1974 ret = btrfs_update_extent_ref(trans, root, lower->start,
1975 lower->start, c->start,
1976 root->root_key.objectid,
1977 trans->transid, level - 1);
1980 /* the super has an extra ref to root->node */
1981 free_extent_buffer(old);
1983 add_root_to_dirty_list(root);
1984 extent_buffer_get(c);
1985 path->nodes[level] = c;
1986 path->locks[level] = 1;
1987 path->slots[level] = 0;
1992 * worker function to insert a single pointer in a node.
1993 * the node should have enough room for the pointer already
1995 * slot and level indicate where you want the key to go, and
1996 * blocknr is the block the key points to.
1998 * returns zero on success and < 0 on any error
2000 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2001 *root, struct btrfs_path *path, struct btrfs_disk_key
2002 *key, u64 bytenr, int slot, int level)
2004 struct extent_buffer *lower;
2007 BUG_ON(!path->nodes[level]);
2008 lower = path->nodes[level];
2009 nritems = btrfs_header_nritems(lower);
2012 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2014 if (slot != nritems) {
2015 memmove_extent_buffer(lower,
2016 btrfs_node_key_ptr_offset(slot + 1),
2017 btrfs_node_key_ptr_offset(slot),
2018 (nritems - slot) * sizeof(struct btrfs_key_ptr));
2020 btrfs_set_node_key(lower, key, slot);
2021 btrfs_set_node_blockptr(lower, slot, bytenr);
2022 WARN_ON(trans->transid == 0);
2023 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2024 btrfs_set_header_nritems(lower, nritems + 1);
2025 btrfs_mark_buffer_dirty(lower);
2030 * split the node at the specified level in path in two.
2031 * The path is corrected to point to the appropriate node after the split
2033 * Before splitting this tries to make some room in the node by pushing
2034 * left and right, if either one works, it returns right away.
2036 * returns 0 on success and < 0 on failure
2038 static noinline int split_node(struct btrfs_trans_handle *trans,
2039 struct btrfs_root *root,
2040 struct btrfs_path *path, int level)
2042 struct extent_buffer *c;
2043 struct extent_buffer *split;
2044 struct btrfs_disk_key disk_key;
2050 c = path->nodes[level];
2051 WARN_ON(btrfs_header_generation(c) != trans->transid);
2052 if (c == root->node) {
2053 /* trying to split the root, lets make a new one */
2054 ret = insert_new_root(trans, root, path, level + 1);
2058 ret = push_nodes_for_insert(trans, root, path, level);
2059 c = path->nodes[level];
2060 if (!ret && btrfs_header_nritems(c) <
2061 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2067 c_nritems = btrfs_header_nritems(c);
2069 split = btrfs_alloc_free_block(trans, root, root->nodesize,
2070 path->nodes[level + 1]->start,
2071 root->root_key.objectid,
2072 trans->transid, level, c->start, 0);
2074 return PTR_ERR(split);
2076 btrfs_set_header_flags(split, btrfs_header_flags(c));
2077 btrfs_set_header_level(split, btrfs_header_level(c));
2078 btrfs_set_header_bytenr(split, split->start);
2079 btrfs_set_header_generation(split, trans->transid);
2080 btrfs_set_header_owner(split, root->root_key.objectid);
2081 btrfs_set_header_flags(split, 0);
2082 write_extent_buffer(split, root->fs_info->fsid,
2083 (unsigned long)btrfs_header_fsid(split),
2085 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2086 (unsigned long)btrfs_header_chunk_tree_uuid(split),
2089 mid = (c_nritems + 1) / 2;
2091 copy_extent_buffer(split, c,
2092 btrfs_node_key_ptr_offset(0),
2093 btrfs_node_key_ptr_offset(mid),
2094 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2095 btrfs_set_header_nritems(split, c_nritems - mid);
2096 btrfs_set_header_nritems(c, mid);
2099 btrfs_mark_buffer_dirty(c);
2100 btrfs_mark_buffer_dirty(split);
2102 btrfs_node_key(split, &disk_key, 0);
2103 wret = insert_ptr(trans, root, path, &disk_key, split->start,
2104 path->slots[level + 1] + 1,
2109 ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
2112 if (path->slots[level] >= mid) {
2113 path->slots[level] -= mid;
2114 btrfs_tree_unlock(c);
2115 free_extent_buffer(c);
2116 path->nodes[level] = split;
2117 path->slots[level + 1] += 1;
2119 btrfs_tree_unlock(split);
2120 free_extent_buffer(split);
2126 * how many bytes are required to store the items in a leaf. start
2127 * and nr indicate which items in the leaf to check. This totals up the
2128 * space used both by the item structs and the item data
2130 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2133 int nritems = btrfs_header_nritems(l);
2134 int end = min(nritems, start + nr) - 1;
2138 data_len = btrfs_item_end_nr(l, start);
2139 data_len = data_len - btrfs_item_offset_nr(l, end);
2140 data_len += sizeof(struct btrfs_item) * nr;
2141 WARN_ON(data_len < 0);
2146 * The space between the end of the leaf items and
2147 * the start of the leaf data. IOW, how much room
2148 * the leaf has left for both items and data
2150 int noinline btrfs_leaf_free_space(struct btrfs_root *root,
2151 struct extent_buffer *leaf)
2153 int nritems = btrfs_header_nritems(leaf);
2155 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2157 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
2158 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2159 leaf_space_used(leaf, 0, nritems), nritems);
2165 * push some data in the path leaf to the right, trying to free up at
2166 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2168 * returns 1 if the push failed because the other node didn't have enough
2169 * room, 0 if everything worked out and < 0 if there were major errors.
2171 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2172 *root, struct btrfs_path *path, int data_size,
2175 struct extent_buffer *left = path->nodes[0];
2176 struct extent_buffer *right;
2177 struct extent_buffer *upper;
2178 struct btrfs_disk_key disk_key;
2184 struct btrfs_item *item;
2192 slot = path->slots[1];
2193 if (!path->nodes[1]) {
2196 upper = path->nodes[1];
2197 if (slot >= btrfs_header_nritems(upper) - 1)
2200 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2202 right = read_node_slot(root, upper, slot + 1);
2203 btrfs_tree_lock(right);
2204 free_space = btrfs_leaf_free_space(root, right);
2205 if (free_space < data_size + sizeof(struct btrfs_item))
2208 /* cow and double check */
2209 ret = btrfs_cow_block(trans, root, right, upper,
2210 slot + 1, &right, 0);
2214 free_space = btrfs_leaf_free_space(root, right);
2215 if (free_space < data_size + sizeof(struct btrfs_item))
2218 left_nritems = btrfs_header_nritems(left);
2219 if (left_nritems == 0)
2227 if (path->slots[0] >= left_nritems)
2228 push_space += data_size + sizeof(*item);
2230 i = left_nritems - 1;
2232 item = btrfs_item_nr(left, i);
2234 if (!empty && push_items > 0) {
2235 if (path->slots[0] > i)
2237 if (path->slots[0] == i) {
2238 int space = btrfs_leaf_free_space(root, left);
2239 if (space + push_space * 2 > free_space)
2244 if (path->slots[0] == i)
2245 push_space += data_size + sizeof(*item);
2247 if (!left->map_token) {
2248 map_extent_buffer(left, (unsigned long)item,
2249 sizeof(struct btrfs_item),
2250 &left->map_token, &left->kaddr,
2251 &left->map_start, &left->map_len,
2255 this_item_size = btrfs_item_size(left, item);
2256 if (this_item_size + sizeof(*item) + push_space > free_space)
2260 push_space += this_item_size + sizeof(*item);
2265 if (left->map_token) {
2266 unmap_extent_buffer(left, left->map_token, KM_USER1);
2267 left->map_token = NULL;
2270 if (push_items == 0)
2273 if (!empty && push_items == left_nritems)
2276 /* push left to right */
2277 right_nritems = btrfs_header_nritems(right);
2279 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2280 push_space -= leaf_data_end(root, left);
2282 /* make room in the right data area */
2283 data_end = leaf_data_end(root, right);
2284 memmove_extent_buffer(right,
2285 btrfs_leaf_data(right) + data_end - push_space,
2286 btrfs_leaf_data(right) + data_end,
2287 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2289 /* copy from the left data area */
2290 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2291 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2292 btrfs_leaf_data(left) + leaf_data_end(root, left),
2295 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2296 btrfs_item_nr_offset(0),
2297 right_nritems * sizeof(struct btrfs_item));
2299 /* copy the items from left to right */
2300 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2301 btrfs_item_nr_offset(left_nritems - push_items),
2302 push_items * sizeof(struct btrfs_item));
2304 /* update the item pointers */
2305 right_nritems += push_items;
2306 btrfs_set_header_nritems(right, right_nritems);
2307 push_space = BTRFS_LEAF_DATA_SIZE(root);
2308 for (i = 0; i < right_nritems; i++) {
2309 item = btrfs_item_nr(right, i);
2310 if (!right->map_token) {
2311 map_extent_buffer(right, (unsigned long)item,
2312 sizeof(struct btrfs_item),
2313 &right->map_token, &right->kaddr,
2314 &right->map_start, &right->map_len,
2317 push_space -= btrfs_item_size(right, item);
2318 btrfs_set_item_offset(right, item, push_space);
2321 if (right->map_token) {
2322 unmap_extent_buffer(right, right->map_token, KM_USER1);
2323 right->map_token = NULL;
2325 left_nritems -= push_items;
2326 btrfs_set_header_nritems(left, left_nritems);
2329 btrfs_mark_buffer_dirty(left);
2330 btrfs_mark_buffer_dirty(right);
2332 ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
2335 btrfs_item_key(right, &disk_key, 0);
2336 btrfs_set_node_key(upper, &disk_key, slot + 1);
2337 btrfs_mark_buffer_dirty(upper);
2339 /* then fixup the leaf pointer in the path */
2340 if (path->slots[0] >= left_nritems) {
2341 path->slots[0] -= left_nritems;
2342 if (btrfs_header_nritems(path->nodes[0]) == 0)
2343 clean_tree_block(trans, root, path->nodes[0]);
2344 btrfs_tree_unlock(path->nodes[0]);
2345 free_extent_buffer(path->nodes[0]);
2346 path->nodes[0] = right;
2347 path->slots[1] += 1;
2349 btrfs_tree_unlock(right);
2350 free_extent_buffer(right);
2355 btrfs_tree_unlock(right);
2356 free_extent_buffer(right);
2361 * push some data in the path leaf to the left, trying to free up at
2362 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2364 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2365 *root, struct btrfs_path *path, int data_size,
2368 struct btrfs_disk_key disk_key;
2369 struct extent_buffer *right = path->nodes[0];
2370 struct extent_buffer *left;
2376 struct btrfs_item *item;
2377 u32 old_left_nritems;
2383 u32 old_left_item_size;
2385 slot = path->slots[1];
2388 if (!path->nodes[1])
2391 right_nritems = btrfs_header_nritems(right);
2392 if (right_nritems == 0) {
2396 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2398 left = read_node_slot(root, path->nodes[1], slot - 1);
2399 btrfs_tree_lock(left);
2400 free_space = btrfs_leaf_free_space(root, left);
2401 if (free_space < data_size + sizeof(struct btrfs_item)) {
2406 /* cow and double check */
2407 ret = btrfs_cow_block(trans, root, left,
2408 path->nodes[1], slot - 1, &left, 0);
2410 /* we hit -ENOSPC, but it isn't fatal here */
2415 free_space = btrfs_leaf_free_space(root, left);
2416 if (free_space < data_size + sizeof(struct btrfs_item)) {
2424 nr = right_nritems - 1;
2426 for (i = 0; i < nr; i++) {
2427 item = btrfs_item_nr(right, i);
2428 if (!right->map_token) {
2429 map_extent_buffer(right, (unsigned long)item,
2430 sizeof(struct btrfs_item),
2431 &right->map_token, &right->kaddr,
2432 &right->map_start, &right->map_len,
2436 if (!empty && push_items > 0) {
2437 if (path->slots[0] < i)
2439 if (path->slots[0] == i) {
2440 int space = btrfs_leaf_free_space(root, right);
2441 if (space + push_space * 2 > free_space)
2446 if (path->slots[0] == i)
2447 push_space += data_size + sizeof(*item);
2449 this_item_size = btrfs_item_size(right, item);
2450 if (this_item_size + sizeof(*item) + push_space > free_space)
2454 push_space += this_item_size + sizeof(*item);
2457 if (right->map_token) {
2458 unmap_extent_buffer(right, right->map_token, KM_USER1);
2459 right->map_token = NULL;
2462 if (push_items == 0) {
2466 if (!empty && push_items == btrfs_header_nritems(right))
2469 /* push data from right to left */
2470 copy_extent_buffer(left, right,
2471 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2472 btrfs_item_nr_offset(0),
2473 push_items * sizeof(struct btrfs_item));
2475 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2476 btrfs_item_offset_nr(right, push_items -1);
2478 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2479 leaf_data_end(root, left) - push_space,
2480 btrfs_leaf_data(right) +
2481 btrfs_item_offset_nr(right, push_items - 1),
2483 old_left_nritems = btrfs_header_nritems(left);
2484 BUG_ON(old_left_nritems < 0);
2486 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2487 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2490 item = btrfs_item_nr(left, i);
2491 if (!left->map_token) {
2492 map_extent_buffer(left, (unsigned long)item,
2493 sizeof(struct btrfs_item),
2494 &left->map_token, &left->kaddr,
2495 &left->map_start, &left->map_len,
2499 ioff = btrfs_item_offset(left, item);
2500 btrfs_set_item_offset(left, item,
2501 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2503 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2504 if (left->map_token) {
2505 unmap_extent_buffer(left, left->map_token, KM_USER1);
2506 left->map_token = NULL;
2509 /* fixup right node */
2510 if (push_items > right_nritems) {
2511 printk("push items %d nr %u\n", push_items, right_nritems);
2515 if (push_items < right_nritems) {
2516 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2517 leaf_data_end(root, right);
2518 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2519 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2520 btrfs_leaf_data(right) +
2521 leaf_data_end(root, right), push_space);
2523 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2524 btrfs_item_nr_offset(push_items),
2525 (btrfs_header_nritems(right) - push_items) *
2526 sizeof(struct btrfs_item));
2528 right_nritems -= push_items;
2529 btrfs_set_header_nritems(right, right_nritems);
2530 push_space = BTRFS_LEAF_DATA_SIZE(root);
2531 for (i = 0; i < right_nritems; i++) {
2532 item = btrfs_item_nr(right, i);
2534 if (!right->map_token) {
2535 map_extent_buffer(right, (unsigned long)item,
2536 sizeof(struct btrfs_item),
2537 &right->map_token, &right->kaddr,
2538 &right->map_start, &right->map_len,
2542 push_space = push_space - btrfs_item_size(right, item);
2543 btrfs_set_item_offset(right, item, push_space);
2545 if (right->map_token) {
2546 unmap_extent_buffer(right, right->map_token, KM_USER1);
2547 right->map_token = NULL;
2550 btrfs_mark_buffer_dirty(left);
2552 btrfs_mark_buffer_dirty(right);
2554 ret = btrfs_update_ref(trans, root, right, left,
2555 old_left_nritems, push_items);
2558 btrfs_item_key(right, &disk_key, 0);
2559 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2563 /* then fixup the leaf pointer in the path */
2564 if (path->slots[0] < push_items) {
2565 path->slots[0] += old_left_nritems;
2566 if (btrfs_header_nritems(path->nodes[0]) == 0)
2567 clean_tree_block(trans, root, path->nodes[0]);
2568 btrfs_tree_unlock(path->nodes[0]);
2569 free_extent_buffer(path->nodes[0]);
2570 path->nodes[0] = left;
2571 path->slots[1] -= 1;
2573 btrfs_tree_unlock(left);
2574 free_extent_buffer(left);
2575 path->slots[0] -= push_items;
2577 BUG_ON(path->slots[0] < 0);
2580 btrfs_tree_unlock(left);
2581 free_extent_buffer(left);
2586 * split the path's leaf in two, making sure there is at least data_size
2587 * available for the resulting leaf level of the path.
2589 * returns 0 if all went well and < 0 on failure.
2591 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2592 struct btrfs_root *root,
2593 struct btrfs_key *ins_key,
2594 struct btrfs_path *path, int data_size,
2597 struct extent_buffer *l;
2601 struct extent_buffer *right;
2602 int space_needed = data_size + sizeof(struct btrfs_item);
2609 int num_doubles = 0;
2610 struct btrfs_disk_key disk_key;
2613 space_needed = data_size;
2615 /* first try to make some room by pushing left and right */
2616 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
2617 wret = push_leaf_right(trans, root, path, data_size, 0);
2622 wret = push_leaf_left(trans, root, path, data_size, 0);
2628 /* did the pushes work? */
2629 if (btrfs_leaf_free_space(root, l) >= space_needed)
2633 if (!path->nodes[1]) {
2634 ret = insert_new_root(trans, root, path, 1);
2641 slot = path->slots[0];
2642 nritems = btrfs_header_nritems(l);
2643 mid = (nritems + 1)/ 2;
2645 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2646 path->nodes[1]->start,
2647 root->root_key.objectid,
2648 trans->transid, 0, l->start, 0);
2649 if (IS_ERR(right)) {
2651 return PTR_ERR(right);
2654 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2655 btrfs_set_header_bytenr(right, right->start);
2656 btrfs_set_header_generation(right, trans->transid);
2657 btrfs_set_header_owner(right, root->root_key.objectid);
2658 btrfs_set_header_level(right, 0);
2659 write_extent_buffer(right, root->fs_info->fsid,
2660 (unsigned long)btrfs_header_fsid(right),
2663 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2664 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2668 leaf_space_used(l, mid, nritems - mid) + space_needed >
2669 BTRFS_LEAF_DATA_SIZE(root)) {
2670 if (slot >= nritems) {
2671 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2672 btrfs_set_header_nritems(right, 0);
2673 wret = insert_ptr(trans, root, path,
2674 &disk_key, right->start,
2675 path->slots[1] + 1, 1);
2679 btrfs_tree_unlock(path->nodes[0]);
2680 free_extent_buffer(path->nodes[0]);
2681 path->nodes[0] = right;
2683 path->slots[1] += 1;
2684 btrfs_mark_buffer_dirty(right);
2688 if (mid != nritems &&
2689 leaf_space_used(l, mid, nritems - mid) +
2690 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2695 if (leaf_space_used(l, 0, mid + 1) + space_needed >
2696 BTRFS_LEAF_DATA_SIZE(root)) {
2697 if (!extend && slot == 0) {
2698 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2699 btrfs_set_header_nritems(right, 0);
2700 wret = insert_ptr(trans, root, path,
2706 btrfs_tree_unlock(path->nodes[0]);
2707 free_extent_buffer(path->nodes[0]);
2708 path->nodes[0] = right;
2710 if (path->slots[1] == 0) {
2711 wret = fixup_low_keys(trans, root,
2712 path, &disk_key, 1);
2716 btrfs_mark_buffer_dirty(right);
2718 } else if (extend && slot == 0) {
2722 if (mid != nritems &&
2723 leaf_space_used(l, mid, nritems - mid) +
2724 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2730 nritems = nritems - mid;
2731 btrfs_set_header_nritems(right, nritems);
2732 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2734 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2735 btrfs_item_nr_offset(mid),
2736 nritems * sizeof(struct btrfs_item));
2738 copy_extent_buffer(right, l,
2739 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2740 data_copy_size, btrfs_leaf_data(l) +
2741 leaf_data_end(root, l), data_copy_size);
2743 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2744 btrfs_item_end_nr(l, mid);
2746 for (i = 0; i < nritems; i++) {
2747 struct btrfs_item *item = btrfs_item_nr(right, i);
2750 if (!right->map_token) {
2751 map_extent_buffer(right, (unsigned long)item,
2752 sizeof(struct btrfs_item),
2753 &right->map_token, &right->kaddr,
2754 &right->map_start, &right->map_len,
2758 ioff = btrfs_item_offset(right, item);
2759 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2762 if (right->map_token) {
2763 unmap_extent_buffer(right, right->map_token, KM_USER1);
2764 right->map_token = NULL;
2767 btrfs_set_header_nritems(l, mid);
2769 btrfs_item_key(right, &disk_key, 0);
2770 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2771 path->slots[1] + 1, 1);
2775 btrfs_mark_buffer_dirty(right);
2776 btrfs_mark_buffer_dirty(l);
2777 BUG_ON(path->slots[0] != slot);
2779 ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
2783 btrfs_tree_unlock(path->nodes[0]);
2784 free_extent_buffer(path->nodes[0]);
2785 path->nodes[0] = right;
2786 path->slots[0] -= mid;
2787 path->slots[1] += 1;
2789 btrfs_tree_unlock(right);
2790 free_extent_buffer(right);
2793 BUG_ON(path->slots[0] < 0);
2796 BUG_ON(num_doubles != 0);
2804 * make the item pointed to by the path smaller. new_size indicates
2805 * how small to make it, and from_end tells us if we just chop bytes
2806 * off the end of the item or if we shift the item to chop bytes off
2809 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2810 struct btrfs_root *root,
2811 struct btrfs_path *path,
2812 u32 new_size, int from_end)
2817 struct extent_buffer *leaf;
2818 struct btrfs_item *item;
2820 unsigned int data_end;
2821 unsigned int old_data_start;
2822 unsigned int old_size;
2823 unsigned int size_diff;
2826 slot_orig = path->slots[0];
2827 leaf = path->nodes[0];
2828 slot = path->slots[0];
2830 old_size = btrfs_item_size_nr(leaf, slot);
2831 if (old_size == new_size)
2834 nritems = btrfs_header_nritems(leaf);
2835 data_end = leaf_data_end(root, leaf);
2837 old_data_start = btrfs_item_offset_nr(leaf, slot);
2839 size_diff = old_size - new_size;
2842 BUG_ON(slot >= nritems);
2845 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2847 /* first correct the data pointers */
2848 for (i = slot; i < nritems; i++) {
2850 item = btrfs_item_nr(leaf, i);
2852 if (!leaf->map_token) {
2853 map_extent_buffer(leaf, (unsigned long)item,
2854 sizeof(struct btrfs_item),
2855 &leaf->map_token, &leaf->kaddr,
2856 &leaf->map_start, &leaf->map_len,
2860 ioff = btrfs_item_offset(leaf, item);
2861 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2864 if (leaf->map_token) {
2865 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2866 leaf->map_token = NULL;
2869 /* shift the data */
2871 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2872 data_end + size_diff, btrfs_leaf_data(leaf) +
2873 data_end, old_data_start + new_size - data_end);
2875 struct btrfs_disk_key disk_key;
2878 btrfs_item_key(leaf, &disk_key, slot);
2880 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2882 struct btrfs_file_extent_item *fi;
2884 fi = btrfs_item_ptr(leaf, slot,
2885 struct btrfs_file_extent_item);
2886 fi = (struct btrfs_file_extent_item *)(
2887 (unsigned long)fi - size_diff);
2889 if (btrfs_file_extent_type(leaf, fi) ==
2890 BTRFS_FILE_EXTENT_INLINE) {
2891 ptr = btrfs_item_ptr_offset(leaf, slot);
2892 memmove_extent_buffer(leaf, ptr,
2894 offsetof(struct btrfs_file_extent_item,
2899 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2900 data_end + size_diff, btrfs_leaf_data(leaf) +
2901 data_end, old_data_start - data_end);
2903 offset = btrfs_disk_key_offset(&disk_key);
2904 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2905 btrfs_set_item_key(leaf, &disk_key, slot);
2907 fixup_low_keys(trans, root, path, &disk_key, 1);
2910 item = btrfs_item_nr(leaf, slot);
2911 btrfs_set_item_size(leaf, item, new_size);
2912 btrfs_mark_buffer_dirty(leaf);
2915 if (btrfs_leaf_free_space(root, leaf) < 0) {
2916 btrfs_print_leaf(root, leaf);
2923 * make the item pointed to by the path bigger, data_size is the new size.
2925 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2926 struct btrfs_root *root, struct btrfs_path *path,
2932 struct extent_buffer *leaf;
2933 struct btrfs_item *item;
2935 unsigned int data_end;
2936 unsigned int old_data;
2937 unsigned int old_size;
2940 slot_orig = path->slots[0];
2941 leaf = path->nodes[0];
2943 nritems = btrfs_header_nritems(leaf);
2944 data_end = leaf_data_end(root, leaf);
2946 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2947 btrfs_print_leaf(root, leaf);
2950 slot = path->slots[0];
2951 old_data = btrfs_item_end_nr(leaf, slot);
2954 if (slot >= nritems) {
2955 btrfs_print_leaf(root, leaf);
2956 printk("slot %d too large, nritems %d\n", slot, nritems);
2961 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2963 /* first correct the data pointers */
2964 for (i = slot; i < nritems; i++) {
2966 item = btrfs_item_nr(leaf, i);
2968 if (!leaf->map_token) {
2969 map_extent_buffer(leaf, (unsigned long)item,
2970 sizeof(struct btrfs_item),
2971 &leaf->map_token, &leaf->kaddr,
2972 &leaf->map_start, &leaf->map_len,
2975 ioff = btrfs_item_offset(leaf, item);
2976 btrfs_set_item_offset(leaf, item, ioff - data_size);
2979 if (leaf->map_token) {
2980 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2981 leaf->map_token = NULL;
2984 /* shift the data */
2985 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2986 data_end - data_size, btrfs_leaf_data(leaf) +
2987 data_end, old_data - data_end);
2989 data_end = old_data;
2990 old_size = btrfs_item_size_nr(leaf, slot);
2991 item = btrfs_item_nr(leaf, slot);
2992 btrfs_set_item_size(leaf, item, old_size + data_size);
2993 btrfs_mark_buffer_dirty(leaf);
2996 if (btrfs_leaf_free_space(root, leaf) < 0) {
2997 btrfs_print_leaf(root, leaf);
3004 * Given a key and some data, insert items into the tree.
3005 * This does all the path init required, making room in the tree if needed.
3007 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3008 struct btrfs_root *root,
3009 struct btrfs_path *path,
3010 struct btrfs_key *cpu_key, u32 *data_size,
3013 struct extent_buffer *leaf;
3014 struct btrfs_item *item;
3022 unsigned int data_end;
3023 struct btrfs_disk_key disk_key;
3025 for (i = 0; i < nr; i++) {
3026 total_data += data_size[i];
3029 total_size = total_data + (nr * sizeof(struct btrfs_item));
3030 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3036 slot_orig = path->slots[0];
3037 leaf = path->nodes[0];
3039 nritems = btrfs_header_nritems(leaf);
3040 data_end = leaf_data_end(root, leaf);
3042 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3043 btrfs_print_leaf(root, leaf);
3044 printk("not enough freespace need %u have %d\n",
3045 total_size, btrfs_leaf_free_space(root, leaf));
3049 slot = path->slots[0];
3052 if (slot != nritems) {
3053 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3055 if (old_data < data_end) {
3056 btrfs_print_leaf(root, leaf);
3057 printk("slot %d old_data %d data_end %d\n",
3058 slot, old_data, data_end);
3062 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3064 /* first correct the data pointers */
3065 WARN_ON(leaf->map_token);
3066 for (i = slot; i < nritems; i++) {
3069 item = btrfs_item_nr(leaf, i);
3070 if (!leaf->map_token) {
3071 map_extent_buffer(leaf, (unsigned long)item,
3072 sizeof(struct btrfs_item),
3073 &leaf->map_token, &leaf->kaddr,
3074 &leaf->map_start, &leaf->map_len,
3078 ioff = btrfs_item_offset(leaf, item);
3079 btrfs_set_item_offset(leaf, item, ioff - total_data);
3081 if (leaf->map_token) {
3082 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3083 leaf->map_token = NULL;
3086 /* shift the items */
3087 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3088 btrfs_item_nr_offset(slot),
3089 (nritems - slot) * sizeof(struct btrfs_item));
3091 /* shift the data */
3092 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3093 data_end - total_data, btrfs_leaf_data(leaf) +
3094 data_end, old_data - data_end);
3095 data_end = old_data;
3098 /* setup the item for the new data */
3099 for (i = 0; i < nr; i++) {
3100 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3101 btrfs_set_item_key(leaf, &disk_key, slot + i);
3102 item = btrfs_item_nr(leaf, slot + i);
3103 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3104 data_end -= data_size[i];
3105 btrfs_set_item_size(leaf, item, data_size[i]);
3107 btrfs_set_header_nritems(leaf, nritems + nr);
3108 btrfs_mark_buffer_dirty(leaf);
3112 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3113 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3116 if (btrfs_leaf_free_space(root, leaf) < 0) {
3117 btrfs_print_leaf(root, leaf);
3125 * Given a key and some data, insert an item into the tree.
3126 * This does all the path init required, making room in the tree if needed.
3128 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3129 *root, struct btrfs_key *cpu_key, void *data, u32
3133 struct btrfs_path *path;
3134 struct extent_buffer *leaf;
3137 path = btrfs_alloc_path();
3139 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3141 leaf = path->nodes[0];
3142 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3143 write_extent_buffer(leaf, data, ptr, data_size);
3144 btrfs_mark_buffer_dirty(leaf);
3146 btrfs_free_path(path);
3151 * delete the pointer from a given node.
3153 * the tree should have been previously balanced so the deletion does not
3156 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3157 struct btrfs_path *path, int level, int slot)
3159 struct extent_buffer *parent = path->nodes[level];
3164 nritems = btrfs_header_nritems(parent);
3165 if (slot != nritems -1) {
3166 memmove_extent_buffer(parent,
3167 btrfs_node_key_ptr_offset(slot),
3168 btrfs_node_key_ptr_offset(slot + 1),
3169 sizeof(struct btrfs_key_ptr) *
3170 (nritems - slot - 1));
3173 btrfs_set_header_nritems(parent, nritems);
3174 if (nritems == 0 && parent == root->node) {
3175 BUG_ON(btrfs_header_level(root->node) != 1);
3176 /* just turn the root into a leaf and break */
3177 btrfs_set_header_level(root->node, 0);
3178 } else if (slot == 0) {
3179 struct btrfs_disk_key disk_key;
3181 btrfs_node_key(parent, &disk_key, 0);
3182 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3186 btrfs_mark_buffer_dirty(parent);
3191 * a helper function to delete the leaf pointed to by path->slots[1] and
3192 * path->nodes[1]. bytenr is the node block pointer, but since the callers
3193 * already know it, it is faster to have them pass it down than to
3194 * read it out of the node again.
3196 * This deletes the pointer in path->nodes[1] and frees the leaf
3197 * block extent. zero is returned if it all worked out, < 0 otherwise.
3199 * The path must have already been setup for deleting the leaf, including
3200 * all the proper balancing. path->nodes[1] must be locked.
3202 noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3203 struct btrfs_root *root,
3204 struct btrfs_path *path, u64 bytenr)
3207 u64 root_gen = btrfs_header_generation(path->nodes[1]);
3209 ret = del_ptr(trans, root, path, 1, path->slots[1]);
3213 ret = btrfs_free_extent(trans, root, bytenr,
3214 btrfs_level_size(root, 0),
3215 path->nodes[1]->start,
3216 btrfs_header_owner(path->nodes[1]),
3221 * delete the item at the leaf level in path. If that empties
3222 * the leaf, remove it from the tree
3224 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3225 struct btrfs_path *path, int slot, int nr)
3227 struct extent_buffer *leaf;
3228 struct btrfs_item *item;
3236 leaf = path->nodes[0];
3237 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3239 for (i = 0; i < nr; i++)
3240 dsize += btrfs_item_size_nr(leaf, slot + i);
3242 nritems = btrfs_header_nritems(leaf);
3244 if (slot + nr != nritems) {
3245 int data_end = leaf_data_end(root, leaf);
3247 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3249 btrfs_leaf_data(leaf) + data_end,
3250 last_off - data_end);
3252 for (i = slot + nr; i < nritems; i++) {
3255 item = btrfs_item_nr(leaf, i);
3256 if (!leaf->map_token) {
3257 map_extent_buffer(leaf, (unsigned long)item,
3258 sizeof(struct btrfs_item),
3259 &leaf->map_token, &leaf->kaddr,
3260 &leaf->map_start, &leaf->map_len,
3263 ioff = btrfs_item_offset(leaf, item);
3264 btrfs_set_item_offset(leaf, item, ioff + dsize);
3267 if (leaf->map_token) {
3268 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3269 leaf->map_token = NULL;
3272 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3273 btrfs_item_nr_offset(slot + nr),
3274 sizeof(struct btrfs_item) *
3275 (nritems - slot - nr));
3277 btrfs_set_header_nritems(leaf, nritems - nr);
3280 /* delete the leaf if we've emptied it */
3282 if (leaf == root->node) {
3283 btrfs_set_header_level(leaf, 0);
3285 ret = btrfs_del_leaf(trans, root, path, leaf->start);
3289 int used = leaf_space_used(leaf, 0, nritems);
3291 struct btrfs_disk_key disk_key;
3293 btrfs_item_key(leaf, &disk_key, 0);
3294 wret = fixup_low_keys(trans, root, path,
3300 /* delete the leaf if it is mostly empty */
3301 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
3302 /* push_leaf_left fixes the path.
3303 * make sure the path still points to our leaf
3304 * for possible call to del_ptr below
3306 slot = path->slots[1];
3307 extent_buffer_get(leaf);
3309 wret = push_leaf_left(trans, root, path, 1, 1);
3310 if (wret < 0 && wret != -ENOSPC)
3313 if (path->nodes[0] == leaf &&
3314 btrfs_header_nritems(leaf)) {
3315 wret = push_leaf_right(trans, root, path, 1, 1);
3316 if (wret < 0 && wret != -ENOSPC)
3320 if (btrfs_header_nritems(leaf) == 0) {
3321 path->slots[1] = slot;
3322 ret = btrfs_del_leaf(trans, root, path, leaf->start);
3324 free_extent_buffer(leaf);
3326 /* if we're still in the path, make sure
3327 * we're dirty. Otherwise, one of the
3328 * push_leaf functions must have already
3329 * dirtied this buffer
3331 if (path->nodes[0] == leaf)
3332 btrfs_mark_buffer_dirty(leaf);
3333 free_extent_buffer(leaf);
3336 btrfs_mark_buffer_dirty(leaf);
3343 * search the tree again to find a leaf with lesser keys
3344 * returns 0 if it found something or 1 if there are no lesser leaves.
3345 * returns < 0 on io errors.
3347 * This may release the path, and so you may lose any locks held at the
3350 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3352 struct btrfs_key key;
3353 struct btrfs_disk_key found_key;
3356 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3360 else if (key.type > 0)
3362 else if (key.objectid > 0)
3367 btrfs_release_path(root, path);
3368 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3371 btrfs_item_key(path->nodes[0], &found_key, 0);
3372 ret = comp_keys(&found_key, &key);
3379 * A helper function to walk down the tree starting at min_key, and looking
3380 * for nodes or leaves that are either in cache or have a minimum
3381 * transaction id. This is used by the btree defrag code, and tree logging
3383 * This does not cow, but it does stuff the starting key it finds back
3384 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3385 * key and get a writable path.
3387 * This does lock as it descends, and path->keep_locks should be set
3388 * to 1 by the caller.
3390 * This honors path->lowest_level to prevent descent past a given level
3393 * min_trans indicates the oldest transaction that you are interested
3394 * in walking through. Any nodes or leaves older than min_trans are
3395 * skipped over (without reading them).
3397 * returns zero if something useful was found, < 0 on error and 1 if there
3398 * was nothing in the tree that matched the search criteria.
3400 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3401 struct btrfs_key *max_key,
3402 struct btrfs_path *path, int cache_only,
3405 struct extent_buffer *cur;
3406 struct btrfs_key found_key;
3414 cur = btrfs_lock_root_node(root);
3415 level = btrfs_header_level(cur);
3416 WARN_ON(path->nodes[level]);
3417 path->nodes[level] = cur;
3418 path->locks[level] = 1;
3420 if (btrfs_header_generation(cur) < min_trans) {
3425 nritems = btrfs_header_nritems(cur);
3426 level = btrfs_header_level(cur);
3427 sret = bin_search(cur, min_key, level, &slot);
3429 /* at the lowest level, we're done, setup the path and exit */
3430 if (level == path->lowest_level) {
3431 if (slot >= nritems)
3434 path->slots[level] = slot;
3435 btrfs_item_key_to_cpu(cur, &found_key, slot);
3438 if (sret && slot > 0)
3441 * check this node pointer against the cache_only and
3442 * min_trans parameters. If it isn't in cache or is too
3443 * old, skip to the next one.
3445 while(slot < nritems) {
3448 struct extent_buffer *tmp;
3449 struct btrfs_disk_key disk_key;
3451 blockptr = btrfs_node_blockptr(cur, slot);
3452 gen = btrfs_node_ptr_generation(cur, slot);
3453 if (gen < min_trans) {
3461 btrfs_node_key(cur, &disk_key, slot);
3462 if (comp_keys(&disk_key, max_key) >= 0) {
3468 tmp = btrfs_find_tree_block(root, blockptr,
3469 btrfs_level_size(root, level - 1));
3471 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3472 free_extent_buffer(tmp);
3476 free_extent_buffer(tmp);
3481 * we didn't find a candidate key in this node, walk forward
3482 * and find another one
3484 if (slot >= nritems) {
3485 path->slots[level] = slot;
3486 sret = btrfs_find_next_key(root, path, min_key, level,
3487 cache_only, min_trans);
3489 btrfs_release_path(root, path);
3495 /* save our key for returning back */
3496 btrfs_node_key_to_cpu(cur, &found_key, slot);
3497 path->slots[level] = slot;
3498 if (level == path->lowest_level) {
3500 unlock_up(path, level, 1);
3503 cur = read_node_slot(root, cur, slot);
3505 btrfs_tree_lock(cur);
3506 path->locks[level - 1] = 1;
3507 path->nodes[level - 1] = cur;
3508 unlock_up(path, level, 1);
3512 memcpy(min_key, &found_key, sizeof(found_key));
3517 * this is similar to btrfs_next_leaf, but does not try to preserve
3518 * and fixup the path. It looks for and returns the next key in the
3519 * tree based on the current path and the cache_only and min_trans
3522 * 0 is returned if another key is found, < 0 if there are any errors
3523 * and 1 is returned if there are no higher keys in the tree
3525 * path->keep_locks should be set to 1 on the search made before
3526 * calling this function.
3528 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
3529 struct btrfs_key *key, int lowest_level,
3530 int cache_only, u64 min_trans)
3532 int level = lowest_level;
3534 struct extent_buffer *c;
3536 while(level < BTRFS_MAX_LEVEL) {
3537 if (!path->nodes[level])
3540 slot = path->slots[level] + 1;
3541 c = path->nodes[level];
3543 if (slot >= btrfs_header_nritems(c)) {
3545 if (level == BTRFS_MAX_LEVEL) {
3551 btrfs_item_key_to_cpu(c, key, slot);
3553 u64 blockptr = btrfs_node_blockptr(c, slot);
3554 u64 gen = btrfs_node_ptr_generation(c, slot);
3557 struct extent_buffer *cur;
3558 cur = btrfs_find_tree_block(root, blockptr,
3559 btrfs_level_size(root, level - 1));
3560 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
3563 free_extent_buffer(cur);
3566 free_extent_buffer(cur);
3568 if (gen < min_trans) {
3572 btrfs_node_key_to_cpu(c, key, slot);
3580 * search the tree again to find a leaf with greater keys
3581 * returns 0 if it found something or 1 if there are no greater leaves.
3582 * returns < 0 on io errors.
3584 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
3588 struct extent_buffer *c;
3589 struct extent_buffer *next = NULL;
3590 struct btrfs_key key;
3594 nritems = btrfs_header_nritems(path->nodes[0]);
3599 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
3601 btrfs_release_path(root, path);
3602 path->keep_locks = 1;
3603 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3604 path->keep_locks = 0;
3609 nritems = btrfs_header_nritems(path->nodes[0]);
3611 * by releasing the path above we dropped all our locks. A balance
3612 * could have added more items next to the key that used to be
3613 * at the very end of the block. So, check again here and
3614 * advance the path if there are now more items available.
3616 if (nritems > 0 && path->slots[0] < nritems - 1) {
3621 while(level < BTRFS_MAX_LEVEL) {
3622 if (!path->nodes[level])
3625 slot = path->slots[level] + 1;
3626 c = path->nodes[level];
3627 if (slot >= btrfs_header_nritems(c)) {
3629 if (level == BTRFS_MAX_LEVEL) {
3636 btrfs_tree_unlock(next);
3637 free_extent_buffer(next);
3640 if (level == 1 && (path->locks[1] || path->skip_locking) &&
3642 reada_for_search(root, path, level, slot, 0);
3644 next = read_node_slot(root, c, slot);
3645 if (!path->skip_locking) {
3646 WARN_ON(!btrfs_tree_locked(c));
3647 btrfs_tree_lock(next);
3651 path->slots[level] = slot;
3654 c = path->nodes[level];
3655 if (path->locks[level])
3656 btrfs_tree_unlock(c);
3657 free_extent_buffer(c);
3658 path->nodes[level] = next;
3659 path->slots[level] = 0;
3660 if (!path->skip_locking)
3661 path->locks[level] = 1;
3664 if (level == 1 && path->locks[1] && path->reada)
3665 reada_for_search(root, path, level, slot, 0);
3666 next = read_node_slot(root, next, 0);
3667 if (!path->skip_locking) {
3668 WARN_ON(!btrfs_tree_locked(path->nodes[level]));
3669 btrfs_tree_lock(next);
3673 unlock_up(path, 0, 1);
3678 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
3679 * searching until it gets past min_objectid or finds an item of 'type'
3681 * returns 0 if something is found, 1 if nothing was found and < 0 on error
3683 int btrfs_previous_item(struct btrfs_root *root,
3684 struct btrfs_path *path, u64 min_objectid,
3687 struct btrfs_key found_key;
3688 struct extent_buffer *leaf;
3693 if (path->slots[0] == 0) {
3694 ret = btrfs_prev_leaf(root, path);
3700 leaf = path->nodes[0];
3701 nritems = btrfs_header_nritems(leaf);
3704 if (path->slots[0] == nritems)
3707 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3708 if (found_key.type == type)
3710 if (found_key.objectid < min_objectid)
3712 if (found_key.objectid == min_objectid &&
3713 found_key.type < type)